User research is one of the most consistently expensive activities in a product organisation. It is also one of the most consistently misused. Not because the research is bad. Because the path from finding to decision is broken in ways that are structural, not accidental, and that most organisations treat as a communication problem when they are actually a governance problem.

I have spent over a decade working in product and growth roles at major digital platforms. In that time, I have commissioned research, presented research, sat in rooms where research was being discussed, and watched research findings disappear without trace into the space between a discovery session and a sprint planning meeting. The pattern is consistent enough that I no longer think it is a people problem. It is a process problem. And the fix is analytical, not interpersonal.

This essay is an attempt to name the mechanisms precisely and describe what closing the gap actually requires.

Why findings do not become decisions

The first thing to understand is that research findings do not fail to get implemented because nobody read them. They fail to get implemented because they arrive in the wrong form for the decision-making context they are entering.

A user research finding looks like this: ‘Users in the 35–50 age group express confusion at the point where the form asks for billing address separately from delivery address. Three of the five participants in the session attempted to skip this step.’ That is a legitimate research finding. It is specific, observed, and traceable to a session.

A product backlog item looks like this: ‘As a user, I want to complete my checkout without confusion so that I can receive my order.’ These are not the same thing. The finding describes a specific friction point with a specific mechanism. The user story describes a desired outcome at a level of abstraction that could be addressed by dozens of different interventions, or by none.

The translation from finding to backlog item is where most of the information content is lost. Not through negligence. Through a mismatch between the format research produces and the format that planning processes consume. Research outputs are typically qualitative, contextual, and causal. Backlog items are typically functional, abstract, and outcome-stated. Bridging those two formats requires explicit work that most organisations have not formalised.

The gap between what user research surfaces and what gets built is not a research quality problem. It is a translation problem. The findings are often correct. The mechanism for converting them into implementable decisions does not exist.

The three points where findings die

There are three specific points in most product processes where research findings stop moving forward. Understanding which one is responsible in a given organisation determines what the fix looks like.

The first is the handover. Research is conducted by one team (UX, research, or insight) and handed to another (product, engineering, or design) in the form of a report, a slide deck, or a session readout. The receiving team was not present for the sessions. They did not hear the tone of voice when a participant said they found something confusing. They did not see the hesitation. They received a document. Documents are easy to acknowledge and easy to deprioritise, because acknowledging is not the same as committing.

The organisations that avoid the handover problem do not solve it with better documents. They solve it by requiring the people who will make the implementation decision to be present, at minimum, for the synthesis session. Not the full sessions. The synthesis. If the product lead sits in the room when the research team says ‘the billing address step caused abandonment in three of five participants,’ the finding has a human observer attached to it. It is harder to deprioritise a finding you witnessed than one you read about.

The second point is prioritisation. Even when findings are well-understood, they compete for backlog position against items that are easier to estimate, easier to scope, and more directly connected to a metric that someone is being measured on. A finding that says ‘mobile users are abandoning at the postcode field because the wrong keyboard type opens’ is correct and specific. But it does not come with a story points estimate or a revenue impact figure. It comes with a quote and an observation. Against a feature request that has a business case attached, it loses almost every time.

The fix here is not to make research findings more compelling. It is to require that findings be translated into a format that the prioritisation process accepts before they can be considered. If prioritisation decisions are made against an effort-impact matrix, findings need to arrive at that meeting with both an effort estimate and an impact estimate. The impact estimate does not need to be precise. It needs to exist. ‘If we fix this, we expect checkout completion on mobile to improve by X percentage points, based on the proportion of sessions where this step caused visible friction.’ That is enough to enter the prioritisation conversation. A quote from a user session is not.

The third point is accountability. Most organisations have no mechanism for tracking what happened to a research finding after it was shared. There is no system that records ‘finding F was surfaced on date D, entered the backlog as item I, was deprioritised at sprint planning S, and has not been revisited.’ Without that tracking, findings exist in a state of permanent ambiguity. Nobody is responsible for ensuring they are addressed. Nobody can report on the percentage of high-severity findings that were implemented within a defined time frame. The research function has no feedback loop on whether its output is generating value.

What closing the gap actually requires

Closing the gap between research and implementation is not a research methodology question. The research methodology may be fine. It is an organisational design question. The following three changes address the three failure points described above.

First: embed research observation into the decision-making role. This does not mean every product manager must conduct research. It means that the people who will decide whether to implement a finding should be present for the moment at which the finding is synthesised, not just the moment at which it is reported. In practice, this means replacing the research readout meeting with a joint synthesis session in which product and design stakeholders participate in organising and prioritising the raw findings before they are written up. What gets lost in the written report never existed in the joint session. The finding is already contextualised.

Second: create a finding-to-backlog translation layer with defined criteria. Before any research finding can be considered in sprint planning, it should pass through a structured translation that produces: a problem statement (what the user cannot do or finds difficult), an estimated impact (what percentage of users are affected and what behaviour is disrupted), an effort range (small, medium, large, validated with engineering), and a proposed success metric (how we will know this has been addressed). This is a one-page template, not a process burden. It exists to ensure that findings arrive at planning meetings in a form that can be compared to other backlog items on the same dimensions.

Third: track finding status as a first-class metric. High-severity findings (those affecting a significant percentage of users at a high-impact funnel stage) should be tracked with the same rigour as open bugs. They should have a status (identified, translated, in backlog, in sprint, implemented, validated), an owner, and a review cadence. The research function should report, on a regular cycle, on the percentage of high-severity findings that have been implemented within a defined window. If that number is consistently low, it is a product process problem, not a research problem, and it should be escalated as one.

The research function cannot fix this alone. And the product function cannot fix it by asking for better research. The fix requires a shared accountability structure that does not currently exist in most organisations.

The analyst’s role in closing the gap

Business analysts are well-positioned to own the translation layer described above. The skills required are already in the BA toolkit: structured problem decomposition, requirements translation, effort-impact analysis, and stakeholder facilitation. What is usually missing is the mandate to apply those skills to the research-to-backlog pipeline.

In practice, this means the BA role in a product team should include a defined responsibility for finding translation: taking research outputs and converting them into backlog-ready items with the full set of attributes the prioritisation process requires. This is not a creative interpretation of BA work. It is a natural extension of the requirements elicitation and analysis function into a domain that has been underserved by structured thinking.

The alternative, which is the current default, is that research findings are translated informally by whoever happens to be in the room, using whatever format feels right, with no consistency and no tracking. The output of that process is a backlog full of items that originated in research sessions nobody can trace, and a research archive full of findings that nobody implemented and nobody can explain why.

That is not a research failure. It is an analytical gap. And it is exactly the kind of gap that structured analytical practice exists to close.

Ammarah Ahmed is the founder of Precision Consulting, a CRO and UX practice for growth-stage e-commerce brands. She spent over a decade leading growth and product teams at Delivery Hero, Foodpanda, Talabat, and Daraz across MENA and Southeast Asia.

References: Krug, S. (2009). Rocket Surgery Made Easy. New Riders. Patton, J. (2014). User Story Mapping. O’Reilly. Nielsen Norman Group research on research-to-design translation: nngroup.com.

The gap between what user research surfaces and what gets built was originally published in Analyst’s corner on Medium, where people are continuing the conversation by highlighting and responding to this story.

There is a specific kind of business problem that looks like success for a long time before it looks like anything else.

The growth curve is steep. The team is growing. The press is good. The investors are happy. And somewhere underneath all of it, something structural is quietly breaking that nobody has the time or the incentive to look at, because the numbers that are visible are all moving in the right direction.

I have seen this pattern more times than I can count, across markets in Asia and the Middle East, at platforms operating at significant scale. And the consistent finding is this: the businesses that scaled the fastest were often the hardest to fix when something went wrong. Not because the problems were more severe. Because the growth had built a layer of insulation between the leadership and the reality of what was happening at the customer level.

The mechanism is not complicated. When a business grows quickly, it generates a lot of signal. Revenue is up. Downloads are up. Orders are up. Partnerships are signing. Headcount is growing. Every week there is a new milestone to celebrate and a new problem to solve, and the most pressing problems are always the ones that are blocking the next milestone, not the ones that are quietly eroding the foundation.

The customers who are churning are not in the room. The buyers who abandoned their cart because the mobile checkout was broken are not in the room. The users who tried the product once and never came back because the onboarding flow gave them no reason to are not in the room. The data about those people exists. But in a fast-growing business, the data that gets attention is the data that supports the growth narrative, because that is the data that everyone is being measured against.

At one platform I worked at, the business had grown rapidly by expanding into new cities. Each new city launch generated a burst of new users, new orders, new press. The aggregate numbers looked extraordinary. What was not visible in the aggregate was that cohort retention in the mature cities, the ones that had launched 12 to 18 months earlier, was declining. The new-city growth was masking a retention problem in the existing base. By the time the expansion slowed and the retained cohorts became the primary driver of revenue, the retention problem had been compounding quietly for over a year.

