Forty ideas, none of them first

A steering committee sets aside a day. Sticky notes go up on a wall, one per idea — predictive maintenance, an internal chatbot, automated invoice matching, a drawing search tool, forty of them by lunch. Each one gets scored for impact and feasibility, one to five, and the wall becomes a table, and the table becomes a slide.

Everyone leaves agreeing on the top of the list. Six months later almost none of it has shipped. Finance built its own chatbot because it could not wait for its turn. Engineering is still arguing with operations about whose problem the drawing search tool actually solves. The slide is still correct, in the sense that nothing on it was wrong — it just never told anyone what to do first.

What should annoy you is not that the list was long. Lists are supposed to be long. It is that nobody in that room was ever going to say no to anything, because the exercise never asked them to.

Why we build it this way

A capability list is not a business case. "We should do something with AI" and "predictive maintenance would save us money" both sound like use cases, but only one of them can be sized, owned, and checked against a real number six months later. The workshop format rewards the first kind, because it is easier to write forty ideas on a wall than to argue two of them down to the ground.

The groups that get somewhere start from the other end. They do not ask what AI could do. They ask what is already costing the business money or time, in problems leadership has already named, and then check whether AI happens to be a sane way to fix that particular one. Sometimes it is not, and the honest answer is a shorter list, not a bigger model.

This only works with both halves of the room at once. Investment has to be approved from the top, or nothing gets funded past a pilot. The actual pain has to come from the people doing the work, or the top only ever funds what looks good in a review. Either half working alone produces the wall of sticky notes again — a roadmap nobody owns, or a request nobody will pay for.

How it works, roughly

We run this as its own short engagement, usually before anything else is built. It reads what the business already has, not a hypothetical:

• Workshops with the people who actually do the work in question, not only their managers
• Any existing list of ideas, and why each one is on it
• Whatever cost or delay the business can already point to and name

What comes back is two or three candidate problems, each with a sized business case behind it — what it costs today, roughly what fixing it is worth, and what would have to be true for AI to be the right tool rather than a simpler fix. Alongside it sits a written "do not do this" list: the ideas that scored well on a wall but do not hold up once someone prices them.

When someone on the leadership team disagrees with the ranking, the engagement does not have a vote or a formula that overrides them. They read the sized case, and either the number is wrong and gets corrected, or they choose to fund something else anyway with their eyes open. The sequencing is a recommendation; the decision about what to spend on first stays theirs.

AI Transformation · drag to turn