The only board missing in your AI program is an Ouija Board
To ask where your AI ROI went after everyone approved it.
This, my smart impatient friend, is a TechTonicShifts field report straight from the European corporate wilderness. And in the trenches, things move at the speed of calendar availability and every good idea must first be ritually cleansed by seventeen governance bodies.
More rants after the messages
Visit the ATLAS website. It’s free. I built it for myself. And now you can use it too.
- Connect with me on Linkedin 🙏
- Subscribe to TechTonic Shifts to get your daily dose of tech 📰
- Please comment, like or clap the article. Whatever you fancy.
Meanwhile, in America, they had a kickoff on Monday
A few months ago, a vendor told me one of those stories that will stay with you throughout this blog. A company in the United States had seen an AI demonstration late in the week. The demonstration worked, as it always does, and the people in the room understood the opportunity.
Senior leadership also happened to like it, so by Monday, the important people had been pulled into a kickoff, ownership had been assigned, funding had entered the conversation, and the company had begun moving.
In the meeting, there were signs of people cheering. There may have also been some table dancing, because when discussing American corporate mythology, you tend to add furniture abuse when enthusiasm needs extra punctuation.
When you tell the same story inside any large European company, you will see the room develop a ‘facial expression’ that is normally reserved for someone proposing, say, indoor falconry. People, of course, like the speed in principle, and they admire the decisiveness and may even use words such as entrepreneurial and outcome-driven.
Then somebody mentions the word procurement and the spell breaks. The demonstration disappears behind an RfP template and a governance calendar and enough stakeholder alignment to make continental drift look reckless.
This is how an AI transformation often begins in Europe. It does not begin with AI, mind you. It begins with . . . choreography.
The technology may be probabilistic and adaptive and capable of processing more information in seconds than a management team can consume in a quarter, but the organization surrounding it is deeply committed to a more deterministic computational model.
Every action needs to pass through a human workflow built out of meetings and the holy corporate belief that adding another decision body makes decisions safer. And in the end, the organization has built a sluggish governance machine around a technology whose economic value depends heavily on speed and iteration. A
nd that machine then behaves exactly as designed. It slows everything down, dilutes accountability until it becomes homeopathic, and it produces evidence that everyone was involved when the value finally evaporates.
The RfP. Urgency goes to die here.
Procurement usually gets the first dance.
There is a good reason why you have something like procurement. Large companies should not hand millions to the first vendor wearing a neural network with a blazer. Commercial terms happen to matter, including the occasional exit clause, and anyone who has watched a technology supplier convert a proof of concept into a ten-year hostage situation knows that procurement is the protagonist in this story.
But in Europe, this otherwise useful control mechanism turns into the operating system for discovery itself.
The RfP has to be written before the organization fully understands the thing it is buying. People who have never operated an agentic system are asked to define detailed requirements for an agentic system, so requirements written for traditional software procurement get copied into a document about probabilistic systems, and everyone knows that the requirements will move the moment the real users touch the thing, yet the document still acquires the ceremonial authority of scripture because a template existeth, and thou shalt filleth the template.
Then summer arrives.
A European transformation program can survive a data breach and an executive reshuffle but, man, August is another category of threat. Entire organizational layers then vanish into France or Italy and small cottages in places where Teams connectivity is blocked by local wildlife.
The project is nonetheless still deemed strategic, and it is sponsored at the highest level, but unfortunately, Legal is back in three weeks, and Finance is away for four; the architecture lead has been handed over to someone who was not in the earlier meetings, and the director who wanted the whole thing done yesterday has activated his out-of-office message.
This is when you, as the responsible schmuck, have to start with stakeholder archaeology. Every manager must be involved, or at least must later be unable to claim they were excluded. The latter matters in Europe because exclusion is one of the most powerful fuels in enterprise politics. The manager who was not invited to an early workshop does not merely lack context. Noooooo, they acquire a renewable license to object and whine.
