# AI Product Adoption Deck

> Use when the user is designing, debugging, or shipping an AI product and asks about adoption, retention, trust, onboarding, prompting UX, explainability, agent permissions, evaluation, privacy/data UX, or team rollout. Also trigger when the user shares a user quote, support transcript, funnel or metric drop, design review, roadmap question, or PRD draft for an AI feature. Maps symptoms to diagnostics, recommends action cards from the AI Product Adoption Deck, and structures workshops.

- Skill: `aiproductcards/ai-product-adoption-deck` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add aiproductcards/ai-product-adoption-deck`
- Raw SKILL.md: https://api.skillmd.com/api/skills/aiproductcards/ai-product-adoption-deck/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Product & Planning
- Author: AIProductCards (https://skillmd.com/u/aiproductcards)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/aiproductcards/ai-product-adoption-deck

---


# AI Product Adoption Deck — Skill File

The **AI Product Adoption Deck** is a 124-page working playbook of **12 diagnostics, 80 action cards across 10 stacks, and 12 workshops** — purpose-built for the moments where AI product adoption breaks.

**This file is the index. The PDF is the source of truth.** Every card's full body — *what it is, why it works, when to use it, how to apply it, what to avoid, example, metrics, pair-with* — lives in the PDF. Use this file to **locate** cards by symptom, name, or number; use the PDF to **read** them.

---

## When to use this skill

Trigger when any of the following appears in the conversation:

- A **user quote** that sounds like an adoption symptom ("I didn't know what to ask," "I used it once and never came back," "I have to rewrite everything").
- A **metric pattern** — first-task funnel collapse, low export/approve rate, high prompt-rewrite rate, day-7/14 retention drop, low connector grant rate.
- A **product question** about AI onboarding, trust, explainability, correction, agent autonomy, evaluation, privacy UX, or team rollout.
- A **design artifact** — PRD, mockup, transcript, postmortem, roadmap item — for an AI feature.
- A **direct ask** like "diagnose this," "what should we add," "which cards apply."

If the user is asking a generic AI question with no product framing (e.g., "explain RAG"), do **not** force this skill. Cite cards only when the symptom is product-shaped.

---

## Core workflow

Every recommendation follows this shape:

```
Symptom  →  Diagnostic (D01–D12)  →  Action Cards (1–80)  →  Workshop (W01–W12)
```

1. **Diagnose first.** Match the symptom to a diagnostic. Skipping this leads to the wrong cards.
2. **Apply 2–4 cards.** Each diagnostic in the PDF lists its *Recommended tactical cards*. Pick the smallest set that addresses the symptom — don't dump the whole stack.
3. **Workshop if shipping.** If the user is moving toward implementation, name the matching workshop (each ships with a fillable deliverable).

When more than one diagnostic plausibly applies, see **Multi-diagnostic sequencing** and **Confidence rule** below.

---

## Diagnostic index (12 cards)

| Code | Name | Triggering symptom |
|---|---|---|
| `D01` | First-Use Dropoff | Signup → tried AI once → never came back |
| `D02` | Empty Prompt Paralysis | Users stare at the prompt box, type "help me," or leave |
| `D03` | Trust Gap | "Impressive" but they don't act on the output |
| `D04` | Output Not Usable | Users heavily edit, reformat, or discard outputs |
| `D05` | Context Failure | Plausible but wrong about user-specific facts |
| `D06` | Overreliance Risk | Users accept incorrect outputs in high-stakes flows |
| `D07` | Automation Anxiety | Suggestions OK, agents/connectors refused |
| `D08` | Correction Loop Breakdown | Same correction needed every week |
| `D09` | Privacy Hesitation | Drop-off at upload/connector/permission steps |
| `D10` | Retention Without Habit | First task succeeds, week-2 retention collapses |
| `D11` | Quality Ops Gap | Team can't explain why quality improves or regresses |
| `D12` | Team Adoption Friction | A few enthusiasts use it, team never standardizes |