Fixing it then was significantly harder than fixing it would have been at month three. Not because the problem had grown more complex. Because by month 18, the team had been structured, incentivised, and culturally shaped around the growth story. Nobody’s job title was ‘fix the thing that’s been quietly breaking while we expanded.’ Nobody was being measured on it. And the organisational habits that had made the fast expansion possible, moving quickly, celebrating new numbers, treating existing customers as a foundation rather than a product problem, were exactly the habits that made the slower, more careful diagnostic work feel wrong.

This is the thing about fast growth that I think is genuinely underappreciated. Growth is not just a financial condition. It is a cultural one. And the culture that produces fast growth is not the same culture that produces durable growth. The skills are different. The incentives are different. The tolerance for ambiguity is different.

The founder who is brilliant at growth, at seeing an opportunity, moving fast, building momentum, telling the story, is often exactly the wrong person to diagnose a structural retention problem. Not because they are not smart. Because their instincts have been calibrated by a period in which moving fast and trusting the top-line numbers was correct. Those instincts do not turn off automatically when the business reaches a stage where they are no longer useful.

I have watched this dynamic play out at the individual level. A founder who built something from nothing to significant scale through sheer velocity and will, who then found themselves in a business that needed patience, nuance, and a willingness to spend six months fixing something invisible before the fix would show up in any number they could show an investor. The skills that got them there were actively working against them.

The businesses that navigated this well had one thing in common. Not a particular strategy or a specific operational framework. They had a founder or a senior leader who had the self-awareness to recognise when the growth phase was over and the consolidation phase had begun, and who was willing to change what they measured, what they rewarded, and what problems they personally spent time on.

That sounds obvious. It is extraordinarily rare.

Measuring retention cohorts is boring compared to measuring new city launches. Fixing the mobile checkout flow in three existing markets is invisible compared to announcing an expansion into two new countries. Spending six weeks on the post-purchase email sequence because it is driving 18% of repeat orders is the kind of work that does not make it into the all-hands presentation. And in a business where the culture has been built around what makes it into the all-hands presentation, the invisible work does not get done.

There is a version of this that I have started calling the scaling trap, for lack of a better name. It describes the condition of a business that has grown quickly enough to have significant operational complexity, a large team, multiple markets or product lines, real fixed costs, but has not yet built the measurement and diagnostic infrastructure to understand what is actually driving performance at a unit level. The business is big enough that the problems are expensive, but not structured enough to see them clearly.

Getting out of the scaling trap requires slowing down before you are forced to. It requires choosing, deliberately, to spend a quarter looking carefully at what is happening in your existing base before you add more complexity on top of it. It requires being willing to find things that are uncomfortable. And it requires having the organisational authority to act on what you find, which means the people who built the growth culture have to be willing to be wrong about some of the assumptions that made the growth possible.

Most businesses do not do this voluntarily. They do it when something breaks publicly, or when growth slows enough that the hidden problems become visible, or when a competitor does it first and the gap in retention rates shows up in the market share data.

The ones that do it voluntarily are, in my experience, the ones that are still standing a decade later.

I do not think this is a strategy problem. The right analytical frameworks for cohort retention, unit economics, and product health are well documented. Any competent operator knows them.

It is a timing problem. The question is not whether to do the diagnostic work. The question is whether you do it while you still have the runway and the organisational energy to act on what you find, or whether you do it after the growth has slowed and the options have narrowed.

Fast growth creates optionality. That is its great gift. But optionality has a shelf life, and it expires faster than most founders expect. The businesses I have seen use it well are the ones that understood, earlier than felt comfortable, that the growth phase was not the end of the journey. It was the beginning of a more complicated one.

The hardest conversation in any fast-growing business is the one where someone says: we need to slow down and look at what we have built. Not because things are going badly. Because things are going well enough that we still have the choice.

Most businesses never have that conversation until the choice is gone.

Ammarah Ahmed is the founder of Precision Consulting (goprecision.co), a CRO and UX consultancy for growth-stage e-commerce brands. She spent over a decade leading growth and product teams at Delivery Hero, Foodpanda, Talabat, and Daraz across MENA and Southeast Asia.

The businesses that scale the fastest are the hardest to save was originally published in Entrepreneurship Handbook on Medium, where people are continuing the conversation by highlighting and responding to this story.

More than half of all e-commerce traffic comes from mobile. Most stores convert at two to three times the rate on desktop. Same products. Same prices. Same brand. The device alone accounts for a two to threefold difference in conversion rate.

The gap is not because mobile users are less willing to buy. It is because most e-commerce sites were designed at a desk, tested at a desk, and optimised at a desk. Then they got scaled down to fit a phone and were assumed to work. They do not work at the same rate.

I have audited over 120 e-commerce funnels. The mobile failures are consistent. Eight patterns account for the majority of the conversion gap. This article covers each one: what it is, why it kills conversion, and the specific fix.

Pattern 1: The above-the-fold problem

A product page built for a 1440px desktop might show the hero image, headline, price, CTA, and star rating all before the scroll. On a 390px phone, the same page often shows only the hero image and part of the headline. The buyer has to scroll before they can even see the price.

The mechanism: buyers make a fast decision about whether to invest further attention within the first few seconds. If the information they need to decide is not visible immediately, most will not scroll to find it. They will leave.

The scroll test: open your best-selling product page on your phone right now. Without scrolling, list every element you can see. Product name, price, star rating, delivery signal, and Add to Cart should all be on that list. If any are missing, this is your first fix.

The fix: restructure the mobile product page so the purchase zone sits above the fold. Reduce image height if needed. Price, star rating, and CTA must be visible without a single scroll. This is a hierarchy decision, not a design one.

Pattern 2: Tap targets sized for a mouse

Apple’s Human Interface Guidelines specify a minimum tap target of 44 by 44 points. Google’s Material Design recommends 48 by 48 density-independent pixels. Most e-commerce sites were designed for mouse clicks, which are precise to a single pixel. A thumb is not.

Add to Cart buttons, size selectors, quantity controls, filter chips, and form fields: if they were sized for a mouse cursor, they are too small for a thumb. A missed tap means trying again. That friction compounds across a funnel where every interaction costs more effort on a small screen than it does on a desktop.

The fix: add padding. A 36px button with 10px of padding on each side becomes 56px and easy to tap. No visual redesign required. Go through every interactive element in your store and check it passes the 44px minimum. The most common offenders are filter chips on category pages, close buttons on modals, and quantity adjusters on product pages.

Pattern 3: Checkout fields built for a keyboard

On a desktop, filling out a checkout form is mildly annoying. On mobile, it is a genuine obstacle. A 16-digit card number entered on a keyboard that covers half the screen. A billing address field that autocorrects to the wrong thing. A postcode that opens a letter keyboard instead of a number pad.

Baymard Institute’s research shows the average e-commerce checkout has 23 form elements. The optimised benchmark is 12 to 14. That gap represents 9 to 11 unnecessary decisions between intent and purchase. On mobile, each one compounds because typing is effortful and every pause is an opportunity to reconsider.

The fix has two parts. First, use the correct input type for every field: inputmode=’numeric’ for card numbers and postcodes, type=’email’ for email fields, autocomplete attributes on address fields. These are single-line HTML changes that require no design work. Second, enable mobile wallets. Apple Pay, Google Pay, and Shop Pay replace every form field with two taps. If your payment processor supports it, this is the highest-ROI change available in mobile UX and requires almost no engineering work to implement.

Pattern 4: Page speed on a mobile connection

53% of mobile visitors leave a page that takes more than three seconds to load (Think with Google). A site that loads in 1.5 seconds on broadband might take 4 to 5 seconds on a 4G connection on a mid-range device with a fraction of the processing power. Every additional second is a measurable conversion loss that compounds across every page in the funnel.

The two most common causes are uncompressed images and unused third-party scripts. Both are fixable without a rebuild. Every product image should be under 200KB where possible and in WebP format. Third-party scripts (chat widgets, analytics, marketing pixels) add load time without contributing to the buying decision. Audit what is running and remove what is not earning its weight.

The fix: run your homepage, product page, cart, and checkout through Google PageSpeed Insights on a simulated mobile connection. A score below 50 is a conversion problem. Address image compression first. It accounts for the majority of mobile speed issues and can be resolved without touching the codebase.

Pattern 5: Navigation that needs three taps

Hamburger menus that require three taps to reach a category. Search bars that expand into a full-screen overlay with no keyboard shortcut. Category filters that are impossible to select with a thumb because they sit in the stretch zone at the middle of the screen.

Mobile navigation should get a buyer to any category in two taps. The search bar should always be visible, not hidden behind an icon. Category filters should sit in the bottom third of the screen where they are reachable one-handed. When filter interactions trigger a full page reload that returns the buyer to the top of the category, engagement drops because the buyer has to re-establish context before continuing.

The fix: map your navigation against the thumb zone. The bottom third of the screen is the easy-reach zone. Primary CTAs, sticky checkout buttons, and high-frequency navigation elements belong here. The top third is the hard-to-reach zone where navigation hamburgers and modal close buttons almost always live, and where most misclicks happen.

