# AI Product Manager Portfolio Outline and Evidence Template

> This is a blank template, not a case study or hiring standard for any company. It cannot predict hiring outcomes. Do not invent users, research, launch states, business metrics, employment, or individual contribution. Label practice work, synthetic data, and tool-assisted content.

How to use it: copy the file once per project. Complete the evidence matrix before writing the summary. Delete an unsupported conclusion, downgrade it to a hypothesis, or add a validation plan. Never present a placeholder as completed work.

---

# {{Project name}}

**Project state:** {{Personal practice / course / internship / prototype / pilot / live}}

**Dates:** {{Start}} – {{End or present}}

**My role:** {{Accurate role}}

**My contribution:** {{Decisions and artifacts I completed}}

**Collaborators and tools:** {{What people and tools contributed}}

**Disclosure note:** {{Data provenance, authorization, de-identification, and confidential material}}

**Last updated:** {{Date}}

## 0. Truthfulness check

- [ ] Project state is not inflated
- [ ] Every number has a definition, source, version, and time window
- [ ] Practice inputs and synthetic data are labeled
- [ ] Team and tool contributions are not claimed as individual work
- [ ] Unauthorized user, customer, and company materials are absent
- [ ] Plans, forecasts, and automated grades are not presented as real outcomes

## 1. One-page summary

### User and task

{{Who needs to transform which input into what output, and in what context.}}

### Problem and evidence

{{Current workflow, problem evidence, impact, and evidence limits.}}

### Why consider AI

{{Potential value over a person, rule, search, or template; what does not need AI.}}

### My critical judgment

{{Most important trade-off, supporting evidence, rejected option, and unknowns at the time.}}

### Current conclusion

{{Whether the evidence supports continuing, narrowing, falling back, or stopping; name unsupported conclusions.}}

## 2. Current workflow and problem evidence

### Current workflow

1. {{Step and owner}}
2. {{Step and owner}}
3. {{Step and owner}}

### Evidence matrix

| Evidence ID | Source and permission | Observation | Affected user / step | Limitation  | Resulting decision |
| ----------- | --------------------- | ----------- | -------------------- | ----------- | ------------------ |
| {{E-01}}    | {{Fill in}}           | {{Fill in}} | {{Fill in}}          | {{Fill in}} | {{Fill in}}        |
| {{E-02}}    | {{Fill in}}           | {{Fill in}} | {{Fill in}}          | {{Fill in}} | {{Fill in}}        |

### Problem statement

> {{Target user}} needs to {{complete task}} during {{trigger context}}, but {{observable problem}}. Current evidence comes from {{source and scope}} and cannot yet establish {{limitation}}.

## 3. Non-AI baseline and AI necessity

| Option                   | Task quality | User steps  | Latency     | Total cost  | Risk        | Conclusion  |
| ------------------------ | ------------ | ----------- | ----------- | ----------- | ----------- | ----------- |
| Human workflow           | {{Fill in}}  | {{Fill in}} | {{Fill in}} | {{Fill in}} | {{Fill in}} | {{Fill in}} |
| Rule / template / search | {{Fill in}}  | {{Fill in}} | {{Fill in}} | {{Fill in}} | {{Fill in}} | {{Fill in}} |
| AI candidate             | {{Fill in}}  | {{Fill in}} | {{Fill in}} | {{Fill in}} | {{Fill in}} | {{Fill in}} |

Explain: {{Where AI is constrained and why; what evidence would move the task to a non-AI option.}}

## 4. Task boundaries and system flow

### Input, output, and non-goals

- Accepted input: {{Fill in}}
- Expected output: {{Fill in}}
- Unsupported input: {{Fill in}}
- Decisions AI does not make: {{Fill in}}
- Actions requiring human confirmation: {{Fill in}}

### System flow

1. {{Receive and validate input}}
2. {{Retrieval / model / rule / tool step}}
3. {{Quality or permission check}}
4. {{Expose sources and uncertainty}}
5. {{Edit, reject, confirm, escalate, or fall back}}