---

## Action card index (80 cards, 10 stacks)

### Stack 01 — Product Fit & Value (cards 01–09)
01 AI-Native Job to Be Done · 02 Outcome Receipt · 03 Capability Contract · 04 Human-Plus-AI Role Split · 05 Trustworthy Defaults · 06 Narrow Wedge · 07 Latency–Value Exchange · 08 Value Before Personalization · 09 Cost and Quota Transparency

### Stack 02 — Mental Models & Onboarding (cards 10–17)
10 Model of the Model · 11 First-Run Expectation Ladder · 12 Example Gallery · 13 Good/Better/Best Ramp · 14 Domain Language Bridge · 15 Context Window Map · 16 Change-Over-Time Notice · 17 Non-AI Escape Route

### Stack 03 — Prompting & Input Design (cards 18–25)
18 Intent Picker · 19 Context Slots · 20 Clarifying Question Gate · 21 Prompt Scaffold · 22 Input Quality Meter · 23 Output Shape Contract · 24 Prompt-to-Workflow Upgrade · 25 Multimodal Input Surface

### Stack 04 — Explainability & Confidence (cards 26–33)
26 Confidence Label · 27 Why This Answer · 28 Source Coverage Map · 29 Evidence Trail · 30 Decision Boundary · 31 Uncertainty Callout · 32 Counterexample Prompt · 33 Comparison Baseline

### Stack 05 — Control, Correction & Feedback (cards 34–41)
34 Inline Correction · 35 Critique Mode · 36 Preference Ledger · 37 Assumption Editor · 38 One-Click Transform · 39 Feedback With Consequence · 40 Editable Memory · 41 Shared Correction Loop

### Stack 06 — Failure Recovery & Safety (cards 42–50)
42 Failure Taxonomy · 43 Fallback Path · 44 Graceful Timeout · 45 Citation Missing State · 46 Human Handoff · 47 Risk-Tiered Warning · 48 Abuse-Resistant Empty State · 49 Prompt Injection Tripwire · 50 RAG Failure State

### Stack 07 — Agent Permissions & Oversight (cards 51–58)
51 Permission Staircase · 52 Reversible Autonomy · 53 Review Before Write · 54 Tool-Use Transparency · 55 Agent Scope Card · 56 Stop Button · 57 Escalation Threshold · 58 Agent Ambiguity Pause

### Stack 08 — Evaluation & Quality Loops (cards 59–66)
59 Golden Task Set · 60 Structured Output Eval · 61 Quality Rubric · 62 Regression Watchlist · 63 Trace Sampling · 64 Human Review Queue · 65 Drift Dashboard · 66 Nondeterministic A/B Test

### Stack 09 — Privacy, Data & Governance (cards 67–73)
67 Data Use Plain Language · 68 Sensitive Data Detector · 69 Minimum Data Start · 70 Data Boundary Label · 71 Audit Trail · 72 Team Prompt Policy · 73 Delete and Forget

### Stack 10 — Retention & Workflow Habit (cards 74–80)
74 Saved Work Surface · 75 Return Hook · 76 Workflow Trigger · 77 Progress Memory · 78 Manual Override Habit · 79 Weekly Value Review · 80 Adoption Cohort Map

---

## Workshop index (12 workshops)

| Code | Workshop | Problem class |
|---|---|---|
| `W01` | AI Adoption Triage | When you don't know which problem you have |
| `W02` | First Session Redesign | First-use / activation |
| `W03` | Trust Calibration Review | Trust / accuracy |
| `W04` | Prompt Scaffold Builder | Prompting / input |
| `W05` | Context Map Session | Context / retrieval |
| `W06` | Output Quality Rubric | Output quality |
| `W07` | Agent Permission Design | Agents / automation |
| `W08` | Failure Recovery Lab | Failure handling |
| `W09` | AI Eval Kickoff | Evaluation |
| `W10` | Privacy & Trust Review | Privacy / data |
| `W11` | Retention Loop Builder | Retention / habit |
| `W12` | Team Rollout Playbook | Team rollout |

Every workshop ships with a fillable deliverable template inside the PDF.