Six months later, while everyone else is discussing deployment, they can still lower their glasses and say that they were never properly involved in the initial process. No one asks if their missing involvement would have changed a darn thing at all, but it simply accepts the complaint as a kind of procedural original sin and reluctantly schedules another meeting to bring the manager up to speed and give him his ceremonial moment of importance.
And when the RfP reaches a state that can survive a review, three suppliers are invited. Three is the traditional enterprise magical number because two looks suspicious and four creates unnecessary diary problems, unless specifically chosen, because a ‘wildcard’ gives the comforting illusion of strategic courage.
Then it’s demo day. Or days, turning into weeks, even, because suppliers are seen on separate days because allowing them to encounter each other at the reception desk just might reveal that they are . . . in a competition. This could apparently disturb the ‘natural’ ecosystem where everyone is equal, and there is never a winner.
The results then enter the scoring phase, and here, the consultant expresses uncertainty through decimals.
A professional RfP consultant converts the observations into a beautiful matrix.
Criteria are agreed by all present, and weights have been debated. Strategic fit gets a percentage. Security has another one. Functional coverage receives the largest percentage of all. And then there’s also some left for Architecture and Cost, and . . . oh yeah, Security. Somebody points out that innovation should count more. Somebody else then rebutes that innovation is already embedded in the functional coverage.
Eventually, a workshop is booked to settle the matter because basic calculus is apparently a stakeholder engagement problem.
But in the end, the recommendation is ready for the steering committee.
Your first real steering committee
This is the moment in time where the organization reveals what it truly believes about authority.
The experts have done the work and procurement has tested the commercials, and as usual, security has raised its concerns; then the program team has assembled the evidence. The decision now travels upward to people whose primary qualification is that the decision is large enough to be dangerous to their career.
The committee meets on the first Thursday of the month because this is apparently when strategy happens. Oooh, shit, one member is missing. Another has to leave after twenty minutes. Oh man, a third has read the pre-read. This person is regarded with suspicion.
The project team presents their recommendation.
The room agrees that AI is strategically important and everyone repeats that speed matters. Then the questions begin. One executive wants more clarity on operating model ownership. The other wants a stronger link to the enterprise data strategy or whatever.
Some other asks a good question if a solution already exists elsewhere in the company, and a leader who has not previously appeared in the process introduces a vague concern about vendor lock-in. The meeting ends with broad support, subject to a few refinements.
Broad support is corporate language for no.
The deck goes back into surgery. You, as the project manager, begin hunting senior people through Outlook like you’re some Victorian naturalist tracking rare birds. Comments arrive in fragments. One executive replies to an old version of your email, and another forwards the whole deck to three colleagues who were not part of the process and asks for their thoughts.
A new concern appears regarding a transformation program that came out of thin air. The recommendation is rewritten, recirculated, and prepared for the next governance slot.
The program is now three months in the making.
When the decision finally lands, everyone from the team who contributed to the outcome experiences a modest wave of accomplishment. People congratulate each other if at all. There are phrases about reaching an important milestone because the steering committee has approved the preferred vendor.
And months will have passed before anything gets deployed, yet the emotional machinery behaves as if a product has already shipped. This makes sense inside a system that rewards decision events. A signed decision therefore generates a cleaner burst of organizational dopamine than a working system ever could.
The original enthusiasm is already in intensive care.
The demonstration that created the momentum has faded into memory. Meanwhile, the executive sponsor has seen six other AI demonstrations, the vendor’s sales rep has been replaced again by another because he wasn’t able to close the deal in time, and the business teams have returned to quarterly targets. The problem that felt urgent has been absorbed by the calendar.
Shit. Every department has an innovation board now
Then someone discovers that every department has an innovation board.
Large organizations do not have one innovation process but instead they have habitats. Finance has an innovation council, where in Operations it is called the digital board. HR then has something like a transformation forum and IT of course has their regular demand process.
