Vertical AI pilot scorecard

Blank worksheet, security / SoR / model-risk parallel-path checklist, filled fictional example, and AI prompt for scoring a Vertical AI (domain workflow) pilot toward PO or MSA.

Blank worksheet

Update weekly. Bring this to midpoint and to the commercial ask. Bound scope to one named workflow / one team / one region unless the agreement says otherwise.

Criterion	Baseline	Target	Current	Status	Owner
1. ___	___	___	___	[ ] G [ ] A [ ] R	___
2. ___	___	___	___	[ ] G [ ] A [ ] R	___
3. ___	___	___	___	[ ] G [ ] A [ ] R	___
4. ___ (optional)	___	___	___	[ ] G [ ] A [ ] R	___
5. ___ (optional)	___	___	___	[ ] G [ ] A [ ] R	___

Security / SoR / model-risk parallel-path checklist

Start these in week one of serious interest, often before or alongside product proof instead of after "loved the accuracy demo." Track them as first-class workstreams beside product proof.

Workstream	Status	Blocker	Owner	Due
Vendor risk questionnaire	[ ] Not started [ ] In flight [ ] Cleared	___	___	___
SOC2 packet	[ ] Not started [ ] In flight [ ] Cleared	___	___	___
Data residency / retention / DPA	[ ] Not started [ ] In flight [ ] Cleared	___	___	___
Model / vendor risk review	[ ] Not started [ ] In flight [ ] Cleared	___	___	___
API / export to system of record	[ ] Not started [ ] In flight [ ] Cleared	___	___	___
Data room / sample corpus	[ ] Not started [ ] In flight [ ] Cleared	___	___	___
MSA / PO path	[ ] Not started [ ] In flight [ ] Cleared	___	___	___

Working steps

Paste written success criteria into the scorecard before kickoff: cycle time, rework/error rate, throughput per FTE, win rate / close rate, time-to-complete matter/claim/job. No soft "loved the accuracy demo," hallucination-only, or NPS-only lines.

If there is a separate unpaid PoC / free-credits / AI-lab eval clock, name it. Put it on a separate sheet and keep it out of the paid pilot scorecard.

Update status every week. Treat amber without a next action as red.

Bring practice lead and economic buyer (or clear proxies) to midpoint. Re-score against written criteria only.

Before the commercial ask, confirm SoR integration, SOC2, data residency, and model risk are cleared or waived in writing, and that phase-two workflow / team / ACV intent is still in the agreement.

Filled example (fictional: personal-auto FNOL / West / 3 adjusting teams)

Teaching example only. Pattern-check your deal; do not cite these numbers as a real case study. Domain: insurance claims Vertical AI, separate from Industrial / AI.

Criterion	Baseline	Target	Current (wk 7)	Status	Owner
FNOL to coverage decision (median days)	4.2d	≤2.5d	2.3d	Green	Shah / you
Rework / reopen rate (pilot cohort)	11%	≤6%	5.4%	Green	Claims ops
Throughput (claims/FTE/week)	18	≥24	25.1	Green	You + finance
Coverage accuracy (spot audit)	n/a	≥97%	97.6%	Green	QA / you
Guidewire API + SOC2 + model risk	Not started	Live before ask	Live wk 6	Green	IT / Vendor risk

AI prompt

Paste into ChatGPT or Claude. Replace bracketed placeholders with your deal facts.

You are helping a Vertical AI founder (domain copilots / workflow AI sold into a specific vertical such as legal, insurance, construction, real estate, or education; not plant-floor Industrial / AI) run a paid pilot that must convert to a PO or MSA with an enterprise or mid-market practice / ops buyer.

Company / product: [YOUR COMPANY + PRODUCT]
Buyer account: [COMPANY + VERTICAL + SEGMENT]
Scope in pilot: [ONE NAMED WORKFLOW / ONE TEAM / ONE REGION / ONE BOOK OF WORK]
Pilot window: [START] to [END]
Target phase-two ACV: [AMOUNT]
Practice / operator champion: [NAME / TITLE]
Economic buyer: [NAME / TITLE, e.g. COO / CFO / practice managing partner]
System of record / workflow owner: [MATTER MGMT / CLAIMS / ERP / EHR / LMS / OTHER + STATUS]
IT / integration (API/export): [STATUS]
Security / SOC2 / data residency / model risk: [STATUS]
Unpaid PoC / free credits / AI-lab eval (if any): [N/A / SEPARATE CLOCK / STATUS]
Compelling event: [SoR MIGRATION / HEADCOUNT FREEZE + VOLUME / AUDIT FINDING / COMPETITOR WIN / BOARD CYCLE-TIME MANDATE / M&A CONSOLIDATION / OTHER]

Produce:
1) A markdown scorecard table with columns:
Criterion | Baseline | Target | Current | Status (green/amber/red) | Evidence | Owner
Include 3-5 underwritable criteria (cycle time, rework/error rate, throughput per FTE, win rate / close rate, time-to-complete matter/claim/job). No soft "loved the accuracy demo," hallucination-only, or NPS-only lines.
2) A second table: Stakeholder | Role | Attitude | Next action this week
(Roles: practice champion, EB, SoR/workflow owner, IT, security/model risk, procurement. Flag it if the contact is AI enthusiast only)
3) A third table for the parallel path: Workstream | Status | Blocker | Owner | Due
(Include vendor risk questionnaire, SOC2 packet, data residency/DPA, model/vendor risk, API/export to SoR, data room, MSA/PO path)
4) End with 5 bullets: biggest risk, whether to ask for PO/MSA this week, scope-cut options (one workflow / one team / one region; hold list price and per-seat/per-matter pricing), one quick-win memo angle for EB + practice sponsor, one thing to stop doing.

Rules:
- Invent no fake metrics; use ___ when unknown.
- Every status needs a next action.
- Prefer reduce workflows / teams / regions over discounting list price or per-seat pricing.
- Practice champion / AI enthusiast is not the economic buyer.
- Unpaid PoC / free credits / AI-lab eval is not a paid customer pilot. Keep the clocks separate.
- Vertical AI is distinct from Industrial / AI (plant-floor OT). Keep the domain workflow framing.
- SoR integration, SOC2, data residency, and model risk often start before or alongside the pilot instead of after the accuracy demo.