Most small businesses that try AI don’t fail because the tool is bad. They fail because the test never ends. Someone signs up for a tool, a few people try it, nobody agrees on what “working” means, and six months later it’s still “something we’re looking at.” The most reliable path to AI implementation for small business teams is a well-run pilot project: one task, a small group, a clear finish line and an honest decision at the end.
That idea sits at the center of The AI Strategy Blueprint by John Byron Hanby IV, which warns about “pilot purgatory”: pilots that run indefinitely, use up time and money, and create the feeling of progress without delivering any. The book is written for large enterprises, but the lesson matters even more for a 20-person company, where nobody has spare hours to lose. Here is a practical way to run a pilot that ends with a decision instead of a shrug.
Why AI pilots stall
The same few problems show up again and again:
- The scope is too big. “Use AI across the company” is not a pilot. It’s a wish.
- Nobody defined success. If you can’t say what “better” looks like, you can’t tell whether you got there.
- There’s no owner. When everyone is responsible for the test, no one is.
- The data isn’t ready. AI working from outdated, scattered or overshared files gives unreliable answers, and people quickly stop trusting it.
- There’s no end date. Without a deadline, the pilot drifts.
Each of these is easy to prevent if you plan for it before you start.
Step 1: Pick one task, not a department
Choose a single task that comes up often, follows a similar pattern each time and is easy for a person to check. Proposal first drafts, meeting summaries, answers to common customer questions and searching long policy documents are all good candidates. If you need help narrowing the list, our guide to what AI can automate in a small business walks through how to choose.
Just as important is what to leave out of a first pilot:
- High-stakes decisions, such as hiring, credit, pricing or anything a regulator cares about.
- Tasks that need many systems connected before they work. Every connection adds time and another place for things to break.
- Tasks that depend on messy data you haven’t cleaned up yet.
- Tasks where nobody can agree on what a good result looks like.
Step 2: Write a one-page pilot plan
A short written plan forces the right conversations before money and time are spent. Keep it to one page and answer these questions:
- What problem are we solving? For example: “Proposal first drafts take about three hours each.”
- Who owns the pilot? One named person who makes the final call.
- Who will use it? A small group of three to five people who actually do the task.
- Which tool and which data? An approved tool and a limited set of example documents, not your whole file server.
- What does success look like? For example: “Drafts take under an hour and need no more editing than today.”
- Where are we starting from? Today’s time, error rate or turnaround for the task.
- When does it end? Four to six weeks is a good target for most small business pilots.
- What could go wrong? For example, confidential details being pasted into the wrong tool, and how you’ll prevent it.
Step 3: Measure where you’re starting, quickly
You need a starting point to compare against, but you don’t need a time-and-motion study. The book makes this point too: detailed measurement projects often stall the work they are meant to support. Time the task on a few real examples, note how many corrections or rework cycles it usually takes, and move on. Choosing a task where the waste is already obvious to everyone makes this step easy.
Step 4: Start with a person checking everything
The book describes a “crawl, walk, run” approach, and it translates well to a small business:
- Crawl: AI produces drafts, and a person reviews every one before it is used. You learn what the tool does well and where it makes mistakes.
- Walk: Once results are consistently good, move to spot checks, with a clear rule for when something must go to a person.
- Run: The tool handles routine work, people handle exceptions, and someone still reviews results on a regular schedule.
Most pilots should stay in the crawl stage for the full four to six weeks. That’s not a weakness. It’s how you build trust in the results and catch problems before a client does.
Step 5: Make an honest decision at the end
When the end date arrives, compare the results with your starting point and choose one of four outcomes:
- Expand it. It met the success measure. Roll it out to more people or similar tasks.
- Improve it. It’s promising but not there yet. Adjust the instructions, examples or data and run a short second round.
- Change the approach. The task is worth solving, but this tool or method isn’t the right fit.
- Stop. The value isn’t there. End it, write down what you learned and pick the next task.
Stopping is a perfectly good result. A pilot that ends cleanly and teaches you something is far more useful than one that quietly runs forever. One simple rule helps: finish or stop the current pilot before starting another.
What a small business pilot might look like
Here is an example of how this could work for a 20-person professional services firm. The owner wants faster proposals. The office manager owns the pilot. Three people who write proposals use an approved AI assistant through their company accounts, working only from five past proposals and the current price list. Today a first draft takes about three hours. The goal is under one hour with no drop in quality, and every draft is reviewed before it goes out. After five weeks, the team compares results and decides whether to expand, improve, change or stop.
Nothing about that plan is complicated, and that’s the point.
Don’t skip the IT and security basics
An AI tool works on top of your accounts and files, so it inherits whatever weaknesses they already have. Before the pilot starts:
- Use an approved tool through business accounts, not personal or free sign-ups.
- Make sure multi-factor authentication is on for email and Microsoft 365 or Google Workspace.
- Check who can see what. AI assistants that search your files can surface anything a user already has access to, including folders that were shared too widely years ago. Our AI data readiness checklist walks through the cleanup.
- Tell the pilot group what must never go into an AI tool, such as passwords, client records under a confidentiality agreement, and health or financial details.
If you don’t have written rules yet, our AI Use Policy page can help you put simple, clear rules in place for your team.
Frequently asked questions
How long should an AI pilot project last?
Four to six weeks works for most small business pilots. That is long enough to see a pattern across real work and short enough to keep attention on it. Set the end date before you start and hold to it.
How many people should be in the pilot?
Start with three to five people who actually do the task every week. A small group gives you real feedback without creating confusion across the company. Expand only after the results justify it.
Do we need special software or new hardware?
Usually not. Many first pilots use a business version of an AI assistant, or AI features in tools you may already pay for, such as Microsoft 365. What matters more is that the tool is approved, set up with the right access and used through company accounts.
What if the pilot doesn’t work?
Then you stop it, write down what you learned and pick the next task. A pilot is a small, time-limited test so that a miss costs very little. Often the lesson is about the data or the task choice, not AI itself.
Can we use client information in a pilot?
Only in a tool your company has approved for that kind of information, and only what the task actually needs. When in doubt, test with sample or redacted documents first. If you handle regulated data, the rules that already apply to it still apply.
Is an AI pilot the same as a proof of concept?
They are close. A proof of concept usually asks “can this work at all?”, often with sample data. A pilot asks “does this work for our people, on our real tasks, well enough to keep?” For most small businesses, a short pilot with real users answers both questions at once.
Who should own the pilot?
One named person who understands the task and has the authority to make the final decision. That might be an office manager, a department lead or the owner. Shared ownership usually means nobody makes the call.
Get help running your first AI pilot
SupportMax helps small and midsize businesses pick the right first task, check that accounts and data are ready, set up approved tools with the right access, and train the people involved. Not sure where you stand today? Start with our free AI Readiness Assessment, learn more about our AI services and AI automation for business, or schedule a consultation. Prefer to talk? Call 310.331.8119.