Each local structure emerged for a some historical reason and they developed their own terminology and intake mechanism with a prioritization method and a collection of people who are deeply certain that their version is the grown-up one.
Your European AI initiative now needs to have a passport.
Each forum asks a slightly different version of the same questions. What problem are we solving, and what value does it create, and the project team answers faithfully, copying material across PowerPoint templates and gradually developing the administrative efficiency of a small customs office.
And every forum also wants evidence that the other forums are aligned, so you get stuck into a distributed merry-go-round. Innovation wants architecture input, whereas Architecture wants Security input.
Security of course, needs the data classification, but Data wants a confirmed solution design which in turn depends on vendor selection, that waits on procurement approval who need evidence of budget that, in turn, depends on portfolio prioritization. But Portfolio only wants confirmation that architecture will approve the initiative before they can approve the budget.
This is a responsible behavior circular dependency graph. It isn’t that anyone is blocking the project or anything like that, they all are simply waiting for someone else to complete the prerequisite required to approve something.
This is when enterprise paralysis starts. Across all vertebrae.
Fantastic. We’re so mature we have a portfolio board
Soon someone notices that innovation has become large enough to require portfolio management.
The portfolio board now compares your AI transformation with every other strategic initiative who is competing for budget and attention. This sounds healthy because it is healthy. Capital is finite and so are people are finite but the comedy lies in the comparison mechanism.
A probabilistic AI capability whose value depends on experimentation is compared to an ERP upgrade whose scope can be estimated through decades of implementation history. A platform investment is plotted against a local cost-reduction initiative. Numbers create an appearance of comparability long before the underlying economics deserve it.
And by now, every sponsor has learned the game. No one is dumb enough to submit an initiative described as useful but ‘optional’. Everything has to be strategic, regulatory, critical, transformational or customer essential.
So, reducing the number of priorities therefore requires a political decision, so the organisation often performs a safer operation and labels twenty-seven things Priority One. This allows everybody to win until resources are assigned, at which point mathematics returns with a knife.
Your AI initiative survives and moves onward to architecture.
Get Marco van Hurne’s stories in your inbox
Join Medium for free to get updates from this writer.
Yeah, the architecture board . . .
Architecture matters enormously in AI because the new stack touches every system and other thing in the business, and poor architectural choices can create spectacular long-term damage. The problem is only when architecture turns into a gate that validates conformance without carrying equal accountability for business delay.
The conversation often begins with an approved technology stack.
Approved stacks are comforting because they simplify support, but they also age, especially in AI, where it has the inconvenient habit of evolving faster than enterprise standards committees. A capability that was peripheral eighteen months ago may now be central, and before you know it, the organization faces an awkward choice between using the best technology or preserving architectural consistency.
Large enterprises often resolve this tension by creating . . . an exception process.
Security then arrives with legitimate concerns and enough questionnaires to flatten a medium-sized mammal.
By the way, AI deserves serious security treatment because autonomous action creates failure modes that ordinary chatbots do not possess, and if you’re deploying AI without security controls, your entire incident report is probabilistic. But as usual, Security steps in when the choice has already been made and when the control system cannot distinguish between risk classes.
Then, a small pilot can inherit the assessment burden of a global production platform, or the employee who wanted an agent to extract six numbers out of a PDF reaches question seventy-three and discovers that they are now personally responsible for the security posture of the entire company. At this point your program has spent a remarkable amount of human effort proving that the machine might save human effort.
Speaking of which . . .
Welcome to the AI program
Then the AI program appears on the portfolio.
The company has realized that scattered AI projects create duplicate costs and chaos. A central program is created to bring order. This is another one of those sensible moves because shared tooling does matter and so does consistent governance, but runtime controls should not be invented by whoever had a free Friday afternoon.
The central program therefore introduces . . . an intake process.
