Vertical AI pilots

Practice leads and AI enthusiasts will hand you data-room access, free model credits, a "try it on our corpus," and an innovation-lab eval without much thought. None of that makes a pilot. A convertible Vertical AI pilot has a named workflow or book of work (one claim queue, one matter type, one job family, or one school-district process, rather than the whole vertical), a compelling event, written criteria the practice lead and economic buyer can underwrite, and a path to a PO or MSA if the proof lands.

Keep unpaid model evals, free-credit experiments, design-partner content co-builds, and open-ended AI innovation labs separate from customer pilots. Lab work can create evidence later, and it seldom closes deals this quarter. If one SOW mixes unpaid eval scope with a paid commercial path, name which clock you are on.

Hunt for compelling events: a practice-management or claims-system migration, a headcount freeze with rising volume, an audit or regulatory finding, a competitor win in the same vertical, a PE or board mandate to cut cycle time, or post-M&A process consolidation.

Map these buyers early: the vertical operator or practice-lead champion, the economic buyer (often the COO, CFO, or practice managing partner), the workflow owners who live in the system of record, IT / integration, security / vendor risk / model risk, legal / procurement, and data stewardship. One AI enthusiast won't get you to a PO. Plant-floor OT work belongs to the Industrial / AI overlay.

Vertical AI buyers expect matter, claims, or ERP integration work to start up front. In the first week of serious interest, start the vendor-risk questionnaire, the SOC2 packet, data residency / retention, model and vendor risk review, data-room access, the API or export path to the system of record (matter management, claims, ERP, EHR, or LMS, depending on the vertical), and procurement. Run them before or in parallel with the product proof so you aren't starting after the champion has said they loved the accuracy demo.

Write down the committee. The AI enthusiast or innovation-lab contact is not the economic buyer. The check-writer is often the COO, CFO, or a practice managing partner. Champion enthusiasm without SoR clearance and a budget line dies once the eval ends.

Work backwards from your target ACV and the MSA process. If production requires model-risk sign-off, data residency, or a named workflow owner in the system of record, start those while the pilot runs instead of after the demo.

Write criteria a practice lead and a finance or economic buyer can both sign, and instrument them inside the pilot window on a bounded scope (one named workflow or book of work): time to complete a matter, claim, or job, error or rework rate, throughput per FTE, and win rate or cycle time.

"Model accuracy on the eval set," "hallucination rate looked fine," and NPS-only lines won't pass as success criteria. Price the pilot when you can, and put the phase-two commercial terms (workflows, seats, modules, ACV intent, MSA path) in writing so the pilot can't turn into permanent free credits.

Land a quick win in weeks 1–2 that the practice lead and economic buyer can see without living in the console. A one-page memo with one underwritable metric (cycle time, rework, or throughput on the named workflow) beats another model-accuracy tour.

Hold a formal midpoint against the written scorecard, with the practice lead and economic buyer in the room or represented. Adjust only against the criteria, and don't expand to all practices, all claim types, or a second vertical without saying so.

Keep vendor risk, SOC2, data residency, model risk, SoR integration, and procurement moving each week. Vertical AI pilots miss the migration, the board mandate, or the renewal window when the team finds the data-residency or model-risk stall in the final week.

Restate the criteria, the evidence, and the commercial path from the agreement, then ask for the PO or MSA schedule. "Loved the demo" without a signature is a stall.

If budget or risk appetite wobbles, cut to one workflow, one team, or one region. Hold list price and per-seat or per-matter pricing, and don't teach procurement that a successful Vertical AI pilot earns a permanent haircut.

Name the Vertical AI stall mode before you improvise.

Treat the pilot scope (one named workflow or book of work) as a proof sample. Expand from the phase-two price and the MSA terms in the agreement.

Add adjacent workflows, teams, or regions after production is live. A published win (with permission) helps you sell the next practice lead. If you keep "pilot seat," free-credit, or pilot-SKU pricing going across the firm, each expansion starts at a discount.

Use this as a pattern drill. It is not a claimed case study, and the numbers are illustrative. Vertical AI here means domain workflow AI (insurance claims), separate from plant-floor Industrial / AI.

Use the Vertical AI pilot scorecard template for the blank worksheet, a filled example, the security / SoR / model-risk parallel-path checklist, and a copy-ready AI prompt.