Key Concepts
Frequency: How often the work actually happens today. Listen for a number, not a feeling like “this will solve a major pain point.” Frequency multiplies the return on an agent.
Impact: How much pain the work removes and how often that pain is felt. A mildly annoying daily task is high impact because the annoyance keeps arriving. Impact also raises the cost of mistakes, so high-impact work that is hard to undo needs a human in the loop.
Quick Reference
Step 1: Ask how often
- Daily or continuous: start here. Every day the team does it is a chance to hand it off.
- Occasionally: not a schedule. Deprioritize it.
Step 2: Weigh frequency against effort
- High frequency, medium effort: ship first. It is the fastest route to something working that the team will actually use.
- Low frequency, high effort: take a closer look. High effort can be worth it even when the work happens only twice a year.
- Low frequency, minimal impact: do not build it.
- Effort is the part people underestimate. Every good agent needs time to scope, build, and test.
Step 3: Check impact
- A quarterly task everyone dreads that several teams depend on can be worth more than a daily task nobody minds.
- High impact and hard to undo: keep a human in the loop.
Quick decision guide
- Build first: frequent, medium effort
- Look closer: rare but high effort, or rare but widely dreaded
- Keep human review: high impact, hard to undo
- Say no: rare and low impact
Common failure to avoid
- Spending two days scoping, building, and testing an agent for a pain point nobody does often enough to change how they work. The demo works, the team finds it interesting, and it never gets used.
Benchmark: Arizona College of Nursing
- Situation: 24 markets, 300+ market-specific prompts, weekly reporting. Weekly across 24 markets is a high-frequency opportunity.
- Approach: built one agent and deployed it in each market, generating the same report on demand 24 times over. It also supported campus-specific content work that used to take hours of research across several tools.
- Results: 51% increase in AI-referred enrollment inquiries in three months, and 20 hours of work back each month
Notes
- If nothing on your list scores well, the framework is working. Your job is not to automate everything people ask for. It is to find the few things worth your time and say no to the rest.
- Ask how often before you ask anything else.