Pattern 6: Trust signals that disappear on mobile

On desktop, returns policy badges, security seals, and payment logos often appear beside the Add to Cart button. On mobile, the same elements are frequently pushed below the fold or removed entirely to save space. That is exactly backwards. The moment of highest friction for a mobile buyer is right at the CTA, which is precisely where trust signals need to be visible.

The mechanism: buying online requires trust, and trust signals reduce perceived risk at the decision moment. A ‘30-day returns’ guarantee visible on desktop but hidden on mobile means mobile buyers are making a higher-risk decision with less reassurance. That asymmetry shows up directly in conversion rate.

The fix: audit your product page on a phone and identify every trust element that is below the fold. Restructure the purchase zone so at least one trust signal, ideally returns policy or star rating, is visible alongside the Add to Cart button without scrolling.

Pattern 7: The invisible checkout button

The cart page on mobile should move the buyer from their items to the checkout button in a single scroll. When the checkout button is below three or four items and a cross-sell block, a significant percentage of buyers who intended to proceed simply do not find it.

This is one of the easiest fixes on the list and one of the most consistently missed. Most cart pages are built to show items and then present the total and checkout option below. On mobile, that sequence means the action the buyer needs to take is the last thing they see, rather than the thing they are guided toward.

The fix: pin a sticky checkout CTA to the bottom of the screen on the cart page. The button should be visible at all times without scrolling. It costs almost nothing to implement. The lift in checkout starts tends to be immediate because you are removing a navigation problem, not changing anything about the product or price.

Pattern 8: Desktop-first team processes

This is the root cause behind all seven patterns above. Most product decisions get reviewed on a MacBook. Most analytics dashboards are open on a monitor. Most design work happens at a desk. The result is a team with systematically less exposure to how the mobile experience actually feels than their customers have.

The experience is rarely broken. It is harder than it needs to be at every single step. That cumulative difficulty is what drives the gap. And because the team never feels it directly, it does not register as a priority until someone pulls the mobile conversion rate separately from the blended number and the gap becomes visible.

The fix is a process change. Before any page change goes live, someone should walk through it on a real phone on a real mobile connection, not a browser simulator. Mobile conversion rate should be a named metric in weekly reporting, tracked separately from the blended rate. Spend one hour watching mobile session recordings filtered to mobile only in Microsoft Clarity (free) and what you find will almost certainly change your priorities faster than any audit checklist.

Where to start

Not everything above is equal in impact. Based on the funnels I have worked through, these are the fixes in priority order:

1. Enable mobile wallets. Replaces every checkout form field with two taps. Biggest lift, lowest complexity.

2. Fix page speed. Image compression and removing unused scripts solve most issues without architectural changes.

3. Restructure the product page purchase zone for mobile. CTA, price, and social proof visible before any scrolling.

4. Increase tap target sizes. Padding adjustments only. No visual redesign needed.

5. Add a sticky checkout CTA on the cart page. Fixed to the bottom, always visible.

6. Correct input types on checkout form fields. One-line HTML changes per field.

Start at the top. These address the highest-frequency, highest-impact friction points. Most can be implemented without significant engineering work, which means the revenue impact arrives quickly.

The mobile conversion gap is not a traffic problem or a product problem. It is a friction problem. And friction is one of the most reliably fixable things in a digital product.

Ammarah Ahmed is the founder of Precision Consulting, a CRO and UX practice for growth-stage e-commerce brands. She has audited over 120 e-commerce funnels across MENA and Southeast Asia. goprecision.co

References: Baymard Institute checkout research (baymard.com). Think with Google mobile speed research. Apple Human Interface Guidelines, minimum tap target sizes. Google Material Design, touch target guidance.

8 mobile UX patterns that kill e-commerce conversion (and how to fix each one) was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.

Between Delivery Hero, Foodpanda, Talabat, Daraz, and Middle East Internet Group, I spent over a decade working across roughly 15 markets. Egypt, Saudi Arabia, the UAE, Pakistan, Bangladesh, Sri Lanka, Myanmar, Singapore, the Philippines, and others. Different languages. Different payment behaviours. Different levels of smartphone penetration. Different logistics realities.

What was consistent was the failure modes.

The businesses that struggled did not struggle for unique reasons. They struggled for the same three reasons, in the same order, across nearly every market I worked in. The context changed. The underlying problem did not.

Here is what they were, and why they were so hard to see from the inside.

The first thing: building for the customer they imagined, not the one who actually showed up

Every market entry starts with a customer assumption. The team imagines who is going to use the product, how tech-savvy they are, what device they are on, how much they earn, what they trust, and what they are willing to spend. These assumptions get built into the product, the pricing, the payment options, and the marketing.

In most of the markets I worked in, the actual customer who showed up was meaningfully different from the assumed one. Not completely different. But different enough that the product built for the imagined customer created friction for the real one at almost every step.

In one market, the team had built a checkout flow assuming bank card penetration that did not match reality. A significant portion of the actual customer base wanted to pay cash on delivery. The checkout did not support it prominently. The team knew cash on delivery existed as an option but had deprioritised it because the assumed customer was expected to prefer digital payment. The actual customer abandoned at checkout at a rate that took months to fully diagnose because nobody had asked the question from the customer’s side first.

The pattern I saw repeat: product decisions made from the inside of the business, based on who the team hoped would use it, rather than from observation of who was actually trying to use it. This is not a failure of intelligence. It is a failure of methodology. The assumed customer and the real customer are always different to some degree. The question is how quickly you find out.

The businesses that recovered fastest had one thing in common: they got to user research before they got to optimisation. Not polished research. Watching real people try to use the product and noting where they stopped.

The second thing: confusing category growth with business growth

When you enter a new market, especially one where your category is underpenetrated, the early growth numbers are deceptive. The business grows because the category grows. New users come in because they are discovering online ordering, or online shopping, or delivery, for the first time. The business looks healthy because the top line is moving.

What the top line hides is how the business is actually performing against the customers who have been around long enough to make a real choice. Are they coming back? Are they spending more on their second and third orders than on their first? Are they choosing this platform over the competitors that have since entered the market?

In markets where category growth was fastest, this confusion was worst. The new-user numbers were so strong that the retention numbers did not get scrutinised until category growth slowed, which it always eventually does. By the time the slowdown arrived, some businesses had never built the product or operational quality needed to retain customers who had options. They had grown with the rising tide and discovered, when the tide went out, that the underlying business was weaker than the headline numbers suggested.

The businesses that avoided this had a specific discipline: they tracked cohort behaviour separately from aggregate growth. Not just overall orders per month, but orders per customer per month for customers in their second, third, and fourth month on the platform. That number tells you whether you are building a real business or riding a wave. Most founding teams I encountered were not looking at it.

The third thing: solving the wrong problem because they measured the wrong thing

This is the one I saw most consistently, across the most different types of businesses, and it is the one that is hardest to fix because it does not feel like a problem from the inside. The team is working hard. The metrics are moving. The dashboards look active. And the business is optimising for something that has a weak relationship with what it actually needs to grow.

I have written about this elsewhere in the context of individual stores, but at platform scale the same dynamic plays out with higher stakes. A team measuring app downloads instead of activated users. A team measuring gross merchandise value instead of contribution margin per order. A team measuring delivery speed as an operational target without connecting it to repeat order rate, which was the actual outcome delivery speed was supposed to drive.

In one platform, the business had a significant operational investment in reducing average delivery time. Delivery speed was the headline metric. It featured in investor materials, in performance reviews, and in team goals. When someone finally connected delivery speed to repeat order rate by cohort, the correlation was weak. Customer satisfaction surveys showed that delivery speed ranked third or fourth in the list of reasons customers chose the platform or stayed with it. Reliability (arriving when promised), selection (having what they wanted), and trust (feeling safe paying) ranked above it. The business had been optimising hard for the fourth most important driver of the outcome it needed.

The speed investment was not wrong. Delivery speed matters. But it was being treated as the primary lever when the data, once someone looked at it correctly, showed it was not. The correct primary lever was selection. More of the items customers wanted, available and in stock. That finding redirected several quarters of product and operational work.

Why these three repeat across every market

They share a structural cause. All three are problems of information asymmetry between the business and its customers.

Building for the imagined customer happens because the team has better access to their own assumptions than to observed customer behaviour. Confusing category growth with business growth happens because aggregate numbers are more visible than cohort behaviour. Optimising for the wrong metric happens because the metric that is easy to measure gets tracked, while the metric with the stronger causal relationship to the outcome stays unmeasured or gets buried in a dashboard nobody opens.

In each case, the business is flying on the information it has rather than seeking the information it needs. This is not laziness. Operating across multiple markets at speed, with limited teams and significant pressure to show results, makes it genuinely hard to step back and ask whether you are measuring the right things and building for the right people.

But the businesses that scaled well across markets did exactly that. Not continuously. But at deliberate intervals, usually triggered by something going wrong, they asked: who is actually using this, what are they doing, and what does the data say is driving the outcome we need? The answer was usually different from what the team had assumed. And acting on it was almost always the most important thing they did that quarter.

