# AI Opportunity Map worksheet

Start with a process and a problem. Complete one record per candidate intervention, then compare the records with the process owner. This worksheet is the successor to the BASICS canvas; its download URL remains available for existing links.

## 1. Process and outcome

- Process name:
- Trigger and endpoint:
- Customer or recipient:
- Accountable owner:
- Current performance gap and baseline:
- Target outcome, measure and period:
- Monthly case volume and variation:

## 2. Constraint and evidence

- Where work waits, repeats, fails or needs clarification:
- Likely cause:
- Competing explanation:
- Why changing this constraint would improve the process outcome:

For each material claim, record:

| Claim | State | Source / owner | Date | Sample / scope | Next evidence needed |
| --- | --- | --- | --- | --- | --- |
| | Unknown / assumed / reported / observed / measured | | | | |

Evidence states describe provenance. A measured claim still needs a suitable sample and method; an operator's account can be useful without being independently verified.

## 3. Compare interventions

| Option | Change to work / roles | Mechanism improving the outcome | Evidence for fit | Limitations |
| --- | --- | --- | --- | --- |
| Process repair / ownership | | | | |
| Integration / workflow | | | | |
| Deterministic rules | | | | |
| Predictive model / analytics | | | | |
| Generative AI support | | | | |

- Preferred intervention or combination:
- Why it is preferable to a simpler alternative:
- Why an alternative was rejected:
- Tasks and handoffs that change:
- Human decisions and exception paths that remain:

Choose architecture and sourcing after requirements justify them. Retrieval, agents, fine-tuning, building, buying and partnering are later choices that can overlap.

## 4. Delivery conditions

For each condition, mark known and satisfied, unresolved, or not applicable. State the evidence and who will resolve it.

- Authority to use inputs and take actions:
- Accessible, suitable data and systems:
- Representative evaluation cases and difficult exceptions:
- Acceptable error and critical-failure checks:
- Human approval, escalation and fallback:
- Integration and operational support:
- Responsible owner and adoption path:

High potential value does not cancel a critical unresolved condition.

## 5. Economics and next decision

- Net time saved per adopted case, including review and rework (low / base / high):
- Adoption and benefit ramp:
- Released capacity and how it will be used:
- Cash savings / avoided costs, with a credible realization path:
- Other outcome improvements, with assumptions:
- Implementation, integration, evaluation and training costs:
- Recurring model, maintenance, monitoring and review costs:
- Reference period, currency and cost owner:
- Shared dependencies and benefits already claimed elsewhere:

Use ranges. Released hours are capacity, not automatically cash savings. Avoid counting the same benefit as both cost savings and additional revenue.

## 6. Decision record

- Decision: measure first / pilot / enable first / choose a simpler intervention / defer
- Reason:
- Evidence that could change this decision:
- Smallest useful experiment:
- Improved non-AI baseline for comparison:
- Process outcome, quality and total-work measures:
- Stop condition:
- Scale condition:
- Owner, deadline and next review:

## Research foundations

This worksheet is a practical synthesis, not an empirically validated ranking model. Its examples and decision rules require testing in the organization using it.

- [APQC guidance on identifying and prioritizing AI use cases](https://www.apqc.org/How-Can-Organizations-Identify-And-Prioritize-Valuable-AI-Use-Cases): business problems, process mapping and alternatives.
- [Dell'Acqua et al., Navigating the Jagged Technological Frontier](https://www.hbs.edu/ris/Publication%20Files/dell-acqua-et-al-2026-navigating-the-jagged-technological-frontier_5c589c8c-fbb5-458f-b285-c944746cd717.pdf): test representative work and exceptions; consulting results with GPT-4 do not predict your process performance.
- [NIST AI Risk Management Framework Core](https://airc.nist.gov/airmf-resources/airmf/5-sec-core/): context, oversight, evaluation and alternatives. NIST does not validate this worksheet.