### Failure states

| State                      | User experience | System behavior | Logging     | Recovery    |
| -------------------------- | --------------- | --------------- | ----------- | ----------- |
| Missing information        | {{Fill in}}     | {{Fill in}}     | {{Fill in}} | {{Fill in}} |
| No reliable evidence       | {{Fill in}}     | {{Fill in}}     | {{Fill in}} | {{Fill in}} |
| Tool or dependency failure | {{Fill in}}     | {{Fill in}}     | {{Fill in}} | {{Fill in}} |
| Insufficient permission    | {{Fill in}}     | {{Fill in}}     | {{Fill in}} | {{Fill in}} |

## 5. Data and evaluation set

### Data note

- Source: {{Public / authorized user data / self-constructed / synthetic}}
- Permission and de-identification: {{Fill in}}
- Version and date: {{Fill in}}
- Coverage: {{Common, boundary, high-risk, and unsupported requests}}
- Not covered: {{Fill in}}

### Evaluation rubric

| Dimension                     | Pass        | Partial     | Fail        | Critical failure |
| ----------------------------- | ----------- | ----------- | ----------- | ---------------- |
| Task completion               | {{Fill in}} | {{Fill in}} | {{Fill in}} | {{Fill in}}      |
| Faithfulness / source support | {{Fill in}} | {{Fill in}} | {{Fill in}} | {{Fill in}}      |
| Boundary handling             | {{Fill in}} | {{Fill in}} | {{Fill in}} | {{Fill in}}      |
| Format and actionability      | {{Fill in}} | {{Fill in}} | {{Fill in}} | {{Fill in}}      |

### Evaluation slices

| Slice                    | Why it matters | Case source | Current finding | Still unvalidated |
| ------------------------ | -------------- | ----------- | --------------- | ----------------- |
| Common input             | {{Fill in}}    | {{Fill in}} | {{Fill in}}     | {{Fill in}}       |
| Long or noisy input      | {{Fill in}}    | {{Fill in}} | {{Fill in}}     | {{Fill in}}       |
| Missing information      | {{Fill in}}    | {{Fill in}} | {{Fill in}}     | {{Fill in}}       |
| High-risk / unauthorized | {{Fill in}}    | {{Fill in}} | {{Fill in}}     | {{Fill in}}       |

## 6. Evaluation findings and failure taxonomy

Do not show only an average. Enter truthful findings. If no evaluation exists, write “not yet evaluated” and retain the plan.

| Version | Baseline or candidate | Data version | Task quality | Critical failure | Latency     | Conclusion  |
| ------- | --------------------- | ------------ | ------------ | ---------------- | ----------- | ----------- |
| {{V0}}  | {{Baseline}}          | {{Fill in}}  | {{Fill in}}  | {{Fill in}}      | {{Fill in}} | {{Fill in}} |
| {{V1}}  | {{Candidate}}         | {{Fill in}}  | {{Fill in}}  | {{Fill in}}      | {{Fill in}} | {{Fill in}} |

| Failure type           | Case ID     | Severity    | Possible cause | Detection   | Product response | Next step   |
| ---------------------- | ----------- | ----------- | -------------- | ----------- | ---------------- | ----------- |
| Task omission          | {{Fill in}} | {{Fill in}} | {{Fill in}}    | {{Fill in}} | {{Fill in}}      | {{Fill in}} |
| Unsupported generation | {{Fill in}} | {{Fill in}} | {{Fill in}}    | {{Fill in}} | {{Fill in}}      | {{Fill in}} |
| Permission or privacy  | {{Fill in}} | {{Fill in}} | {{Fill in}}    | {{Fill in}} | {{Fill in}}      | {{Fill in}} |
| System failure         | {{Fill in}} | {{Fill in}} | {{Fill in}}    | {{Fill in}} | {{Fill in}}      | {{Fill in}} |

## 7. Prototype control and recovery

For each critical screen, attach an image or link and answer:

- How does onboarding explain capability and limits?
- How does the product ask for missing information?
- How are sources, uncertainty, and version shown?
- How can the user edit, reject, retry, or escalate?
- Which high-impact actions require confirmation?
- What happens on timeout, dependency failure, or insufficient permission?
- How is feedback used and retained?

