# Idea 3 — AI Transformation Scorecard & Use-Case Triage

**What Adrian builds:** the scorecard he will show Dana and the exec team every week on the Claude rollout — who has activated, who is actually using it, whether leadership is "walking the walk" — plus a triage of the 45 use cases employees submitted, sorted into *Now / Next / Later* with a reason and an owner.

**Why this one for Adrian:** Strategic Initiatives is driving AI transformation across Aurex. Claude went enterprise-wide on March 3; the bottoms-up momentum is real but leadership is behind and Dana wants top-down acceleration visible. Adrian is the person who has to report on it. This build turns the admin export, the comfort survey, the use-case intake form and the `#ai-wins` channel into something he can stand behind — and it is the most natural thing for him to demo at 14:30, because it is *about* the program the offsite is part of.

**Session fit:** best as the afternoon build (13:45–14:30) or the advanced-track build, after a morning win on Idea 1 or 2. Steps 1–2 are quick; Step 3 is the triage, which is where the judgment lives.

---

## How to run it

1. Claude desktop app → new Project *AI Transformation* (a Project is worth it here; he'll reuse it weekly).
2. Add the `data/` folder.
3. Paste the prompt; iterate with the follow-ups.

---

## The prompt

```text
You are helping me, Adrian Vale, SVP Strategic Initiatives at Aurex. My team (Sulaiman Rao leads it) is driving AI adoption across the company. We rolled Claude Enterprise out to ~1,500 people on March 3, 2026; Engineering has had early access since October 2025. Adoption so far has been bottoms-up; the CEO, Dana Brooks, wants leadership visibly driving it, and she wants to see progress weekly. I need two things: a scorecard I can stand behind, and a triage of the use cases people submitted so we fund the right ones first.

Attached data:

- claude_enterprise_usage_export_2026-03-09.csv — the admin-console usage export, one row per seat (~1,500). Columns: email, name, department, office, level, cohort, seat_assigned, seat_status (Active / Invited — not activated), first_login, last_active, messages_7d, messages_30d, projects_created, connectors_authorized, claude_code_used. The executive team is in here by name.
- ai_comfort_survey_responses.csv — Google Forms export from the first training cohort (68 responses): comfort 1–5, tools used, capability of interest, tasks they'd hand off, biggest blocker.
- ai_use_case_intake.csv — Google Forms export, 45 submitted use cases: department, title, description, systems it touches, frequency, hours per week, how many people do it, data sensitivity, willingness to pilot in 30 days.
- approved_ai_tools.csv — the sanctioned tool inventory with each tool's connectors and the maximum data classification allowed. Note: Claude is cleared for Confidential but NOT for Restricted (client PII / MNPI) until the data addendum is signed, expected end of March.
- slack_export/ — standard Slack export of #ai-wins (users.json, channels.json, one JSON per day). It contains wins, one honest failure, a policy question and an IT gap.

Work in four steps, pausing after each.

STEP 1 — The adoption picture.
From the usage export, compute for the March 3 enterprise cohort (exclude the Engineering early-access cohort, then show it separately as the benchmark): seats invited, % activated, % weekly-active among activated, median messages_7d, % with at least one connector authorized, % who created a Project. Break each down by department and by level (IC / Manager / Director / VP+). Then a separate small table for the executive team by name: activated yes/no, last active, messages, connectors. Be matter-of-fact about who on the exec team has not activated; I will handle the conversation.
Finish with five sentences on what the numbers say, and two hypotheses for the department gaps that I could test (the Slack export contains one concrete explanation for one department — find it).

STEP 2 — What people want and what's stopping them.
From the comfort survey: distribution of comfort scores overall and by department; the top 10 "tasks I'd hand off" themes (cluster the free text and count); and the top blockers. Map the blockers to what would remove each one (training, a policy one-pager, connectors, the data addendum). Pull the three most instructive posts from #ai-wins — including the honest failure — as quotable examples for the exec team.

STEP 3 — Triage the 45 use cases.
Score every submitted use case on an explainable scale you define and state, using at least: impact (hours per week x people doing it), feasibility now (touches only Gmail / Drive / Calendar / Slack = connectable today; Jira / Salesforce / NetSuite / internal systems = needs integration work), data-sensitivity fit (Restricted = blocked until the addendum; Confidential = OK; flag MNPI separately), and the submitter's willingness to pilot in 30 days. Put each into NOW (start this month), NEXT (after the data addendum / an integration), or LATER, with a one-line reason and a suggested owner department. Output a table sorted by bucket then score. Then call out: the five highest-impact NOW items, any duplicates across departments that should be built once, and anything that should be declined with a reason.

STEP 4 — Build the scorecard.
Produce one self-contained HTML file, ai_transformation_scorecard.html (inline CSS/JS, no external libraries, offline). Sections: (1) headline tiles — invited, activated, weekly active, exec-team activation; (2) department bars for activation and weekly-active, with the Engineering early-access benchmark shown as a line; (3) the exec-team table; (4) comfort distribution and top blockers; (5) the use-case triage table with a bucket filter; (6) three quotes from #ai-wins; (7) a "This week's asks" box with what I need from IT, Legal (data addendum) and each exec; (8) footer with data as-of date and sources.
Then write a 150-word note from me to the exec team for Tuesday staff: the numbers, what's working, the one thing I'm asking each of them to do this week.

Constraints:
- Use only what is in the files; do not estimate numbers that aren't there.
- People's names are fine inside the company; do not rank individual ICs by usage in anything that would be shown outside the exec team — aggregate to department.
- Keep the tone neutral about who is behind. The point is to make it easy to catch up, not to embarrass anyone.
```