Your initiative has now passed business intake, procurement, steering, innovation review, portfolio review, architecture, and security. It now needs to become an AI use case in the official AI use-case portfolio. And this starts with a fresh form that asks about business value, sponsor, data, risk, architecture, users, model type, autonomy, expected benefits and delivery readiness. These are good questions. You only have answered most of them six times already.
That intake funnel develops the same immune system as the rest of the enterprise. Easy use cases move through because they are easy to explain and easy to govern, whereas complex, high-value transformation cases acquire dependencies and the merry-go-round.
Their business cases are larger, their architecture is broader and their risk surface is, um, richer? The portfolio therefore has a natural tendency to reward small productivity applications and the strategic slides are left talking about enterprise reinvention.
The result is that your enterprise is publicly pursuing transformative AI because the CEO wanted to become AI-First, but internally you’re funding a civilization of copilots.
AI transformation should be about changing the shape of work itself and real value lives across functions and between systems and decision boundaries. And those boundaries are exactly where governance multiplies. An example: a local assistant can fit inside one department, and an agentic process that crosses Finance, Operations and IT inherits three organizational histories, four ownership models and enough committees to qualify as a federal system.
The more transformative the idea, the more governance surfaces it touches and the longer the lead time becomes and the business case ages. Then sponsors move on, budgets shift. and the portfolio board then points at the deteriorating business case as evidence that the original initiative lacked sufficient certainty.
This is how governance can manufacture the uncertainty itself, that it was created to reduce.
Few governance bodies (or loss functions as I call them) are measured on the cumulative time they add to enterprise value creation. Security and architecture get blamed for incidents, but not for revenue that arrived nine months late.
A six-month governance cycle push value six months into the future and it also freezes the organization’s understanding for six months while capabilities continue moving and by the time approval arrives, the decision is about a market that no longer exists.
The difference people often describe is partly a product of Europe’s political and social model.
European institutions have spent decades building systems around risk control, and the idea that everyone affected should have a voice. That instinct has obvious democratic value, but it also leaks into corporate decision-making.
Some organizations are still willing to put accountable people in a room, bound the risk, fund a contained move and learn through execution. But most enterprises interpret good governance as making sure every function has been heard and all stakeholders have been consulted so none of them can later claim they were excluded.
A transformation should not need to explain itself independently to procurement and architecture, security and the rest of the boards and multiple steering committees when those bodies are evaluating overlapping dimensions.
Shared evidence should travel with the initiative. I think that risk should determine the depth of review with low-risk experimentation given a fast route and cross-functional decisions should only happen with the relevant authorities in the same room instead of being passed sequentially through the organisation’s intestinal tracts.
That requires giving people authority. And authority comes at a risk that you’re making a decision that could jeopardize your career. In the Netherlands that risk is amplified by a deeply ingrained cultural instinct not to stick your neck out too far.
There is still something distinctly Calvinist about the idea that ambition should remain modest and that visibility should be earned reluctantly and you shouldn’t look too pleased with yourself for taking charge. In that environment, consensus feels safe and another round of consultation is considerably less dangerous than being the person whose name is attached to the decision.
Real delegation creates the possibility that someone lower in the hierarchy makes a consequential decision. And we wouldn’t want this to happen now would we? This is especially dangerous in cultures where seniority and control have grown together for decades. The safer alternative is escalation since you give everyone psychological cover because when a decision goes haywire, a committee approved it.
The price of that comfort is speed.
The board that manages all other boards
At some point in time, the number of boards in a European company become visible even to the people who are creating the boards. The enterprise discovers there are conflicting portfolios and overlapping steering structures and stakeholders continue to attend meetings whose only purpose is to prepare for other meetings.
Then, a transformation council is proposed to align the governance landscape.
Yup. The board that manages boards has arrived. This is, what I call, the larval stage of meta-governance, where representatives of existing governance bodies meet to resolve conflicts and harmonize priorities. A common dashboard is commissioned. and someone creates a RACI that has enough colored cells to trigger a seizure in a program manager.
