SageCreek AI
Two colleagues working together on laptops and code, starting their first AI project
← ALL BRIEFS

SAGECREEK BRIEF

How to choose your first AI initiative

Most companies pick their first AI project because it's easy to demo, not because it moves revenue. Here's a better filter.

SAGECREEK AI · SEPTEMBER 18, 2026

The easiest AI project to greenlight is usually the wrong one. A chatbot on your website, a summarizer for internal memos, an assistant that drafts emails — these are easy to demo in a board meeting and easy to forget about six months later, because they don't touch the number anyone is actually accountable for.

The better filter is simpler than it sounds: where does this business lose money or leave money on the table because a human has to do something slow, repetitive, and judgment-light? Comp analysis that takes 80 hours instead of one. A research process that runs on founder pattern-matching instead of a repeatable system. A category screen that takes a quarter instead of an afternoon.

Those are the projects worth building first, because the return shows up in hours saved, deals closed faster, or revenue that was sitting on the table the whole time. Impressive demos are optional. Moving the number is not.

Four questions to run every candidate through

You can run this filter yourself on a whiteboard in an hour. List the candidate projects, then ask the same four questions of each one.

  1. 01Which number does it move?Name the metric and the person accountable for it: revenue, margin, hours, or enterprise value. If nobody owns the number, nobody will fight for the project when it gets hard.
  2. 02How often does the work happen?A task that runs every day or on every deal pays back quickly. A task that runs twice a year rarely justifies a custom build, however painful it is.
  3. 03Is the judgment in it repeatable?Look for work where a skilled person follows roughly the same steps every time. If your best analyst can describe how they do it, a system can do the first pass. If every case is a one-off, keep it human for now.
  4. 04Can you get to the data?The inputs have to live somewhere a system can reach: a database, a booking site, documents, a shared drive. Messy is workable. Locked in one person’s head is not.

A project that clears all four is a strong first build. One that clears only the last two is a nice tool that nobody will miss.

What a good first project looks like

Axia Partners, a real estate private equity firm that buys RV parks, is a useful example. A full comp analysis took 80 analyst hours per park. At the firm’s acquisition pace that math broke, and the plan was to hire two analysts just to keep comps ahead of the pipeline.

That workflow clears every question. The number was analyst hours per deal, and the acquisitions lead owned it. It ran on every park. The steps were the same each time. The booking and occupancy data could be pulled with browser automation. We built a comp analysis agent side by side with the acquisitions lead, and a park comp now comes back the same day. The two planned hires came off the roadmap.

Nobody would have picked comp analysis for a board demo. It was the right first project because the return was obvious as soon as it shipped.

Three first projects to avoid

  • The showcase. It demos well and touches no number anyone is accountable for. Six months later it is still a demo.
  • The moonshot. Rebuilding the whole business around AI is sometimes the right destination, and it is almost never the right first step. Start with one workflow and earn the next one.
  • The orphan. A project with no owner on the business side stalls as soon as it needs a decision. We wrote about that failure in why most AI pilots stall before they ship.

Keep the first one small

A first initiative has two jobs. It has to return something measurable, and it has to teach the organization how to do the second one. Both are easier when the scope is one workflow, one owner, and one number.

Small does not mean trivial. Pick a workflow that matters, define done as running on live data every day, and measure the before and after. Then use the result to decide what comes next. If you are still working out where the candidates are, how to find the right AI use case covers the search in more detail.

If you want a second opinion on your shortlist, that is what the first call is for. We start from how the business makes money, name two or three places AI can move revenue or cut cost, and follow up in writing.

QUESTIONS

About this brief.

Ask which number it changes and who is accountable for that number. If you can't name a metric, whether hours saved, deals closed faster, or revenue that was sitting on the table, it's a demo, not a project.

Work where a human does something slow, repetitive, and judgment-light today: a comp analysis that takes 80 hours instead of one, a research process that runs on pattern-matching, a category screen that takes a quarter instead of an afternoon.

Start from how the business makes money, not from the technology. On the free call we do exactly that, name two or three places AI can move revenue or cut cost, and follow up in writing.

START HERE

Want this applied to your business? Start with a conversation.

30 minutes · No obligation · You own everything we build

Book my free 30-minute call

YOUR CALL IS WITH CONNOR OR BOB, THE FOUNDERS.