I am not describing edge cases. These three patterns accounted for the majority of underperformance I observed across a decade of work in markets that were otherwise quite different from each other. The lesson is not that e-commerce is hard in these markets. It is that e-commerce is hard everywhere when the business loses contact with what the actual customer is doing and why.

The good news is that regaining that contact is not structurally complicated. It requires watching real customers use the product, tracking the right cohort metrics, and being willing to redirect resources when the data contradicts the assumption. None of those require budget or headcount that most scaling businesses do not have. They require the discipline to look at uncomfortable information and take it seriously.

Fifteen markets taught me that the failure modes are universal. So are the fixes.

Ammarah Ahmed is the founder of Precision Consulting (goprecision.co), a CRO and UX consultancy for growth-stage e-commerce brands. She spent over a decade leading growth and product teams at Delivery Hero, Foodpanda, Talabat, and Daraz across MENA and Southeast Asia.

I ran e-commerce across 15 markets. The same three things killed businesses in all of them. was originally published in Analyst’s corner on Medium, where people are continuing the conversation by highlighting and responding to this story.

There is a line in interaction design between guiding a user toward an action and manipulating them into it. Most design teams would say they are on the right side of that line. Most of them have not drawn the line precisely enough to know.

The problem is not intent. Most manipulative patterns in digital interfaces were not designed to deceive. They were designed to convert, to retain, to activate. The designers who built them were optimising for measurable outcomes and reached for patterns that worked. The fact that those patterns work by exploiting cognitive weaknesses rather than by genuinely serving the user is rarely examined as a separate question.

This essay tries to draw the line precisely. Not as a moral argument, but as a design one. Manipulation and guidance are technically distinguishable. Knowing the difference is part of being a competent practitioner.

What guidance looks like

Guidance reduces friction between a user’s intent and the action that fulfils it. The user already has a goal. The design makes it easier to reach. Nothing about the goal changes. Nothing about what the user wants is altered. The design simply removes obstacles.

A checkout flow that auto-detects a user’s city from their postcode is guidance. The user intended to enter their address. The interface made that faster. A search bar that surfaces the most relevant result at the top is guidance. The user intended to find something. The interface made it more findable. A progress indicator in a multi-step form is guidance. The user intended to complete the form. The interface made the remaining distance visible.

The common thread: the user’s goal is unchanged. The design serves that goal. The user finishes the interaction closer to what they came for.

Guidance answers the question the user is already asking. Manipulation answers a question they were not asking and frames the answer in a way that makes it look like the answer to the question they were.

What manipulation looks like

Manipulation changes what a user believes about a situation in order to produce an action the designer wants, regardless of whether that action serves the user’s interests. The goal of the design is not to help the user reach their goal faster. It is to redirect the user toward a different goal entirely, usually one that benefits the product or business.

Harry Brignull coined the term ‘dark patterns’ in 2010 to describe exactly this category of design. His taxonomy includes roach motels (easy to get into, hard to leave), trick questions (deliberately ambiguous opt-in language), misdirection (visual design that draws attention away from an important element), and confirmshaming (making the decline option read as self-deprecating). Each of these works not by removing friction from the user’s path but by adding friction to the user’s escape route, or by altering the user’s perception of their options.

The mechanism, in each case, is cognitive. Manipulation exploits the same heuristics and biases that guidance works with, but in the opposite direction. Guidance uses proximity to reduce search effort. Manipulation uses proximity to associate a desired action with something trusted. Guidance uses visual salience to make the primary action findable. Manipulation uses visual salience to make the cancellation option unfindable.

The grey area: persuasion

Between guidance and manipulation sits persuasion, and this is where the design community has the most unresolved argument.

A limited-stock indicator is persuasive. It communicates genuine scarcity and activates loss aversion, the tendency to weight losses more heavily than equivalent gains, first documented by Kahneman and Tversky in their 1979 Prospect Theory paper. If the stock information is accurate, the indicator guides the user toward a decision they might otherwise delay and later regret. If the stock information is fabricated or perpetually refreshed to create artificial scarcity, the same indicator is manipulation. The design is identical. The ethics depend entirely on the accuracy of the underlying information.

This is the most useful test I know for the grey area: is the information true? A countdown timer for a sale that ends on Friday is persuasion if the sale actually ends on Friday and returns to the original price on Saturday. It is manipulation if the timer resets every 24 hours regardless of what the price does. The psychological mechanism is the same. The distinction is factual, not visual.

Social proof works the same way. Showing that 47 people are viewing a product right now is guidance if 47 people are actually viewing it. It is manipulation if the number is fabricated or algorithmically inflated. The design element looks identical in both cases. Its character as guidance or manipulation is entirely determined by whether it accurately represents reality.

Why the distinction matters to designers specifically

The argument is sometimes made that this is a business ethics question, not a design question. Designers design. Leadership decides what information to surface. If leadership decides to show fabricated stock counts, that is their decision, not the designer’s.

This argument fails on two grounds. The first is practical. Designers are the people who build the systems that display the information. A designer who builds a ‘people viewing this now’ counter knows, or should know, what data source it pulls from. A designer who builds a countdown timer knows, or should know, whether it resets. The technical knowledge is on the designer’s side of the table. The decision to use it manipulatively or honestly may be made elsewhere, but the capacity to flag the distinction belongs to the designer.

The second ground is the user relationship. Every design decision is an implicit communication with the user. When an interface uses a pattern associated with genuine scarcity to signal artificial scarcity, it is borrowing trust that was earned by honest use of that pattern elsewhere and spending it on a fiction. That trust belongs to the user. Spending it without their knowledge is a relationship decision, not just a business one. And the long-term consequence of that decision, eroded trust, reduced willingness to return, increased scepticism toward future communications, falls on the product and its users, not on the metric that was optimised in the short term.

Susan Weinschenk’s work on the psychology of design describes this as the trust loop: interfaces that behave predictably and honestly build a mental model in the user that allows them to act with confidence. Interfaces that behave manipulatively break that model. Once broken, it is not easily rebuilt. Users who have been tricked by a countdown timer that resets will disbelieve the next countdown timer they see, even if it is accurate.

A working test

Before shipping any pattern that uses a cognitive bias or psychological mechanism, ask three questions.

First: is the information this pattern communicates accurate? Not approximately accurate. Accurately accurate. If a stock counter shows ‘only 3 left’ and the actual count is 47, the pattern communicates something false. That is manipulation regardless of whether the intent was to deceive.

Second: if the user knew exactly how this pattern worked, would they feel it had served them? This is the transparency test, described by Ryan Calo in his 2014 paper on digital market manipulation. A user who understood that the ‘most popular’ badge is assigned by sales volume rather than by review score might make a different choice. If full transparency about the mechanism would change the user’s behaviour, the pattern is using information asymmetry to its advantage. That is manipulation.

Third: does this pattern make it harder for the user to do something they are entitled to do, like cancel, decline, or leave? If the answer is yes, the pattern is a dark pattern regardless of how it is positioned internally. Friction on the user’s exit is not retention design. It is a roach motel.

The design community’s responsibility

There is a growing body of work on this from researchers and practitioners, including Brignull’s ongoing dark patterns database, the work of the Center for Humane Technology on attention manipulation, and the emerging academic field of algorithmic persuasion. The conversation exists. What it sometimes lacks is the precision to be actionable at the level of an individual design decision.

The guidance versus manipulation distinction is not settled by intent. It is not settled by aesthetics. It is not settled by whether a pattern appears on a list of known dark patterns. It is settled by asking whether the pattern serves the user’s goal or redirects it, and whether the information it communicates is true.

Those are testable questions. They can be asked about any pattern, on any product, at any stage of the design process. The fact that they are not routinely asked is a practice gap, not a knowledge gap. The knowledge is there. The discipline to apply it consistently is what the field is still building.

Ammarah Ahmed is the founder of Precision Consulting, a CRO and UX practice for growth-stage e-commerce brands. She has worked on product and growth teams at major platforms across Asia and the Middle East.

References: Brignull, H. (2010). Dark Patterns. darkpatterns.org. Kahneman, D. & Tversky, A. (1979). Prospect Theory. Econometrica. Calo, R. (2014). Digital Market Manipulation. George Washington Law Review. Weinschenk, S. (2011). 100 Things Every Designer Needs to Know About People. New Riders. Center for Humane Technology: humanetech.com.

Where interaction design stops guiding and starts manipulatin was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.

The first 60 seconds of an onboarding flow are the most consequential interaction a product will ever have with a new user. This is not an exaggeration. Research by Appcues found that users who do not experience value within the first session rarely come back. And most onboarding flows are designed in a way that makes experiencing value in the first session structurally impossible.

I have spent a decade running growth and product teams at platforms handling millions of daily active users. The onboarding problem looked the same everywhere. Not because the products were bad. Because the flows were built around what the product wanted to show, not what the user needed to feel.

Here is what I have seen go wrong, and why it goes wrong at such a consistent rate.

