PROVE: five steps to AI agents that pay off
Most AI projects stall because nobody wrote down what success looks like, or the rules came after the build. PROVE puts both first. It's how every Mindlab engagement runs, and it's free for your team to reuse.
- Pick One process, one owner, one written target.
Choose a task your team repeats every week, name the person who owns it, and write down the result you want, for example “invoice triage from 3 days to 4 hours”.
You end with: A one-line goal with an owner - Record Measure how the work runs today.
Hours spent, wait time, errors and cost, from two weeks of tracking or the data you already have. Without a baseline, nobody can say later whether the AI helped.
You end with: A baseline card - Own the guardrails Set the rules before anything is built.
Who owns the agent, what it may read, what it may do, who approves its actions, and a monthly spend cap. The same rules apply whether it runs on Copilot, Claude or ChatGPT.
You end with: An Agent Fact Sheet - Validate Build small, then test with real questions.
Build the smallest agent that does the job on the right platform, test it with at least 20 real questions or cases, fix what it gets wrong, and only then release it.
You end with: A test set with a pass rate - Evidence Compare with the baseline 30 days after go-live.
Measure the same things as in step R. Then decide: scale it, change it or switch it off. Start the next process with what you learned.
You end with: A results report and a decision
Four rules behind it
- Measure before you build.
- Guardrails before agents.
- The same rules for every AI tool.
- Switch off what doesn’t pay.
Works with any AI tool
PROVE doesn't care whether the agent runs in Microsoft Copilot Studio, Claude or ChatGPT. Steps P, R and E measure the work. Step O sets the rules your risk team needs. Only step V depends on the platform, and that's where we help you pick the right one for the job.
Use it on your own. If you'd like a second pair of eyes, the free AI readiness review walks through steps P to O for your first process.
How PROVE maps to our services
Free workshop
All five steps in miniature. You leave with a working agent and its Fact Sheet.
Governed Agent Workshop
Steps P to V with your whole team on a real process. $4,000 CAD per team.
Pilot to Production
All five steps on one process, in production in six weeks, with the 30-day results report. From $27,500 CAD.
Want help with your first PROVE cycle?
Start with a free 30-minute call. You talk to the architect who would do the work.