# The Opportunity Funnel

The canonical process flow. Run every stage, in order. The funnel narrows: many opportunities enter, a prioritized roadmap exits.

```
1. Brainstorm AI opportunities (manually, with the executive)
        |  rule: exclude RPA / deterministic-code-only candidates
        |  rule: revisit the icebox any time we brainstorm
2. ADD: generated list of additional AI opportunities
        |  (you generate, grounded in their company profile)
3. ADD: newly viable AI opportunities
        |  (economics of paying a human vs. an AI agent)
4. Evaluate EVERY opportunity on the four key metrics
        |  Risk · Business Impact · Feasibility · Human vs. AI
5. High-priority cut
        |  top of the list -> high priority
        |  everything else -> the ICEBOX (never deleted)
6. Deep Economic Assessment (high-priority only, 13-criterion rubric)
        |
7. Roadmap: highest-priority opportunities that PASS deep assessment
```

## Stage rules (from the process slide, verbatim intent)

1. **Brainstorm (interview).** Consider existing processes, workflows, interactions, roles. The executive's own list comes first; their knowledge of the business is the input you cannot generate.
2. **The RPA exclusion.** Don't include opportunities that are easily addressed using RPA and/or deterministic code exclusively. If every run of the task takes identical steps with no judgment, it's automation, not AI, and it should be cheaper automation. Teach this distinction the moment the first RPA-shaped idea appears; it recalibrates everything the executive suggests afterward.
3. **The icebox revisit rule.** Revisit the icebox any time we are brainstorming AI opportunities. On a repeat audit, load the previous matrix's icebox and re-evaluate before generating anything new. What was infeasible or uneconomical last cycle may be viable now.
4. **Generated additions.** Go beyond what comes to mind. Generate opportunities the executive did not name, grounded in their industry, functions, and stated pain. Every generated item carries a rationale; unexplained generation teaches nothing.
5. **Newly viable opportunities.** Consider opportunities that are now economically viable given the expense of paying a human vs. an AI agent to execute a task or task flow. These are categorically different from the enriched list: not "AI does an existing task better" but "this was never worth doing at human prices." Teach the arithmetic in the open: task volume x human cost per task vs. agent cost per task.
6. **Four-metric evaluation.** Evaluate every opportunity on the four key metrics that determine their priority (see evaluation-criteria.md). Every opportunity means every: brainstormed, generated, and newly viable alike.
7. **The cut.** Those opportunities that do not pop to the top of the list when rated by the four key metrics move to the icebox.
8. **Deep economic assessment.** High-priority opportunities only (see deep-assessment-rubric.md).
9. **Beyond the sandbox.** Implement the highest-priority opportunities that pass deep economic assessment, and use those as the roadmap moving forward. Every high-priority opportunity MUST make it beyond the sandbox. A pilot that never graduates is a decision that was never made.

## The repeat cadence

Repeat this process regularly, given how quickly the technology is changing, and to evaluate new opportunities as those that were high priority before are implemented. Quarterly is the default recommendation; every 6 months is the floor. The matrix deliverable is designed to be re-opened, re-sorted, and re-run: the icebox section is the starting backlog for the next cycle.