Running the selection · Chapter 13
How to scope an AI automation pilot
A pilot is the safest way to start. It tests the work and the agency at once. This chapter shows how to scope one.
“The way to get started is to quit talking and begin doing.”
What is an AI automation pilot?
A pilot runs on your real data, not a demo. It has a metric, a target and an end date. It keeps risk low for both sides.
Why start with a pilot
A pilot answers the questions a pitch cannot. It gives you proof, not promises.
- It proves value on real data.
- It tests how the agency works.
- It limits your risk and spend.
- It builds a case for a full rollout.
How to scope a pilot
Keep the scope to one page. Agree these six things before you start.
Pilot brief
Workflow
Invoice data entry
Success metric
Hours saved per week
Baseline
10 hrs / week today
Target
6+ hrs saved
Timebox
3 weeks
Budget
One fixed pilot fee
Pick one workflow with a clear, countable result. Set a measurable metric and a target. Then timebox it and fix the fee.
Set the success metric
A pilot without a metric proves nothing. Choose one number that matters. Record the baseline before any work starts. Set a target you would call a win. Measure the same way at the end.
Pilot vs proof of concept
The two get confused, but they differ. Here is the distinction.
| Proof of concept | Pilot | |
|---|---|---|
| Tests | Is it possible? | Does it add value? |
| Data | Sample or demo | Your real data |
| Output | Often throwaway | A working build |
| Best for | Technical risk | Business value |
What a good pilot looks like
The go or no-go decision
A pilot ends in a clear choice. Decide it against the target, not a feeling.
- Hit the target? Plan the full build.
- Close but short? Extend once, then decide.
- Missed badly? Stop and rethink the workflow.
Before you scale, agree ownership, pricing and support. Then roll out one workflow at a time.
Pilots by company size
For small teams
One workflow and a small fixed fee is ideal. Keep the timebox tight and the metric simple.
For enterprises
Run the pilot in one team before a wider rollout. Involve security and IT in the scope early.
Common pilot mistakes
Key takeaways
- Pilot one real workflow before a full build.
- Set a baseline, a target and a timebox.
- Pay a small fixed fee to keep focus.
- End with a clean go or no-go call.
Ready to run a pilot?
Find agencies open to a small, fixed-scope pilot.
Browse the directoryFrequently asked questions
What is an AI automation pilot?+
A pilot is a small, paid build on one real workflow. It proves value before a bigger commitment. You set a metric, a target and a timebox. At the end, you decide to scale or stop. It lowers risk for you and the agency.
How long should an AI automation pilot take?+
Most pilots run two to four weeks. That is long enough to build and test one workflow. It is short enough to keep risk low. Set a firm end date up front. If it drags on, the scope was too big.
What is the difference between a pilot and a proof of concept?+
A proof of concept tests whether something is technically possible. A pilot tests whether it delivers real value in use. A proof of concept can be throwaway. A pilot should run on your real data. For most buyers, a pilot is the better first step.
Should I pay for an AI automation pilot?+
Usually yes. A paid pilot gets you real work and real focus. Free pilots often get rushed or deprioritised. A small fixed fee keeps both sides serious. Treat it as a low-risk test, not a discount.
How do I measure a pilot's success?+
Pick one metric that matters, like hours saved. Record the baseline before you start. Set a clear target to hit. Measure the same way at the end. If it hits the target, scale it; if not, stop.
What happens after a successful pilot?+
You make a go or no-go decision. If it hit the target, plan the full build. Agree ownership, pricing and support before you scale. Then roll it out one workflow at a time. A pilot should lead into a clear next step.