### Prototype validation record

| Task        | Participant / case source | Observation | Design change | Conclusion limit |
| ----------- | ------------------------- | ----------- | ------------- | ---------------- |
| {{Fill in}} | {{Fill in}}               | {{Fill in}} | {{Fill in}}   | {{Fill in}}      |

## 8. Quality, latency, cost, and risk decision table

| Dimension                      | Current evidence | Minimum condition | Candidate option | Trade-off   | Owner / next step |
| ------------------------------ | ---------------- | ----------------- | ---------------- | ----------- | ----------------- |
| Task quality                   | {{Fill in}}      | {{Fill in}}       | {{Fill in}}      | {{Fill in}} | {{Fill in}}       |
| Critical failure               | {{Fill in}}      | {{Fill in}}       | {{Fill in}}      | {{Fill in}} | {{Fill in}}       |
| P50 / P95 latency              | {{Fill in}}      | {{Fill in}}       | {{Fill in}}      | {{Fill in}} | {{Fill in}}       |
| Total cost per successful task | {{Fill in}}      | {{Fill in}}       | {{Fill in}}      | {{Fill in}} | {{Fill in}}       |
| Human review                   | {{Fill in}}      | {{Fill in}}       | {{Fill in}}      | {{Fill in}} | {{Fill in}}       |
| Privacy and permission         | {{Fill in}}      | {{Fill in}}       | {{Fill in}}      | {{Fill in}} | {{Fill in}}       |

**Current decision:** {{Continue / narrow / fall back / stop}}

**Decision evidence:** {{Point to evidence IDs, evaluation versions, and user validation}}

**This decision does not establish:** {{Unsupported conclusions}}

## 9. Release validation and rollback

If the project is not live, label this as a plan.

- Included users: {{Fill in}}
- Excluded users: {{Fill in}}
- Current-workflow baseline: {{Fill in}}
- Primary behavioral metric: {{Definition}}
- Quality and safety guardrails: {{Fill in}}
- Cost and reliability guardrails: {{Fill in}}
- Continue condition: {{Fill in}}
- Adjust condition: {{Fill in}}
- Stop condition: {{Fill in}}
- Rollback method and owner: {{Fill in}}

## 10. Iteration log

| Date     | Evidence / issue | Change      | Expected effect | Observation | Decision    |
| -------- | ---------------- | ----------- | --------------- | ----------- | ----------- |
| {{Date}} | {{Fill in}}      | {{Fill in}} | {{Fill in}}     | {{Fill in}} | {{Fill in}} |
| {{Date}} | {{Fill in}}      | {{Fill in}} | {{Fill in}}     | {{Fill in}} | {{Fill in}} |

## 11. Contribution and tool-use matrix

| Artifact / decision | My contribution | Other contribution | Tool assistance | How I verified it |
| ------------------- | --------------- | ------------------ | --------------- | ----------------- |
| {{Fill in}}         | {{Fill in}}     | {{Fill in}}        | {{Fill in}}     | {{Fill in}}       |
| {{Fill in}}         | {{Fill in}}     | {{Fill in}}        | {{Fill in}}     | {{Fill in}}       |

## 12. Three-minute presentation template

### 0:00–0:30 | Task and user

> {{Who experienced which task problem, in what context, and what the true project state is.}}

### 0:30–1:00 | Evidence and baseline

> {{Problem evidence, non-AI workflow, and evidence limitations.}}

### 1:00–1:40 | Critical trade-off

> {{Why this option, rejected alternatives, and where people retain control.}}

### 1:40–2:20 | Evaluation and failure

> {{Cases and rubric, critical failure found, and detection, recovery, and regression prevention.}}

### 2:20–3:00 | Conclusion and next step

> {{What evidence supports and does not support; continue, narrow, fall back, or stop; next validation.}}

## Final check

- [ ] The first page identifies task, state, contribution, and current conclusion
- [ ] Every material conclusion points to evidence or an explicit hypothesis
- [ ] The AI option is compared with at least one non-AI baseline
- [ ] Evaluation set, rubric, slices, and critical failures are inspectable
- [ ] The prototype includes source, edit, reject, confirmation, and fallback
- [ ] Quality, latency, cost, human effort, and risk inform one decision
- [ ] Practice, synthetic, pre-launch, and confidential materials are labeled
- [ ] The PDF explains the core story without the website, and broken links have a fallback
- [ ] Secrets, internal links, personal identifiers, and unauthorized materials are removed
- [ ] The three-minute presentation matches the portfolio and adds no unsupported result

Template updated: 2026-08-20