The first mistake: leading with features

The most common onboarding pattern I encounter is a product tour. Arrows pointing at interface elements. Tooltips explaining what buttons do. A progress bar showing the user they are on step 2 of 6. The implicit logic is: if the user understands how the product works, they will use it. This logic is wrong.

Understanding a product and feeling the value of a product are different experiences. A user who has read all six tooltip screens knows what every button does. They do not yet know whether any of it is relevant to them. They are more informed than they were at the start of the flow. They are not more motivated.

BJ Fogg’s Behaviour Model describes three conditions for behaviour to occur: motivation, ability, and a prompt arriving at the right moment. Product tours address ability (can the user do the thing?) but consistently neglect motivation (does the user want to do the thing?). When motivation is low, ability is irrelevant. The user leaves not because they could not figure out the product, but because they did not yet see a reason to stay.

The question an onboarding flow should answer in the first 60 seconds is not ‘do you understand how this works?’ It is ‘do you understand why this is worth your time?’

The second mistake: asking too much before giving anything

Sign up. Verify your email. Set up your profile. Add a photo. Connect your calendar. Invite a colleague. Complete these five steps before you can start.

Every request made before the user has experienced value is a withdrawal from an account that has no deposits yet. There is no reason for the user to comply. They have no evidence that the product is worth the effort. They are being asked to invest before they have been shown a return.

The Ikea Effect, documented by Michael Norton and colleagues at Harvard Business School, shows that people place higher value on things they have invested effort in building. The implication for onboarding design is precise: you want effort after the user has experienced value, not before it. Effort before value reads as friction. Effort after value reads as investment.

The practical version of this principle: defer everything that is not strictly necessary for the user to experience a meaningful moment in the product. Profile photos, notification preferences, team invitations: none of these are prerequisites for value. Build the flow so the user reaches a result before any of those requests appear. Once they have experienced value, they are motivated. Motivated users complete setup steps. Unmotivated users abandon them.

The third mistake: designing the flow for the median user

Onboarding flows are typically designed for a single persona: the user the team imagines when they think about who they are building for. This user knows roughly what the product does. They are reasonably motivated. They have about five minutes to explore.

Real users arriving at a product for the first time are wildly heterogeneous. Some found it through a recommendation and already understand the problem it solves. Some clicked an ad and are not sure what they signed up for. Some were added by a colleague and resent being there. Some are evaluating three competing products simultaneously and are actively looking for a reason to eliminate this one.

A single linear onboarding flow serves none of these users well. The already-informed user finds the explainer condescending and skips through it. The confused user finds it too fast. The reluctant user finds it irrelevant. The evaluator finds it generic.

The most effective onboarding flows I have seen use a single early question to segment users and branch the experience accordingly. ‘What are you hoping to do first?’ is not a UX nicety. It is a routing mechanism. The user who says ‘I want to do X’ gets taken directly to X. The user who says ‘I am not sure yet’ gets a shorter, more exploratory path. Both reach a meaningful moment faster than a one-size tour would take them.

The fourth mistake: measuring completion instead of activation

Most teams track onboarding completion rate. What percentage of users who started the flow finished it? This is the wrong metric for exactly the same reason that traffic is the wrong metric for an e-commerce store. It measures what happened, not whether it mattered.

A user who completes every step of the onboarding flow and then never opens the product again has a 100% completion rate and a 0% activation rate. A user who skips step four because they spotted something interesting, went straight to it, and came back the next day has a lower completion rate and a higher activation rate. The second user is the one the product needs.

Activation is the moment a user first experiences the core value of the product. Every team should be able to name that moment specifically: not ‘when they understand the dashboard’ but ‘when they complete their first X’ or ‘when they see Y for the first time.’ If the team cannot name the activation moment, they cannot design toward it.

Once the activation moment is named, the onboarding flow has a single job: get the user to that moment as quickly and with as little friction as possible. Everything else in the flow is either contributing to that goal or in the way of it.

The fifth mistake: treating onboarding as a one-time event

Onboarding is not a set of screens the user sees once and then leaves behind. It is the entire arc of a new user’s experience through their first meaningful engagement with a product. That arc typically spans multiple sessions across multiple days.

A flow that tries to accomplish everything in session one will always fail, because users do not retain what they are shown before they understand why it matters. Information absorbed before context is established is information discarded. Samuel Hulick’s work on user onboarding, particularly his framework of the product as a guide rather than a destination, describes this precisely: the user is the hero of their own story, not a visitor in yours. The product’s role is to give them what they need when they need it, not to deliver everything at once and expect it to be retained.

Empty states are one of the most powerful and consistently underused onboarding surfaces. The moment a user sees an empty dashboard for the first time is a critical design moment. It is an opportunity to show them exactly what this space will look like when they have taken the action that fills it. Done well, an empty state is an invitation. Done badly, it is a blank room that offers no indication of what should happen next.

What good onboarding actually looks like

It does one thing first: it gets the user to a moment where the product has done something for them. Not shown them something. Done something. Produced a result. Made something visible that was not visible before. Saved them a step they previously had to take manually.

That moment does not have to be impressive. It has to be real. A user who completes their first task in a project management tool, runs their first report in an analytics product, or gets their first recommendation from a recommendation engine has experienced value. They now have a reason to return. Everything that comes after, the profile setup, the team invitation, the notification preferences, lands differently when the user arrives at it with that experience behind them.

Onboarding does not fail because users do not understand products. It fails because products do not give users a reason to care before asking them to invest.

Ammarah Ahmed is the founder of Precision Consulting, a CRO and UX practice for growth-stage e-commerce brands. She spent over a decade leading growth and product teams at Delivery Hero, Foodpanda, Talabat, and Daraz across MENA and Southeast Asia. goprecision.co

References: Fogg, B.J. (2009). A Behaviour Model for Persuasive Design. ACM. Norton, M., Mochon, D., & Ariely, D. (2011). The IKEA Effect. Journal of Consumer Psychology. Hulick, S. (2015). The Elements of User Onboarding. UserOnboard.

Why onboarding flows lose users in the first 60 seconds was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.

Here is something I noticed after auditing over 120 e-commerce funnels. Every store that was losing revenue at the product page stage had the same surface problem: the page looked fine. It had a hero image, a product title, a price, reviews, and a CTA. All the right elements. And yet buyers were not converting at the rate the funnel should have supported.

When I started looking more carefully, the issue was almost never a missing element. It was a broken hierarchy. The page was talking to the buyer in the wrong order. The CTA was visually competing with a promotional banner. The price was smaller than the product category label. The ‘Add to Cart’ button was the same visual weight as ‘Save to Wishlist’ sitting directly beside it. The page was arranged around the brand’s organisational logic, not the buyer’s decision sequence.

Visual hierarchy is one of the most well-documented principles in design. It is also one of the most consistently violated in practice, because the people making layout decisions are not experiencing the page the way a first-time visitor does. They know what matters. So they see it regardless of where it sits.

What hierarchy is actually doing

Visual hierarchy is the sequence in which a viewer processes the elements on a page. Not the sequence you intended. The sequence they actually follow, driven by size, contrast, colour, whitespace, and position.

The Gestalt psychologists documented in the 1920s that the brain groups visual information into patterns before analysing individual elements. Max Wertheimer and Wolfgang Köhler’s work established that people perceive wholes before parts, and that visual properties like proximity, similarity, and figure-ground relationships shape what gets processed in which order. That groundwork is still the backbone of every layout decision in UI design.

What this means practically: when a buyer lands on your product page, they are not reading it. They are scanning it. Their eye is following a path determined by the visual weights of the elements on the screen. If you have not designed that path deliberately, you have created one accidentally. And accidental hierarchies almost always prioritise the wrong things.

If everything demands attention, nothing gets it. The visual hierarchy problem on most product pages is not that the important elements are hidden. It is that too many elements are competing at the same visual weight, so the buyer’s eye has nowhere clear to go.

The Von Restorff effect and why your CTA keeps losing

In 1933, Hedwig von Restorff documented what is now called the isolation effect: items that visually stand out from their surroundings are more likely to be remembered and acted on. This is the scientific basis for contrasting CTA colours, for ‘Most Popular’ badges on pricing tiers, and for highlighted elements in feature tables.

The implication that most designers understand in theory but violate in practice: your primary CTA should be the only element on the page in its colour. The moment another element shares that colour, the CTA’s isolation is broken. The visual signal that says ‘this is the thing to do’ gets diluted.

I audit product pages where the ‘Add to Cart’ button is the same colour as a sale badge, a trust seal, and a category navigation tag. None of those decisions were made together. Each was made independently, and each made sense in isolation. Accumulated, they created a page where the primary conversion action is visually indistinguishable from secondary and tertiary elements.

The fix is subtractive, not additive. Reserve the primary CTA colour for the primary CTA only. Reduce the visual weight of everything competing with it. When you lower the volume on secondary elements, the CTA gets louder without changing anything about it.

Hierarchy and the sequence of trust

The order in which elements appear on a page is not just an aesthetic decision. It is a decision about which questions you answer first.

