Most small firms try AI the same way: someone signs up for a tool, uses it for a few weeks, and the firm is left with a general sense that it might be helping. That is not a pilot. A pilot ends with a decision you can defend.
1. Pick one workflow
Choose work that happens often, involves a lot of documents, and can be measured. Proposal writing, permit applications, RFIs and takeoffs are good candidates. "Use AI more" is not.
2. Measure the baseline
How many hours does the task take today? How many rounds of revision? How many per month? Rough numbers from timesheets or a two-week log are enough. Without a baseline, you will be judging on feel.
3. Check your data
List the documents the tool will see. Check your client contracts for confidentiality clauses. Confirm where the tool stores data and whether it trains on it. Canadian firms should check their obligations under PIPEDA and provincial privacy law, and public-sector clients may have stricter rules.
4. Agree on the number that counts
Before you start, write down what success looks like, for example "proposal first drafts in under three hours, with no drop in win rate". Also write down what would make you stop.
5. Run it side by side
For two to four weeks, run the new way next to the old way on real work. Have the people who do the job use it, and give them time to do so.
6. Decide
Compare against the baseline. Go, adjust, or stop. All three are good outcomes if they are based on numbers.
Mistakes we see most
- Buying a platform before choosing a use case.
- Giving the pilot to your busiest person with no time set aside.
- Measuring "it feels faster" instead of hours.
- Skipping the data check and finding a confidentiality problem later.
Our Pilot Sprint follows exactly this process over six weeks.