Free playbook
Your first AI project: 10 places to start.
Most AI spending starts with a tool and goes looking for a problem. This playbook works the other way: ten places a strong first project usually hides, a simple way to pick one, and the four questions to answer before anyone builds anything. Plain language, no tool pitches.
What's inside
The ten places, at a glance.
- The report you rebuild every week
- The quote or proposal you redo from scratch
- The questions customers ask over and over
- Moving information from one place to another
- Scheduling and the back-and-forth
- Chasing people
- Summarizing long things
- Turning one thing into many
- Research you do the same way every time
- Onboarding and intake that repeats
Each one comes with what it looks like in a real week, why it makes a strong first project, and the realistic quick win. Then: how to choose one, and the four-question gut check we use before any build.
Before you ask
Straight answers on first projects.
What should my first AI project be?
Your first AI project should be one task that is frequent, repetitive, and built on information you can already reach: a weekly report you rebuild by hand, quotes you redo from scratch, questions you answer over and over, or data you move between systems. Start with the task that costs the most hours and repeats most predictably, not with a tool. A first project chosen this way is small enough to prototype in a couple of weeks and pays for itself in time back.
What are realistic AI quick wins for a small business?
Realistic AI quick wins are drafts and summaries with a person in the loop: a first-draft weekly report with the numbers pulled in, draft replies to common customer questions built from your own past answers, structured summaries of long calls or documents, automated follow-up nudges that stop when someone responds, and intake that lands information in the right system without retyping. Each is a real, common win because the inputs already exist and a human reviews the output before it counts.
How do I run an AI proof of concept?
Run an AI proof of concept as a working prototype on one real task, not a slide deck. First answer four questions: what exactly is not working, where the information lives, what security and privacy boundaries the build must respect, and what the finished thing should hand you. Then build the smallest version that runs on real examples, try it for a couple of weeks with a person reviewing the output, and measure the time it gives back. Greenlight the full build after you have watched it work.
Found your candidate? Let's pressure-test it.
A free, 20-minute readiness call. Bring the task you'd hand off first. If it's not a fit, you still leave with a clearer view.