On a product page, a buyer has a sequence of questions that must be resolved before they will purchase. What is this? Is it right for me? Can I trust this seller? What does it cost? What happens if something goes wrong? These are not equal-priority questions. They are sequential. A buyer does not move to the cost question until they have resolved the ‘is this right for me’ question. They do not move to the trust question until they know what the product is.

Most product pages are hierarchically sequenced around the brand’s structure, not the buyer’s question sequence. The brand logo is prominent. The category navigation is large. The promotional banner is high on the page. These are not the buyer’s first questions. They are the brand’s first statements about itself.

The purchase zone, which is the cluster of elements the buyer needs to decide, should be visible without scrolling on any device. Product name, price, star rating with review count, a delivery or returns signal, and the Add to Cart button. That is the sequence that maps to the buyer’s decision process. Everything else is secondary.

Where reviews appear relative to product descriptions is a small but measurable example of this. Reviews answer trust questions. Product descriptions answer detail questions. Trust questions come earlier in the decision sequence. If reviews appear after the description, the page is answering detail questions before trust questions. Most buyers who leave at the product page level were not ready to answer the detail question yet. They were still resolving trust.

Whitespace is not about aesthetics

I want to address whitespace specifically because it is the element most consistently justified aesthetically and most consistently misunderstood functionally.

Whitespace controls visual weight. An element with space around it draws more attention than an identical element surrounded by competing content. This is not a style preference. It is how visual salience works. When you give a CTA more whitespace, you are increasing its visual weight without changing anything about the element itself. When you compress elements together to fit more content above the fold, you are reducing the visual weight of every element on the page simultaneously.

Research published in 2026 by Loopex Digital found that minimal, content-first layouts increase time-on-page by roughly 59% compared to cluttered alternatives, with cluttered designs showing bounce rates approximately 50% higher. The mechanism is not that cluttered pages look bad. It is that they create flat hierarchies where the eye has no clear place to go and defaults to leaving.

The most common whitespace error I see is the assumption that more content above the fold is better because it shows buyers more. It is not better. It shows buyers more at the cost of showing them anything clearly. A single product image, a product title, a price, and an Add to Cart button in a clean layout will outperform a hero image, a badge strip, a feature summary, an offer banner, and a CTA packed into the same space, because the first arrangement answers one question clearly and the second arrangement answers five questions simultaneously at equal volume.

Colour hierarchy and the two rules worth keeping

Colour is the most powerful and most misused tool in visual hierarchy. Most design systems accumulate colours over time as features get added, promotions run, and elements get branded. By the time someone runs an audit, the page has six or seven distinct colours, none of which are playing a clear hierarchical role.

Two rules are worth applying before any others. First: build a grayscale version of the page first. If the hierarchy does not work in grayscale, colour will not fix it. Colour adds emphasis; it does not create structure. A page that is hierarchically correct in grayscale becomes more hierarchically correct when colour is added. A page that is flat in grayscale remains flat when colour is added.

Second: assign each colour a tier. Primary colour for primary actions only. Secondary colour for secondary elements. Tertiary and neutrals for everything else. When a new element needs a colour, ask which tier it belongs to and use the colour for that tier. This sounds mechanical, and it is. That is the point. Colour hierarchy decisions made case-by-case accumulate into incoherence. Colour decisions made against a tier system accumulate into clarity.

A three-tier colour hierarchy applied to a product page: primary for the CTA only, secondary for trust signals and scarcity indicators, neutrals for everything else. When the tiers are respected, the buyer’s eye follows the intended path.

What I changed my mind about

I came into conversion work from product and growth, not from design. My instinct for a long time was that visual hierarchy was a downstream concern: you sorted the strategy first, then the design team made it look right.

What I changed my mind about is the sequencing. Visual hierarchy is not downstream of the product or content decision. It is part of the same decision. The order in which elements are arranged on a page is a statement about what matters. Getting the order wrong means getting the product strategy wrong at the point where it meets the buyer. You can have a strong product, strong copy, and correct placement of trust signals, and still underperform because the visual hierarchy is telling the buyer to look at the wrong things first.

Every layout is a priority statement. The question is whether you made it deliberately or accidentally.

Ammarah Ahmed is the founder of Precision Consulting, a CRO and UX practice for growth-stage e-commerce brands. She has audited over 120 e-commerce funnels across MENA and Southeast Asia over the past decade. goprecision.co

References: Wertheimer, M. (1923). Laws of Organization in Perceptual Forms. Von Restorff, H. (1933). Über die Wirkung von Bereichsbildungen im Spurenfeld. Yablonski, J. (2020). Laws of UX. O’Reilly Media. Loopex Digital research on layout and viewing time (2026).

Visual hierarchy is not a design decision. It is a revenue decision. was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.

A UX audit that starts with a checklist is not a UX audit. It is a presence check. Checklists tell you whether the right elements exist. They do not tell you whether the elements that exist actually function.

The distinction matters more than most design reviews acknowledge. A returns policy placed correctly in the purchase zone but written in legal language no buyer reads is not a functioning trust signal. A star rating above the fold that takes three seconds to render does not contribute to conversion. An Add to Cart button in a ghost outline style with no fill does not communicate ‘click here’ to a buyer who is uncertain. All three would pass a standard checklist. None of them is doing the job.

I have run structured UX audits on over 120 e-commerce funnels. The single most common finding is not missing elements. It is elements that are present but broken at the functional level, and nobody has tested them as a first-time buyer would encounter them because the team has stopped being able to see their own store clearly.

Here is what a UX audit is actually evaluating, and where the gap between presence and function shows up most reliably.

The starting point: Jakob’s Law and the translation tax

Users spend most of their time on other sites. This means they arrive at yours with a set of expectations formed elsewhere, and they judge your interface not against what you intended but against what they already know. Jon Yablonski documents this in Laws of UX as Jakob’s Law, and its implications for navigation design are direct: every label, pattern, or interaction model that deviates from convention asks the buyer to translate before they have even started.

‘Discover’ instead of ‘Shop’. ‘Our World’ instead of ‘About’. ‘The Edit’ instead of ‘New Arrivals’. These are not just naming choices. They are cognitive costs paid upfront, before the buyer has a reason to stay. The translation tax is real, and it compounds with every unconventional element that follows.

The audit question here is not ‘is there a navigation?’ but ‘does this navigation require translation?’ Run five seconds with a timer. Before it ends, ask whether someone who has never heard of this brand can tell what the store sells and who it is for. If the answer is not immediate, you have a first-impression problem that predates every other finding.

The five-second test is not a design preference exercise. It is a test of whether the interface works within the attention window a real visitor actually grants it. Nielsen Norman Group’s research puts the average decision window at 10 to 20 seconds before a visitor commits to staying or leaves.

Where presence and function diverge on the product page

The product page is where the gap between presence and function is most consequential, because it is where buying decisions happen. Most product pages pass a checklist review. Most of them are still losing buyers at the point of decision.

Mango’s mobile product page is a useful case study because the failures are specific and instructive rather than catastrophic. The product image fills the entire screen. To see the product name, price, or any purchase action, the buyer must scroll. When they do, the CTA reads ‘ADD’. Not ‘Add to Bag’. Not ‘Add to Cart’. Just ‘ADD’, stripped of the value exchange framing that makes a call to action functional. There is no star rating visible anywhere on the product page.

Directly below the purchase section, before reviews appear, a cross-sell block shows the full outfit. The cross-sell mechanic is sound in principle. The placement is the problem. When complementary products appear before the trust signals a buyer needs to confirm their main purchase, the mechanic designed to grow the basket creates doubt about the item already in it.

Compare that to Ugmonk. The price appears inside the CTA itself: ‘Add to Cart ($159)’. The buyer can see the price and take action in a single element. A scarcity signal sits directly below the button. Feature checkpoints provide trust information without requiring a scroll. Not every store can match Ugmonk’s design execution. Every store can apply Ugmonk’s sequencing logic, because the sequencing is a functional decision about what the buyer needs to know and when, not an aesthetic one.

The audit question is the scroll test. Before a scroll, list every element visible on the product page on mobile. Price, star rating, primary CTA, and at least one trust signal should all be in that list. If any are missing, the page is asking the buyer to invest effort before they have decided the product is worth their attention.

The checkout audit and why most designers have never run it honestly

The Baymard Institute’s benchmark for an optimised checkout is 12 to 14 form elements. The web average is 23. Most design teams know this. Most of them have not personally completed their own checkout flow on a mobile device, without autofill, as a first-time buyer, from the cart page to the confirmation screen.

When that test gets run, specific things happen. A postcode field opens a letter keyboard instead of a number pad. An email field is missing the @ symbol on the first keyboard view. Address fields require manual entry on a screen that offers no autocomplete. Express payment options exist but are positioned below the card entry form, which means they are not functioning as express options because the buyer has already been asked to type before they are offered the alternative.

These are not minor details. They are interaction failures that break the completion flow at the moment of highest intent. A buyer who has to switch keyboard types mid-form pauses. A pause prompts re-evaluation. Re-evaluation on a purchase is frequently abandonment. Multiplied across real traffic, these moments represent a measurable gap in checkout completion that does not appear in analytics as a UX failure. It appears as a number.

