Your first project has one job. Making the second project possible.
That changes what you should pick. The best first project is often the second or third best opportunity in your business, chosen because it will finish, work, and be seen.
Four filters
Every candidate needs all four.
1. Small enough to finish in a month
Value inside four weeks, or momentum dies.
Long projects lose their sponsor, lose their team to something urgent, and lose everyone who was curious. By the time a six month build lands, the business has moved on.
Cannot show something working in a month? Cut the scope until you can.
2. Painful enough that someone cares
Find the person who complains about this task.
Their complaint is your best asset. They will test it, tell you the truth about it, and champion it to their peers, because you fixed something that was genuinely annoying them.
A project with no complainant has no advocate. It ends up technically fine and socially invisible.
3. Measurable in a way people believe
You need a before number and an after number. Both should be things people already recognize.
Hours per week. Turnaround time. Error rate. Backlog size.
Avoid anything that needs explaining. "Improved capability" convinces nobody and funds nothing.
4. Visible outside the team
The result has to reach people who were not involved.
This is what turns one project into a program. Other teams see something real, done by colleagues they know, and start asking whether their version is possible.
A win buried in the back office produces value and no momentum.
What to avoid first
Four categories that make excellent projects and terrible first projects.
Anything customer facing. The stakes are wrong for a first attempt. Get your failures out of the way internally.
Anything touching three or more systems. Integration work is where timelines go to die, usually for reasons unrelated to AI.
Anything a senior person is emotionally attached to. You are learning. Learn somewhere the politics are quiet.
A chatbot. Sorry. Most requested first project, one of the hardest to do well. Open ended input, high visibility, easy to embarrass yourself. Come back to it at project four.
Where good first projects usually hide
The same five places, across most businesses.
- The Monday report. Someone rebuilds it every week from the same sources. Hours, every week, forever.
- Inbox triage. Reading what came in, working out what it is, sending it on with context.
- Quote and proposal drafts. The first version takes an hour and is eighty percent the same every time.
- Data entry between systems. Someone reads from one screen and types into another. The most automatable work in existence.
- The onboarding pack. Same documents, same questions, assembled by hand for every new client or staff member.
None of these are exciting. All of them work.
Getting it approved
Two numbers and one sentence.
The numbers are what it costs today and what it will cost to fix. Both fit on one line.
The sentence is what changes for a named person. "Sarah gets four hours back every Monday" lands better than any efficiency framing. It is concrete, and someone can go and ask Sarah about it afterward.
Keep the pitch short. A long deck signals a project that needs defending.
Running it
Six steps, in this order.
- Measure the before. One hour, done properly. Everyone skips this and everyone regrets it.
- Name the owner. Someone whose objectives include this working, with protected time.
- Build small. The narrowest version that produces value. Resist every feature request until version one is in use.
- Put it where the work happens. Same screen, same tool, same rhythm people already have.
- Sit with the users. Watch them use it. Watching beats asking. The gap between what people say and what they do is where the real problems live.
- Measure the after. Same method as the before, so the comparison holds.
Then stop and look
After a month, three questions.
Did it work? Are people using it without being asked? Did anyone outside the team notice?
Three yeses, go again, bigger. Two, fix the missing one before expanding. One or zero, look at why before spending more.
That pause is what separates a program from a pile of pilots.
Why the first one matters
Every business has a memory of change.
Your first AI project works and people can see it? The next one starts with the benefit of the doubt. People turn up curious.
It stalls quietly? The next one starts underwater. Your team has learned that this gets announced and then fades, and they will treat the next announcement accordingly.
You are setting the expectation for everything after it.
Score your own workflows in two minutes.
The AI Opportunity Audit maps your workflows, scores them for AI fit, and hands you a ranked list with effort, cost, and payback. Your first project is usually near the top of it.
Weighing up whether you need dedicated AI leadership before you start? The Leadership Fit check takes about the same time.