We almost made search worse by making it smarter
What expert users taught us about AI-first design, and when typing beats talking
A contract you already know now opens in about 3 to 4 seconds, down from about 10 in our first AI-first design. Four of our top five early adopter clients asked for direct search back, and the shipped design gives them that while AI keeps the spotlight.
A quick lookup turns into a conversation
Legal asks a contract manager for signed agreement C006336. In Classic Search, she would type the ID and open it in about 5 seconds.
Our redesign put Ask Sirion first and moved search into second place, to position Sirion around AI-first search. Now she had to start a conversation, wait for the AI, read its answer, and find the link. The same lookup took about 10 seconds.
Enterprise CLM users are expert users. Sometimes they have a real question, such as which supplier contracts renew this quarter. Just as often, they want to open C006336, and the new design made that harder.
Three questions, and a trade-off under deadline
I asked the product stakeholders three questions.
Why not take a contract ID typed into Ask AI straight to the contract, as Classic Search did? What happens to users who open the app expecting Classic Search? And how is the business positioning AI search, for every client or as a paid feature?
The pull towards AI was stronger, since every product in our market was repositioning around it. Classic Search stayed in the side panel below recent conversations. A senior stakeholder preferred that option, and a tight release window pushed us to ship.
Search was technically still there, yet the task people did every day had become the hardest thing to find. I kept raising it in every review with Akash.
Four of five clients asked for it back
We opened the redesign to our top five clients first. Four came back with the same feedback: finding a contract by its ID was much easier in Classic Search, and we should keep it.
Where should search live?
Akash and I explored three placements, each a different bet on where people would look for search.
The side panel was easy to miss, and a link below the AI input crowded the conversation. The global header is where expert users already look for a known item, one click away on every screen. We chose it, and Ask Sirion kept centre stage.
AI should lead the experience, without becoming the only way to get work done.
Using the first win to ask for more
The header fix showed results, and we shared them with senior leadership. That gave me the credibility to raise the gaps AI left open. Many lookups only need typing and clicking, so improving Classic Search would help users without asking them to learn something new.
Get Shivin Mittal’s stories in your inbox
Join Medium for free to get updates from this writer.
I backed it with a cost case. Every contract-ID lookup sent through Ask AI spends tokens on an answer the user already had, and direct search keeps that spend for questions where AI adds value. The better experience was also the cheaper one, and leadership agreed.
One search for both intents
A keyboard shortcut or the header now opens search on any screen, with recent searches ready. Matches highlight as you type, and an exact contract ID goes straight to the contract. Beside the result, search suggests a question for Ask AI, such as “What are the key risks and obligations in C006336?”
Search became AI’s on-ramp. People meet AI inside a task they already understand, and a known contract opens in about 3 to 4 seconds, faster than Classic Search.
Engineering and business backed both searches, and the experience is now live in production.
What I’d tell another design leader
Our first plan was a reasonable bet, and it gave the team a direction. What changed it was four clients telling us the simplest task in the product had become harder. Before your next AI launch, I would put three questions to your team:
- Are we solving the user’s problem, or showing off the technology?
- What happens when the user already knows what they want?
- Are we creating adoption through usefulness, or through restriction?
Let AI lead, and keep the fastest path open for people who already know what they want.
