# Strategy and Commercialization Reference

## Strategic stack

Build choices from top to bottom:

1. Market and timing: where change is possible now.
2. ICP and workflow: whose important task to win.
3. Value mechanism: which measurable outcome improves.
4. Winning system: product, AI, data, operations, trust, and distribution.
5. Economics: price, cost, margin, implementation, support, and scale.
6. Organization: capabilities, decision rights, portfolio, and governance.

Strategy requires explicit exclusion. List attractive segments, features, channels, or platform investments that will not be pursued in the current horizon.

## ICP scoring

Assess problem severity/frequency, budget, buying authority, implementation friction, data access, measurable value, risk, time to proof, sales cycle, retention, expansion, and reference potential.

Separate:

- End user who changes behavior.
- Beneficiary who receives the outcome.
- Buyer who controls budget.
- Administrator who deploys and governs.
- Risk owner who accepts consequences.

## Capability-workflow fit

Do not sell a generic assistant. Identify a repeatable workflow where AI changes speed, quality, reach, personalization, or cost. Specify human/AI boundary and how the output changes a downstream business result.

## Moat tests

| Moat hypothesis | Test |
|---|---|
| Workflow embedding | How costly is replacement and retraining? |
| Authorized data loop | Does use create lawful, unique improvement data? |
| Evaluation/error knowledge | Are hard cases and decision thresholds proprietary? |
| Domain operations | Does delivery depend on accumulated expert processes? |
| Trust and distribution | Is adoption cheaper/faster due to reputation or channel? |
| Unit economics | Does scale structurally reduce cost or improve yield? |
| Network/ecosystem | Does each participant increase value for others? |

Ask a capable, funded competitor to copy each moat in 12 months. If the answer is “buy the same model and rebuild the UI,” the moat is weak.

## Value quantification

Estimate annual value from:

- Time saved × loaded labor cost × realizable utilization.
- Errors/rework avoided × expected loss.
- Throughput/capacity added × contribution value.
- Revenue conversion/retention lift × contribution margin.
- Risk reduction × probability-adjusted impact.
- Faster decision/cycle time × business consequence.

Avoid counting all theoretical saved time as cash value. Apply adoption, quality, realization, and attribution factors.

## Full unit economics

Calculate cost per successful task:

```text
(model + embeddings/retrieval + tools + storage/network
 + retries/failed work + human review + support/operations
 + allocated evaluation/security/engineering/vendor costs)
/ number of qualified successful tasks
```

Model conservative/base/upside traffic, input/output length, steps, tool calls, retry, success, human review, price, and vendor change.

## Pricing and packaging

Choose a value metric that customers understand, can budget, cannot easily manipulate, and that scales with value without encouraging harmful behavior.

- Seat: value and usage correlate with active users.
- Usage: resource or task volume varies materially.
- Outcome: result is attributable, auditable, and dispute cost is manageable.
- Hybrid: platform/base value plus variable usage.
- Enterprise contract: predictability, SLA, governance, and implementation matter.

Package by workflow, control, service, integration, capacity, governance, and outcome—not only model access.

## Portfolio management

For each initiative record strategic thesis, user evidence, technical evidence, economic evidence, risk, dependency, next milestone, investment, and kill criteria. Classify as accelerate, maintain, experiment, partner, or stop.

Review sunk-cost bias. Continuing requires future expected value, not past investment.

## Build, Buy, Partner

Build differentiated workflow, proprietary evaluation/data loops, control, and critical orchestration when the organization can sustain them. Buy mature commodity capability for speed. Partner for domain distribution, data, operations, or implementation that is not sensible to own.

Compare multi-year TCO, switching, vendor change, data/control, time, talent, support, and exit.

## Platform threshold

Platformize when several teams repeatedly need stable common capabilities and duplication/risk exceed platform cost. Require a platform product owner, customers, service catalog, SLO, adoption plan, showback/chargeback, governance, and roadmap. Avoid abstracting unstable one-off use cases.