“Bring AI into sales” does not define a problem. “We spend two hours reconstructing meeting commitments every week” identifies work to investigate. Observe who does it, how often, which information is used and where errors occur. Inaccessible data or unclear responsibilities may require process improvement before automation.
Separate creation, transformation and decisions. Drafting a first version differs from extracting an exact date or approving a discount. AI can prepare meeting information while a business owner retains the decision.
A selection brief
Describe frequency, current effort, observable quality, data availability and error consequences. These are comparison dimensions, not a universal scoring formula. Use them to explain why a candidate deserves a pilot.
Frequent, reviewable tasks provide learning opportunities. Rare tasks may also be valuable but require different designs. Where errors have serious consequences and independent validation is unavailable, narrow the pilot rather than trying to solve everything.
Measure the whole process
Include preparation, execution, review and correction. Saving ten drafting minutes while adding fifteen review minutes reduces productivity. Released hours do not automatically become financial savings: realized value requires a change in capacity, costs or outcomes.
Create a baseline from real tasks, compare similar work and record failures. Favorable averages can conceal important errors in particular customer categories.
Design a pilot that supports a decision
Specify whether the future decision is continue, adjust or stop. Set a deadline, owner and criteria before results arrive. Keep an alternative operating procedure. Start with reading and drafts if that provides enough evidence to learn.
Worked case
Fictional example: preparing notes takes 40 minutes, reviewing agreements 15 and correcting omissions 5. With AI these become 10, 20 and 10 minutes. Total time falls from 60 to 40: twenty minutes saved per case, not fifty. Twelve weekly cases release four hours of capacity.
Quality still needs verification. An invented commitment can make a fast pilot unacceptable. Time is one measure, not a substitute for evaluating consequences.
Practice instructions
Compare three candidate tasks using frequency, total effort, available data and error consequences.
Identify missing information. Propose a pilot and baseline.
Do not invent savings or treat released capacity as revenue.
Your turn
Design a pilot brief for commercial meeting preparation. Include task, input, output, owner, evaluation sample and two stopping conditions. Separate expected from measured benefits.
Show the commented solution
A defensible pilot uses authorized records from twenty meetings, prepares sourced briefs and requires commercial review. Compare total time and omissions against the current method. Stop for unauthorized disclosure or false commitments. Twenty cases are an initial proposed sample, not a statistical guarantee or an observed benefit.
Evaluate your work
The problem is observable.
The baseline includes review.
Success conditions precede results.
Expected benefits are separate from measured benefits.
Before moving on
Check your exercise against the criteria. Keep sources and your changes: they will support the final project.
Editorial review: September 29, 2026 · All case examples and figures are fictional. Study times are estimates.