The discovery step that finds you already solved this

A director of support at a hardware company came to us wanting an AI assistant. His agents kept getting the same question, over and over, about which replacement part fit which product model. It ate time, it caused errors, and he was ready to fund a project to build a model that could answer it. He had the budget approved. He wanted to start.

We asked one question before touching the model. How do your best agents answer this today. He did not know exactly, so he went and watched them for an afternoon. He came back a little embarrassed. His senior agents were not guessing. They were pulling up an internal compatibility lookup, a simple search tool his own team had built two years earlier, that mapped part numbers to models. It gave the right answer in seconds. The problem was not that the answer did not exist. The problem was that half his agents did not know the tool existed.

He did not need an AI project. He needed to surface a report he already had. That afternoon of watching saved him a build.

The need feels new because you skipped a step

When a problem is painful and repetitive, the instinct is to build something new to kill it. AI makes that instinct stronger, because AI feels like the new thing that finally handles the messy cases. What gets skipped is the least glamorous discovery step there is: finding out how the problem is solved today, by the people who are good at it.

Almost every organization has more solutions than anyone can remember. A report a data analyst built for one request and everyone forgot. A macro that automates a task three people know about. A lookup tool that answers exactly the question you are about to spend six figures re-answering with a model. These things do not show up in a roadmap. They show up when you watch the work.

A head of operations at an insurance firm told me a version of this that stuck with me. Her team wanted AI to summarize policy documents for adjusters. When she looked, she found the summaries already existed. A previous team had generated them and stored them in a folder nobody linked to from the tool the adjusters used. The work was done. The wiring to reach it was missing, and wiring is a lot cheaper than a model.

Discovery is inventory, not just imagination

Good discovery is not only imagining what AI could do. It is taking inventory of what already exists and honestly asking whether the gap is intelligence or access.

A few questions surface the solution you forgot you built:

How do your best people handle this today, watched in person, not described in a meeting.
What reports, tools, or macros exist that touch this problem, even ones nobody uses.
Is the real gap that the answer does not exist, or that people cannot find or reach the answer that does.
If you deleted the AI idea, what would a well-informed employee do instead.

If the answer to the last question is “use the tool we already have,” you have your project, and it is not an AI project. It is a discoverability project, and it will ship faster and break less.

This is not an argument against building AI. The hardware company did eventually build something, later, for a genuinely harder question the lookup tool could not answer. The point is that they built the right thing, because discovery separated the problem they had already solved from the one they had not.

How we approach it at Density Labs

The AI Opportunity Assessment, our fixed two-week engagement at $2,500, starts by mapping how the work is done now before we scope anything new. We watch the process, list the tools already in play, and test whether the gap is real intelligence or just plumbing and awareness. Sometimes the honest finding is that the best move is not to build a model at all, and that finding is worth the engagement on its own. It is a lot cheaper to learn it in two weeks than in two quarters.

Before you fund the model, find out whether your team already answered the question and forgot.