---

## Follow-up prompts for iteration

1. **Weekly repeatability.** "Write `WEEKLY_SCORECARD_RUNBOOK.md`: the admin export to download each Monday, the exact prompt, and a 'what changed since last week' section I paste previous numbers into."
2. **Per-exec pack.** "For each executive, produce a half-page: their department's numbers, the two use cases from their team in the NOW bucket, and the one thing I'd ask them to do this week."
3. **The addendum case.** "Sum the impact (hours x people) of every use case that is blocked only by the data addendum. Draft three sentences for Stefan on why it matters that it is signed by March 31."
4. **Training plan.** "Using the survey's comfort distribution and blockers, propose a three-session training plan for the next month: who it's for, what it covers, and what each session's 'ship something' exercise would be."
5. **Kill the vanity metric.** "Which of the metrics on the scorecard would you remove because it can be gamed or doesn't predict real adoption? What would you add instead?"

---

## Making it real the week after

- The usage export is the shape of the Claude Enterprise admin console's seat/usage export → Sul downloads it each Monday to a Drive folder.
- The survey and intake form are already Google Forms → Drive connector on the response Sheets.
- `#ai-wins` → Slack connector.
- Ask Mariel's team to (a) turn on the Slack connector for the Treasura workspace, and (b) confirm the data-addendum date with Stefan — both show up as gaps in the data.

---

## Facilitator notes (not for Adrian)

**Planted in the data**

| What | Where |
|---|---|
| Exec team: Stefan Adler and R. Visser have not activated; Bellamy and Marc Ashford logged in once; Adrian, Monica and Sul are the heaviest users | `claude_enterprise_usage_export_2026-03-09.csv` (search by name) |
| Treasura and Compliance lag badly; Engineering/Product/Marketing lead | same, `department` |
| Treasura staff are still on their own Slack workspace — Slack connector isn't on for them; explains part of the gap | `#ai-wins` 03-06 (Ravi) |
| Priya's honest "confidently wrong on FX reval" post — the failure example the prompt asks for | `#ai-wins` 03-05 |
| Nadia's policy question + Sul's answer — Confidential OK, Restricted not | `#ai-wins` 03-04 |
| Sul's "~55% activated" Slack estimate is computed from the export; Claude can verify it | `#ai-wins` 03-06 |
| Several high-impact intake items touch Restricted data (sanctions triage, intercompany rec, margin calls) → NEXT until the addendum | `ai_use_case_intake.csv` |
| Near-duplicates across departments: "regulator letter summarizer" (Compliance) vs. "case law & reg update digest" (Legal); "meeting follow-up drafter" vs. "pre-call briefing" (Sales); two "ticket response" items (CS, Treasura) | intake |
| Two of the intake items are Adrian's own (integration roll-up, market activation view) — i.e. Ideas 1 and 2 in this folder | intake `UC-008`, `UC-045` |
| The approved-tools list shows Perplexity/ChatGPT as unsanctioned-but-used, mirroring the real exec survey | `approved_ai_tools.csv` |

**Timing:** Step 1 ≈ 10 min, Step 2 ≈ 10 min, Step 3 ≈ 15 min (the scoring debate is the point), Step 4 ≈ 15 min.

**Sensitivity:** the exec-team usage table will be the most talked-about thing in the room if Adrian demos it. Check with him before the demo whether he wants to show the named table or the department aggregate. The data is synthetic, but the names are the real exec team's, so the room will read it as real.