The express payment finding deserves specific attention. Apple Pay, Google Pay, and Shop Pay replace every form field with two taps. From the buyer’s position, the difference between two taps and a three-minute keyboard entry is often the difference between a completed purchase and an abandoned one. A store that offers these options but positions them below the card form is not offering them at the decision moment. It is offering them as an afterthought the buyer has to discover after the friction has already occurred.

Mobile is where the audit becomes physical

The most revealing part of a mobile UX audit is the thumb-only test. Navigate the entire store using only one thumb, one-handed, as most people actually hold a phone. Anything missed or hit by accident is a tap target problem. This test takes ten minutes and consistently surfaces issues that no desktop review finds, because the constraints are physical and the failures are immediate.

Fitts’s Law describes the relationship between target size, distance, and the time required to reach it accurately. On a phone, small interactive elements are not just difficult to tap. They are systematically missed at a rate that varies by finger size, phone size, and holding position. The minimum for reliable tap accuracy is 44 by 44 pixels. The most common offenders are filter chips on category pages, close buttons on modals, and quantity adjusters on product pages. All of these receive significant interaction volume. None of them are usually built to the minimum.

The filter experience on a category page also deserves its own test. Apply a filter. Observe whether the page updates in place or reloads from the top. A page reload that returns the buyer to the start of a category page is not a bug. It is a design decision that has not been tested against real buyer behaviour. The buyer found the category, scrolled to establish context, applied a filter, and the page reset. Most do not restart. They exit.

Edge cases as the fastest signal of intentional design

The edge case audit is the part of a UX review that most teams skip, because edge cases feel like exception handling rather than experience design. They are not. They are the fastest signal of whether experience design was intentional or incidental.

Search for a product that does not exist. The empty result page with no navigation, no suggestions, and no path forward is not a neutral experience. It is a negative one. The buyer expressed intent and received nothing. The question is not whether this path is common. It is whether the design team has thought through what happens when it occurs.

Navigate to a sold-out product. If there is no back-in-stock notification, that is a buyer who raised their hand and was given no mechanism to follow through. Enter an invalid discount code and read the error message. ‘Code not valid’ is a system output. ‘This code has expired, check your email for the current offer’ is an experience decision. Remove an item from the cart. Whether the cart updates without a page reload is an implementation detail that becomes a UX decision when the buyer uses it mid-session.

Edge cases are disproportionately revealing because they are the paths a store’s designers were not thinking about when they built the main flow. Intentional design extends to them. Incidental design stops where the happy path ends.

What the audit produces and how to use it

After a structured review, the output is a list. The error most teams make at this stage is treating every item as an equal priority fix. It is not a UX audit problem. It is a triage problem.

Prioritise by funnel stage and buyer frequency. Checkout and product page findings are first-tier: they affect every buyer on every visit. Category and homepage findings are second-tier. Navigation, edge cases, and empty states are third-tier unless analytics show they are causing a measurable drop-off. Checkout friction affecting 100% of buyers is not the same finding as a broken empty-state page affecting 3%, even if both represent genuine UX failures.

The audit’s value is not the length of the list. It is identifying the two or three things that, if fixed, change the experience for the largest number of buyers in the next sprint. Everything else belongs in the backlog, not the current conversation.

One last thing worth naming: the most honest UX audit is run by someone who has not built the store. Steve Krug’s research in Rocket Surgery Made Easy shows that three people attempting defined tasks on a site surface most of the major usability problems. They do not need to be professional researchers. They need to not know the store, have a phone, and be asked to buy something while thinking aloud. That combination reveals more in 90 minutes than most internal reviews find in a quarter.

Ammarah Ahmed is the founder of Precision Consulting, a CRO and UX practice for growth-stage e-commerce brands. She has audited over 120 e-commerce funnels across MENA and Southeast Asia over the past decade.

References: Yablonski, J. (2020). Laws of UX. O’Reilly Media. Krug, S. (2009). Rocket Surgery Made Easy. New Riders. Baymard Institute checkout research: baymard.com/lists/cart-abandonment-rate. Nielsen Norman Group: nngroup.com/articles/how-long-do-users-stay-on-web-pages.

What a UX Audit Is Actually Testing (And Why Most Designers Get It Wrong) was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.

There is a particular kind of founder delusion that is hard to catch because it looks exactly like focus.

You pick a number. You put it on the dashboard. You build your week around moving it. The team knows the number. The investors get updates on the number. You feel like you are running a real business because you are tracking a real metric.

The problem, which I did not fully understand until I had spent years inside some of the largest e-commerce operations in the Middle East and Southeast Asia, is that the metric you choose to track shapes every decision you make. Choose the wrong one and you can spend years optimising for something that has almost no relationship to what you actually need to grow.

I have seen this at scale. I have done it myself. Here is what it looks like from the inside.

The number that felt important

In e-commerce, the metric that attracts the most attention is traffic. Sessions. Visitors. Monthly active users. The dashboard number that goes up when your ads are working and down when they are not.

Traffic is not a vanity metric in the pejorative sense. It is a real input. But it becomes dangerous when it becomes the primary number because it tells you nothing about what happens after someone arrives.

I spent time in a business where traffic was growing steadily. Every month, the curve went up. The team was energised. The campaigns were performing. Then someone pulled the revenue-per-session figure, which almost nobody had looked at, and the picture inverted. Revenue per session was declining. We were getting more visitors and extracting less value from each one. The traffic growth was masking a conversion problem that had been compounding for months.

The awkward part: the conversion problem was not subtle. It was sitting in the checkout funnel in plain sight. We had added steps to the checkout process to collect more customer data. Every step added was a decision point where a buyer could change their mind. The team knew this in principle. But because nobody was looking at the revenue-per-session number, the cumulative effect had not registered as a crisis. It registered as a data collection improvement.

What the wrong metric does to a team

This is the part that I think gets underappreciated in conversations about metrics and measurement.

A metric is not just a number. It is a signal about what the organisation values. When traffic is the primary number, the team optimises for traffic. Ad spend goes toward reach rather than intent. Content gets produced for volume rather than quality. Features get built to bring people in rather than to help the people who are already there.

Every one of those decisions is locally rational. It moves the metric you are tracking. The problem is that none of them necessarily move revenue, and in some cases they actively work against it.

At one of the platforms I worked at, we had a metric for the number of restaurant partners listed on the app. More listings meant more choice, which we believed meant more orders. The metric went up for eighteen months. Then we ran an analysis on order completion rates by city and found that the cities with the most restaurant options had the lowest order completion rates. The choice was overwhelming buyers. Hick’s Law operates at the platform level too: more options, more paralysis, fewer decisions.

We had spent eighteen months and significant resources optimising for a metric that was negatively correlated with the outcome we wanted.

The replacement question

The shift that actually changed how I think about this is deceptively simple. Instead of asking ‘what metric should we track?’ I started asking ‘what is the decision this metric is supposed to inform?’

Traffic informs acquisition decisions: where to spend, which channels to prioritise, whether the top of the funnel is healthy. It is the right metric for those decisions. It is the wrong metric for decisions about whether your product is working.

Revenue per session is a better metric for the latter because it collapses conversion rate and average order value into a single number. It tells you what each visitor is actually worth, not how many of them arrived. A business with 50,000 monthly sessions and a revenue per session of £3.20 understands its lever differently than a business that tracks sessions and AOV separately and has never multiplied them together.

The metric you choose should force the right question. Traffic forces the question ‘how do we get more people here?’ Revenue per session forces the question ‘how do we make the experience worth more for the people already here?’ Those are different problems with different solutions, different team priorities, and different investment decisions.

The version of this that founders get wrong most often

I run a conversion optimisation consultancy now, and I work with growth-stage e-commerce founders. The version of this problem I see most often is not the dramatic case of the wrong metric hiding a declining business. It is the quieter case of the right metrics not being looked at together.

A founder knows their conversion rate. They know their AOV. They know their traffic. But they have never calculated what a 1% improvement in conversion rate is worth in annual revenue, held constant everything else. So when the engineering team says the checkout redesign will take eight weeks and the growth team says the new ad campaign can launch in two, the ad campaign wins by default. Not because the founder decided that traffic was more valuable than conversion. Because the comparison was never made in the numbers.

On a store doing £75,000 in monthly revenue, a 40% improvement in conversion rate is worth £360,000 per year. A 35% improvement in average order value on the same baseline, compounded with the conversion improvement, produces over £800,000 in annual revenue uplift. Without a single additional visitor.

These are not complicated calculations. They take about ten minutes. But most founders have never done them, because they have never been prompted to, because the metric dashboard they look at every morning does not surface the question.

What I would do differently

If I were building the measurement framework from scratch, I would start with one question: what does it look like when the business is healthy? Then I would work backward to find the smallest number of metrics that together answer that question, and I would resist adding any metric that does not directly inform a decision I am making in the next 90 days.