---

## Triage — symptom → diagnostic

### By what users say
| Quote | Diagnostic |
|---|---|
| "I didn't know what to ask" | `D02` |
| "Impressive but I wouldn't send this without checking" | `D03` |
| "I have to rewrite everything it gives me" | `D04` |
| "You missed the obvious thing" | `D05` |
| "It approved it — I assumed it was right" | `D06` |
| "I'm not comfortable letting it act on its own" | `D07` |
| "I corrected this last week and it did it again" | `D08` |
| "I don't know what happens to my data" | `D09` |
| "I used it once, it was cool, I just haven't gone back" | `D10` |
| "Our team can't agree on what good looks like" | `D11` / `W06` |
| "Our admin won't approve it for the team" | `D12` |

### By what users do
| Behaviour | Diagnostic |
|---|---|
| Signup → one AI action → no return | `D01` |
| Stare at prompt box, type "help me," or leave | `D02` |
| Generate output, then don't export/save/approve | `D03` |
| Heavily edit, reformat, or discard outputs | `D04` |
| Plausible but wrong on user-specific facts | `D05` |
| Accept incorrect outputs in high-stakes work | `D06` |
| Refuse agents, connectors, or write-actions | `D07` |
| Re-correct the same mistake repeatedly | `D08` |
| Drop off at upload/connector/permission steps | `D09` |
| First task succeeds, week-2 retention collapses | `D10` |
| Team can't explain why quality changes | `D11` |
| A few enthusiasts use it, team never standardizes | `D12` |

### By what metrics show
| Metric pattern | Most likely diagnostic |
|---|---|
| Signup → first-task funnel collapses before output | `D01`, `D02`, `D09` |
| High generation count, low export/save/approve | `D03`, `D04` |
| High prompt rewrite rate | `D02`, `D05` |
| Repeated identical corrections | `D08` |
| High approval rate + low review time | `D06` |
| Low connector or permission grant rate | `D07`, `D09` |
| Day-7 or day-14 retention collapses | `D10` |
| Surprise regressions after model/prompt changes | `D11` |
| Pilot stalls before team rollout | `D12` |

---

## Multi-diagnostic sequencing

When two diagnostics match equally:

- **`D01` + `D02`** → Run `D02` first; fixing the prompt surface usually unlocks `D01`.
- **`D03` + `D06`** → One product, two cohorts. `D03` for the cohort that doesn't act; `D06` for the cohort that acts too quickly.
- **`D04` + `D05`** → Run `D05` first; fixing output shape on top of wrong context only polishes a wrong answer.
- **`D07` + `D09`** → Run `D09` first if friction is at connector/upload; `D07` first if friction is at agent action.
- **`D10` + `D11`** → Run `D11` first; you can't build a habit on an unpredictable surface.
- **`D12` + anything** → Diagnose the individual problem first, then layer `D12` for team rollout.

---

## Confidence rule

Calibrate confidence to evidence:

- **One clean match** → commit. Name the diagnostic and recommend 2–4 cards.
- **Two plausible matches** → name both, apply the multi-diagnostic sequencing rules above, and tell the user which to run first and why.
- **Three or more plausible matches** → don't guess across the deck. Recommend `W01 AI Adoption Triage` — the meta-workshop exists for exactly this case.
- **No clean match yet** → ask one targeted clarifying question (a user quote, a metric, an observed behaviour). Don't speculate; one good question beats four hedged diagnoses.

Never spread a recommendation thin across all twelve diagnostics. A diagnosis the user can't act on is worse than asking one more question.

---

## Agent safety default

For agent or autonomy questions, prefer conservative recommendations:

- Start **read-only or draft-only**.
- **Require review** before write actions.
- Make **tool calls visible**.
- Show **scope and stop conditions**.
- Require **approval before irreversible or external actions**.

