The short answer
Your first AI win should be one real piece of work your team already does often, that someone senior cares about, that you can measure, and that a small team can ship in a month. Spend week one choosing it and defining “done”. Spend weeks two and three building it with the people who will use it. In week four, put it into daily use and measure it against the number you set. One working result in a month teaches your business more than a quarter of experiments.
Why most AI efforts stall before they change anything
Most leadership teams have tried AI. A few people draft emails with it, and there was a promising demo in the spring. Yet the way the business wins and delivers work looks much the same.
That is rarely a technology problem. The effort stalls for ordinary reasons:
- Nobody chose. “Explore AI” is a direction, not a project, so effort spreads thin across ten small trials.
- Nobody owns it. Experiments live with whoever was curious that week. When they get busy, the work stops.
- “Done” was never defined. If you can’t say what finished looks like, you can’t finish.
- It sat beside the work, not inside it. A clever prompt in someone’s personal account is not a process.
- The first target was too big. “Rebuild our whole marketing operation with AI” is a year, not a first win.
The fix is a smaller, sharper first project that actually ships.
How to choose your first AI project: a five-point test
Run every candidate through these five questions. A good first win passes all of them.
- It is real, recurring work. The task happens every week or every month, and today a person does it by hand.
- Someone senior cares how it turns out. If the founder or a department head feels the pain, the project gets attention, decisions get made, and the result gets used.
- You can measure it with one number. Hours per week, days to turn something around, proposals sent, leads followed up. Pick one, and write down today’s figure before you start.
- A small team can ship it in a month. Two to four people, existing tools, no new hires, no procurement cycle. If it needs a vendor contract to begin, it is not your first project.
- A mistake is cheap and easy to catch. Start where a person reviews the output before it reaches a customer. Save the high-stakes, fully automated work for later, once you trust the process.
If two candidates pass, choose the one your team will use most often.
What a 30-day AI project looks like, week by week
Week 1: decide
Choose the project and name one owner. Write a one-page brief: the job being done, who uses the result, the number you will move, today’s baseline, and what “live” means. Collect real examples of the work as it is done today, good and bad. Those examples become the standard the AI version has to meet.
Weeks 2 and 3: build it with the team
Build a first working version in the first few days, then improve it against real cases. The people who do the job today should test it daily. They know the edge cases, and their sign-off is what turns a tool into a habit.
Keep it inside the tools the team already uses, such as the shared drive, the CRM or the inbox. Write the steps down as you go.
Week 4: go live and measure
Switch it on for real work. Run it alongside the old way for a few days if the stakes call for it, then retire the old way. At the end of the week, compare the number against your baseline and record what you learned. That short write-up becomes the case for your second project.
This pace is achievable because AI handles much of the legwork while people keep the direction and the judgment. On a platform we built this way, one person set the direction and approved the work, and 528 reviewed changes shipped in 60 days. Your first project will be far smaller than that. The principle is the same: people decide and review, AI does the volume.
Examples of good first AI wins
Four first wins that tend to pass the test, one from each part of the business:
Strategy: a monthly market and competitor brief. Today someone skims competitor sites, news and customer calls, and a summary appears when they have time. The AI version gathers the same inputs on a schedule, drafts a two-page brief in a fixed format, and a leader edits it in fifteen minutes. The number: hours to produce it, and whether it now arrives every month.
Brand: on-voice first drafts for the content you already publish. Write down your voice, your audience and the topics you own, along with a set of your best past pieces. The AI version turns a short brief from a subject expert into a first draft that sounds like you. An editor still approves every piece. The number: drafts published per month, or the days from idea to post.
Go-to-market: proposals and follow-ups after sales calls. Sales calls end and follow-ups go out late, or not at all. The AI version takes the call notes, drafts the follow-up email and a first-pass proposal from your existing templates, and puts both in front of the account owner the same day. The number: time from call to proposal, and the share of calls that get a follow-up within 24 hours.
Operations: a smoother client onboarding. New clients get the same welcome pack, questionnaire and setup checklist, assembled by hand each time. The AI version produces a tailored pack from the signed agreement and flags anything missing. The number: days from signature to kickoff.
Notice what these share. Each one is recurring, owned, measurable, small, and reviewed by a person before it reaches anyone outside the team.
Common mistakes to avoid with a first AI project
- Starting with the tool. Choose the work first, then the tool that fits it.
- Picking the flashiest idea. A customer-facing chatbot looks impressive in a meeting and carries real risk. An internal process that saves your team eight hours a week is a better first win.
- Skipping the baseline. If you don’t record today’s number before you start, you can’t prove the result, and the second project gets harder to fund.
- Building it off to the side. If the people who do the job aren’t involved in week two, they won’t use it in week four.
- Running five pilots at once. One finished project beats five half-built ones. Finish, measure, then choose the next.
- Letting it go live without an owner. Name the person who keeps it running before launch, not after.
Where to start this week
Write down three candidates and run each through the five-point test. If one passes cleanly, you have your first project. If none do, that tells you something too, and it usually means the right first win sits somewhere you haven’t looked yet.
If you’d like a structured way to find it, Next Moves asks 12 questions about your business and takes about four minutes. It shows where AI could make the biggest difference first, and you see the result straight away.
Frequently asked questions
How do I choose my first AI project?
Pick work your team does every week, that a senior person cares about, that one number can measure, that a small team can ship in a month, and where a person reviews the output before it goes out.
Can an AI project really go live in 30 days?
Yes, if the scope is one job, one owner and one number. A month is enough to choose, build with the people who will use it, and put it into daily work. It is not enough for a company-wide rollout, which is why the first project should be small.
Do we need to hire AI specialists first?
Usually not for a first project. You need an owner, the people who do the work today, and enough AI skill to build the first version. Outside help can speed that up, but the team that will use it should help build it.
How do I measure whether an AI project worked?
Record one number before you start, such as hours per week or days to turn something around, and compare it at the end of week four. Then check whether the team still uses it unprompted. That is the clearest sign it stuck.
What should our second AI project be?
Often the step just before or after the first one in the same process. Run it through the same five-point test and keep the same 30-day shape.