For most e-commerce businesses, that is: revenue per session (product working), contribution margin per order (unit economics sound), repeat purchase rate (customers coming back), and checkout completion rate (no obvious friction). Everything else is a supporting metric that you pull when one of those four moves in the wrong direction.

The dashboard should prompt questions, not provide comfort. If you look at it every morning and feel reassured, it is probably showing you the number that is easiest to move rather than the number that matters most.

I do not think this is a metrics problem, fundamentally. It is an incentive problem. The metric that gets tracked is usually the metric that someone is compensated to move, or the metric that is easiest to show in a board deck, or the metric that went up last month when everything else felt uncertain.

Founders are not less analytical than operators. They are just operating with fewer resources, more pressure, and a natural tendency to reach for the number that confirms the business is moving. That is a human response to an uncertain situation, not a failure of rigour.

The rigour comes in building the habit of asking, once a quarter at minimum: is this still the right number? Is it telling me what I think it is telling me? And is there a number I am not looking at that would change the decisions I am making?

Usually there is.

Ammarah Ahmed is the founder of Precision Consulting (goprecision.co), a CRO and UX consultancy for growth-stage e-commerce brands. She spent over a decade leading growth and product teams at Delivery Hero, Foodpanda, Talabat, and Daraz across MENA and Southeast Asia.

The Metric That Felt Like Progress Was the One Costing Me the Most was originally published in Entrepreneurship Handbook on Medium, where people are continuing the conversation by highlighting and responding to this story.

Most organisations treat conversion rate optimisation as a marketing function. It lives with the growth team, reports into the CMO, and gets measured by traffic and click-through rates. The result, in my experience, is a discipline that produces a lot of activity and not enough diagnosis.

Business analysts are better placed to run a CRO audit than most practitioners in the field. The core competencies are already there: structured problem decomposition, stakeholder interviews, process mapping, requirements elicitation, and a bias toward data over opinion. What is usually missing is a framework for applying those competencies to conversion problems specifically.

This essay outlines the framework I use. It draws on over a decade of running growth programmes at major e-commerce platforms across Asia and the Middle East, and on the audit methodology I have developed since founding a conversion optimisation consultancy. The structure is deliberately analytical rather than tactical.

Start with a model, not a checklist

The most common mistake in CRO audits is starting with a list of things to check. Button colours, headline copy, image placement. These are outputs, not inputs. Checking them without a model of how your funnel works is the analytical equivalent of reviewing requirements without a use case.

Before touching any page, map the conversion funnel as a process flow. Every visitor entering your funnel follows a sequence of states: awareness, consideration, intent, action, confirmation. At each transition between states, some percentage of visitors do not move forward. The audit’s job is to identify where the drop-off is largest, understand why it is happening, and prescribe interventions with a clear causal mechanism.

This framing resets the question from ‘what should we change?’ to ‘where is the system failing and why?’ The first question produces opinions. The second produces hypotheses you can test.

Phase 1: Establish the baseline

Before any analysis, you need three numbers at every funnel stage: the entry volume, the exit volume, and the conversion rate between them. Most organisations have these numbers in their analytics platforms but have not assembled them into a single view.

Build a funnel table: landing page to product page, product page to cart, cart to checkout initiation, checkout initiation to purchase confirmation. For each stage, calculate the drop-off rate and rank the stages by absolute revenue impact. A 5% improvement in checkout completion on a store doing 10,000 orders per month is worth more than a 20% improvement in landing-to-product-page click-through on the same store, if the order values differ significantly. Always convert funnel leaks into revenue terms before prioritising them.

The goal is to identify the stage where the largest volume of motivated buyers are exiting. Not the stage with the lowest conversion rate. Those are often different things.

Segment the baseline data by device type, traffic source, and new versus returning visitors before drawing any conclusions. A checkout completion rate of 68% that is actually 81% on desktop and 54% on mobile is a mobile problem, not a checkout problem. The intervention is different. So is the priority.

Phase 2: Qualitative diagnosis

Quantitative data tells you where people are leaving. It does not tell you why. The qualitative phase answers the why.

Three inputs are most useful here. The first is session recordings: actual recordings of visitor behaviour on the pages with the highest drop-off rates. Watch for rage clicks (repeated tapping on non-interactive elements, which indicate failed expectations), scroll depth anomalies (pages where visitors stop scrolling before reaching key content), and hesitation patterns at form fields or CTAs. Tools like Microsoft Clarity provide this at no cost.

The second input is exit-intent survey data. A single open-text question placed at checkout exit, ‘What stopped you from completing your order today?’, consistently surfaces friction points that do not appear in quantitative data. Unexpected shipping costs, distrust of the payment method, and confusion about returns policies are the three most common responses in my audits. None of these are visible in session metrics alone.

The third input is a structured heuristic evaluation of the pages with the highest drop-off. Baymard Institute’s checkout research provides the most rigorous publicly available benchmark: the average e-commerce checkout has 23 form elements, against an optimised benchmark of 12 to 14. Applying a structured heuristic against that benchmark identifies friction that visitors experience but rarely articulate in surveys.

Phase 3: Hypothesis construction

Each significant finding from phases 1 and 2 becomes a hypothesis in the following format: we believe that [changing X] will result in [outcome Y] because [mechanism Z]. The mechanism is the critical part. It is what distinguishes a hypothesis from a guess.

An example: we believe that displaying the free shipping threshold gap dynamically on the cart page (‘You are £8 away from free shipping’) will increase average order value by 10 to 15% because it activates the Goal Gradient Effect, the documented tendency for people to accelerate effort toward a visible and achievable goal.

The mechanism citation is not decorative. It is what tells you whether the hypothesis is worth testing and what result you should expect. A hypothesis without a mechanism is a preference with a test attached.

Prioritise hypotheses using an effort-impact matrix. Impact is the estimated revenue uplift if the hypothesis is correct; estimate this by applying the expected conversion improvement to your baseline revenue at that funnel stage. Effort is the implementation cost in engineering time. Hypotheses in the high-impact, low-effort quadrant go first.

Phase 4: Test design and measurement

CRO testing has a statistical validity requirement that most organisations underestimate. An A/B test requires a minimum detectable effect, a significance threshold (typically 95%), and a sample size derived from both. Running a test to 200 conversions and declaring a winner is the equivalent of drawing conclusions from an insufficiently sized requirements sample. The finding is not reliable.

Calculate your required sample size before running any test. For most e-commerce stores doing fewer than 1,000 monthly conversions, the sample size for a properly powered test exceeds available traffic. In these cases, sequential testing (version A for two weeks, version B for two weeks, direct comparison) is a pragmatic alternative. It is not statistically equivalent to a simultaneous A/B test, but it replaces pure opinion with structured observation, which is a meaningful step forward.

Define your primary metric and your guardrail metrics before the test starts. The primary metric is what you are trying to move. Guardrail metrics are signals that would indicate harm if they moved in the wrong direction: average order value, return rate, customer satisfaction score. A change that improves checkout completion but doubles the return rate has not produced a net positive result.

Phase 5: Documentation and iteration

Each completed test, whether it wins or loses, belongs in a shared test log with the following fields: hypothesis, mechanism, variant description, sample size, result, confidence level, and what you learned. A losing test is not a failure. It is evidence that the mechanism you assumed was less powerful than expected, or that the friction you identified was not the primary cause of drop-off. Both findings have value for the next hypothesis.

Over time, the test log becomes an institutional knowledge base. Organisations with mature CRO programmes do not start each audit from scratch. They interrogate the log first: what have we already tested on this page? What mechanisms have we confirmed? Where are there still open questions? This is the analytical equivalent of reusing validated components rather than rebuilding from requirements every time.

The analyst’s advantage

CRO done well is applied behavioural science inside a structured analytical process. It requires comfort with ambiguous data, disciplined hypothesis construction, statistical literacy, and the ability to translate quantitative findings into prioritised interventions. These are analyst competencies.

The practitioner who approaches CRO as a checklist of best practices will always be outperformed by the one who approaches it as a system with measurable states, identifiable failure modes, and testable interventions. The framework above is not new. It is the same decompose-diagnose-hypothesise-test-document cycle that appears in business analysis methodology applied to a conversion problem.

The difference is that most conversion optimisation practice has not named it clearly enough for analysts to recognise it as home territory.

Ammarah Ahmed is the founder of Precision Consulting, a conversion rate optimisation and UX practice for growth-stage e-commerce brands. She spent over a decade leading growth and product teams at Delivery Hero, Foodpanda, Talabat, and Daraz across MENA and Southeast Asia.

Further reading: Baymard Institute checkout research (baymard.com/lists/cart-abandonment-rate). Ariely, D. (2008). Predictably Irrational. HarperCollins. Kivetz, R., Urminsky, O., & Zheng, Y. (2006). The Goal-Gradient Hypothesis Resurrected. Journal of Marketing Research.

How to Structure a CRO Audit: A Business Analyst’s Framework was originally published in Analyst’s corner on Medium, where people are continuing the conversation by highlighting and responding to this story.