AI Opportunity Map
Write up one candidate use case properly, or see what a Sprint delivers.
A one-to-two-page structure for deciding whether one piece of work is worth taking to production with AI. It is the deliverable at the end of a Sprint, and it works just as well as a self-assessment. Fill it in for one candidate at a time. Write it for the person who has to decide, in plain words.
How to use it
- Pick one piece of work. Not "customer service": "answering order-status emails".
- Fill in sections 01 and 02 before you think about technology. If you can't, you don't know the work well enough yet. Go and watch it.
- Fill in the rest. Where you are guessing, write "assumed". Where you don't know, write "unknown". Both are fine. Hiding them is not.
- Comparing several candidates? Score each finished Map with the use-case scoring rubric.
01 · The work
What is the job, who does it, how often, and what does "done" look like? One paragraph, no technology words.
| Field | Answer |
|---|---|
| The job | |
| Who does it (roles, how many people) | |
| How often (per day or week) | |
| What "done" looks like |
02 · Where it hurts
Where do people lose time, re-enter data, wait for someone, search across documents, or rely on what one person knows? Be specific.
| Friction | Minutes per instance | Instances per week | Who is affected | Measured, assumed or unknown |
|---|---|---|---|---|
03 · The candidate future workflow
Describe the job again with the proposed change. Say what a person does, what the system does, and where a person checks the system's output before it counts.
04 · Recommendation
Pick one, and say why the alternatives lost.
| Option | Chosen? | Why it won or lost |
|---|---|---|
| AI | ||
| Simpler automation (rules, forms, integrations) | ||
| Process change (who does what, when) | ||
| No change |
05 · What it would need
| Area | What it needs | State today |
|---|---|---|
| Data (what, where, how clean) | ||
| Tools and integrations | ||
| Who maintains it in production |
06 · Controls
Answer each one. "Not applicable" needs a reason.
- Privacy and security. What data does it touch? Who may see it? Where may it be processed?
- Human review. What must a person check before an output counts? Who?
- Failure handling. How is a wrong output caught, logged and escalated? What happens when the system is down?
- Cost. What does one use cost to run, roughly, and what is the monthly limit?
07 · Pilot scope and production exit criteria
| Field | Answer |
|---|---|
| What the pilot must prove | |
| With whom, and for how long | |
| Explicitly out of scope | |
| Exit criteria: once met, it goes live | |
| Stop criteria: if hit, it stops | |
| Who owns it after go-live |
08 · Success measures
The two or three numbers that would change the decision, and the baseline you need before starting.
Measure growth as well as time: quote or response turnaround, requests or orders handled per person, share of enquiries answered the same day.
| Measure | Baseline today | Target | How it is measured |
|---|---|---|---|
09 · Effort, risks and the next decision
| Field | Answer |
|---|---|
| Rough effort (people, weeks) | |
| Biggest risks and how they are handled | |
| The one decision the sponsor makes next | |
| Decision owner and date |
Worked example (fictional)
A fictional 60-person industrial supplies distributor, for illustration only.
- 01 The work. Three customer-service staff answer "where is my order?" emails, about 40 a day. Done means the customer has a correct delivery date.
- 02 Where it hurts. Each answer takes about 6 minutes: look up the order, check the carrier site, write the reply. About 200 a week. Measured over one week.
- 04 Recommendation. Simpler automation first: a nightly carrier-status import and a reply template cut most of the lookup. AI drafting of the unusual cases is a phase two candidate. AI alone lost because the hard part was data access, not writing.
- 07 Exit criteria. Replies correct in 98% of a 100-email test set, median handling under 2 minutes for four weeks, no customer-facing reply sent without a person approving it.
Use and reuse. © 2026 Maple Spark Labs Inc. You may use and adapt this material within your organization. Keep this notice. Please ask before republishing, redistributing or selling it.