This default maps directly onto Stack 07 (cards 51–58): `51 Permission Staircase`, `52 Reversible Autonomy`, `53 Review Before Write`, `54 Tool-Use Transparency`, `55 Agent Scope Card`, `56 Stop Button`, `57 Escalation Threshold`, `58 Agent Ambiguity Pause`. If the user proposes an agent design that skips any of these, surface the gap before recommending cards.

---

## Response patterns

### Entry points by input type

- **User quote or transcript** → identify the strongest single symptom; name the diagnostic; recommend 2–4 cards; name a workshop if shipping.
- **Metric drop** → translate metric → likely diagnostic(s). Ask one clarifying question only if it would change the diagnostic; otherwise commit.
- **PRD, design, or roadmap** → identify which adoption moments are unaddressed (empty prompt? trust gap? agent scope? privacy UX?). For each gap, name the diagnostic, then the cards. Flag cards that pair (the PDF lists pair-with relationships under each card).
- **"What cards exist for X"** → use the stack index above. Don't invent card numbers or names — every card 1–80 is listed.
- **Card body lookup** → quote from the PDF rather than paraphrasing. Every card has eight named fields; the structure is part of the value.

### Citation conventions

- Action cards: `NN Card Name` — e.g., `29 Evidence Trail`, `36 Preference Ledger`. Card numbers are stable.
- Diagnostics: `DNN Name` — e.g., `D03 Trust Gap`.
- Workshops: `WNN Name` — e.g., `W06 Output Quality Rubric`.

### Default response schema

Use this structure unless the user asks for another format.

#### Diagnosis
`DNN — Name`
Reason: one sentence on why this diagnostic fits the symptom.

#### Recommended cards
1. `NN Card Name` — why this applies here.
2. `NN Card Name` — why this applies here.
3. `NN Card Name` — why this applies here.

#### Implementation move
The smallest next product/design action is… (one or two sentences — what to ship this week, not the full roadmap).

#### Metrics to watch
- Metric 1 — what it tells you.
- Metric 2 — what it tells you.

#### Workshop
`WNN — Name`, if the user is ready to operationalize. Omit if the conversation is still in diagnosis.

### When to depart from the schema

- **Single card body request** → quote the PDF, skip the schema.
- **PRD or design critique** → use a per-section critique format instead.
- **Yes/no question** → answer directly first; add schema only if useful.

---

## What not to do

- **Don't invent cards.** Cards 1–80, D01–D12, and W01–W12 are the entire taxonomy. If something doesn't fit, say so.
- **Don't recommend more than 4 cards per response** unless asked. The deck's value is selectivity.
- **Don't skip the diagnostic step** to jump straight to cards — the diagnostic frames whether the cards are even pointed at the right problem.
- **Don't paraphrase the eight-field card structure** as your own framework. Treat it as the deck's contribution.
- **Don't treat the deck as a checklist.** Every card carries a *when to use it* gate — surface it.
- **Don't restate the deck's philosophy** in every response. Diagnose, cite, move on.

---

## If the PDF is not attached

This skill file alone gives you the index, the triage table, and the workflow. Without the PDF you can:

- Diagnose by symptom and name the right cards by number and name.
- Sequence multi-diagnostic cases.
- Point to the matching workshop.

Without the PDF you **cannot**:

- Quote a card's *why it works*, *how to apply it*, *what to avoid*, *example*, or *metrics* — those bodies live in the PDF.
- Render the fillable workshop deliverable templates.

If the user needs the body of a card and the PDF isn't attached, ask them to attach it (or share the relevant page).

---

## Provenance

- **Title:** The AI Product Adoption Deck
- **Author:** Paulius Uza
- **Format:** 124-page PDF · 104 cards total (12 diagnostics + 80 action cards + 12 workshops)
- **Home:** https://aiproduct.cards
- **License:** Org-wide sharing rights included with every tier.