Finding the Big AI Idea

The most expensive decision in AI isn't which model to use or which coding tool to adopt. It's what to build.

95% of AI pilots never deliver measurable ROI. Most don't fail in the build — they fail before it, because the team aimed at the wrong thing or stood it up on architecture that couldn't hold it. By the time that shows up, you've spent a quarter and six figures to learn it.

The fix is to move that decision to the front, where it's cheap. Name the single highest-leverage thing AI should build in your product — scored on value and feasibility, with the reasoning a technical leader can pull apart. Not a brainstorm. Not a strategy deck. A worked diagnosis.

Mapping the Territory

Before you score anything, you need a map of where AI actually has leverage in your business. That means going deep on four things:

Your product and where it's headed — the jobs your customers are trying to get done, and where intelligence would change the outcome.

The data you already hold — what's structured, what's messy, what's locked in systems you don't control.

The systems AI would have to live inside — integrations, APIs, auth, deployment constraints.

The candidate ideas that emerge from what's really there — not from a list of AI trends.

This is where the whiteboard matters. Candidate ideas should surface from your business, not from a vendor pitch deck.

Scoring Value vs Feasibility

Every candidate idea gets scored on two axes: the value it creates for the business, and the feasibility of actually building it on your stack.

Value is not "how cool is this." It's whether the idea moves a metric your leadership cares about — revenue, retention, cost, risk, speed.

Feasibility is not "can AI do this in general." It's whether your team can build it on your data, your systems, and your timeline — with the architecture you have today.

The Big AI Idea is the one that wins on both. The runners-up get documented too, with the reasoning, so the choice is defensible when your CEO or board asks why this and not that.

The Architecture Gap Read

Once you have a target, read the architecture standing between you and building it. This is the step most strategy work skips — and the one that saves the most money.

What data-model gaps exist? What integration points are undefined? What standards and patterns are missing? Surface these now, while they cost a conversation instead of a mid-build rewrite.

A strategy firm can name an idea but can't read your architecture. A dev shop will build whatever you hand them, right or wrong. The person doing this work needs to have shipped production software — not just slide decks.

Roadmap Sequencing

The last step is sequencing the build — phases in order, dependencies, what to build first and what each phase needs. Not a wish list. A plan scoped to your team and your stack.

Here's what a complete roadmap artifact set looks like:

You walk in with a sense that AI matters and a list of maybes. You walk out with one named, scored idea, the gaps between you and shipping it, and a roadmap you can act on. And if it turns out AI isn't the right move right now, you'll know that too — before you spend on finding out the hard way.