Mature organizations love their dashboards because they create the sensation that complexity has been domesticated. Initiatives become rows and statuses receive colors. Green means on track, amber requires attention, and when red you will be asked to explain yourself.
The AI transformation remains green for months.
Procurement completed ‘on time’. Green. Architecture review completed. Green. Security assessment progressing. Green. Portfolio approval obtained. Green. Steering decision taken. Green. AI intake accepted. Green. Every governance mechanism has met its internal milestone. The company is now extraordinarily successful at delivering the process around delivery.
There is still no meaningful transformation.
This is the point where I would like to propose the final missing board.
The answer is the Ouija Board.
Once a quarter, the executive committee, procurement, architecture, security, portfolio management, AI governance and the transformation annex value office gather in a darkened conference room. The original business case is placed in the middle of the table. Everyone joins hands and the lights are lowered. A project manager who has aged visibly since vendor selection asks the spirit world to make contact with the ROI.
There may be movement.
The glass slides toward a letter. Then another. Slowly, painfully, the message forms.
The message says that the ROI was here in Q2.
This sounds ridiculous because it is ridiculous, yet it points at the one governance question that is never asked. Every board wants to have their domain satisfied, but very few ask whether the total governance burden remains economically proportional to the value and risk of the initiative.
Governance itself has a cost, people.
That cost can easily be measured in expert hours, but also in calendar delay, duplicated analysis, management attention, lost learning and decayed opportunity. Enterprises are comfortable calculating the cost of software, and licenses, but they’re far less equipped with calculating the cost of making thirty senior people repeatedly discuss the same decision across seven forums.
Governance labor is distributed across salaried people and therefore feels free. Calendar time does not appear as a line item on the project charter and the architect was already employed and the steering committee would have met anyway. And this way, hours disappear into overhead and the transformation carries only the visible external spend.
This accounting illusion makes governance expansion cheap for the people adding governance.
The enterprise can therefore spend half a million euros of internal time protecting a hundred-thousand-euro decision without ever seeing the contradiction on a purchase order or it delays a valuable capability for two quarters and congratulates itself on commercial discipline. AI makes the contradiction harder to ignore because AI is often sold as a mechanism for reducing coordination cost.
We build agents to move information between systems, and then we deploy them through a transformation process made almost entirely of human handovers.
I think that the design goal should be good governance per unit of friction.
That means reducing repeated evidence and separating advisory review from decision authority, matching process depth to actual risk and placing business value inside the governance objective instead the outside. And decision rights matter even more. If a steering committee approves a budget, it should not also need to approve every technical choice inside the approved boundary.
Ever thought about that? Decentralized authority.
If architecture has defined a compliant pattern, teams using that pattern should move. When security has classified a low-risk use case, the lighter route should actually be lighter and when your AI program has created standard controls, business units should gain speed by using them.
Central governance only earns its existence when reuse reduces friction.
The alternative is what I call governance gravity. Where every central capability attracts more intake and more reporting, resulting in even more meetings and more mandatory artefacts until the thing you created to accelerate the enterprise is another one of the planets circling in orbit.
Now, when the transformation finally reaches production after several years, only a few remember the launch, and even less can recall how much value was consumed getting there, and the original ROI is adjusted to reflect current reality
Then another executive sees an AI demo.
The room becomes quietly excited. Someone says this one feels different. The opportunity is enormous. Speed will be critical. Leadership alignment is strong. People leave the meeting energized.
Then procurement opens the RfP template.
And somewhere, in a locked cabinet near the transformation office, the Ouija Board begins to vibrate.
Signing off,
Marco
For my daytime job, I’m a researcher and factory builder at Eigenvector, a commercial research lab operating at the frontier of unsexy AI. In the evenings, I write about why building this stuff is a lot harder than the vendors would have you believe.
This story is published on Generative AI. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories.
Subscribe to our newsletter and YouTube channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!
