Kav AI — The 90-Day Pilot
A structured evaluation against your own facility data, with success criteria agreed in writing before it starts.
The fastest way to test whether Kav AI delivers value at your facility is a 90-day pilot run against your own inspection data. The pilot is not a demo: it runs on your real imagery, your real facility, and your real operational questions — and success or failure is measured against criteria agreed in writing before data ingestion begins, never assessed retrospectively.
Scope
| Parameter | Standard definition |
|---|---|
| Asset class | One per pilot. Recommended: pressure vessels, heat exchangers, or piping circuits — the classes with full API 581 automation support. Expansion available in Phase 2. |
| Facility | One facility or one defined area (a process unit, tank farm, or compressor station). |
| Data | Visual inspection imagery (RGB, thermal, or OGI) from at least one completed campaign. SCADA / historian connection is optional — included only if you elect to connect. |
| Duration | 90 days, from data ingestion to a structured success-evaluation meeting. |
Scope to clarify up front: atmospheric storage tanks are screening-only under API 653 methodology and require engineer review — they do not generate automated Tier 1 actions. Rotating equipment and pipelines are out of scope for v1. This equipment-class boundary is communicated before onboarding so the pilot never tests an expectation the platform does not claim.
Timeline
| Weeks | Phase | What happens |
|---|---|---|
| 1–2 | Onboarding | Data transfer and ingestion; 3D facility model generated from your imagery; OPC UA configuration if elected; baseline calibration begins. |
| 3–4 | Baseline | First anomaly detections reviewed by your pilot lead against known historical defects. |
| 5–8 | Active use | Your team uses the platform for real integrity queries; mid-pilot review at Week 6. |
| 9–10 | Validation | Findings cross-validated against your inspection reports and CMMS records; false-positive and false-negative rates measured; confidence calibration curve generated. |
| 11–12 | Evaluation | Structured debrief against every agreed criterion. Decision: proceed, extend, or conclude — with data deletion executed if not proceeding. |
Note on the first weeks: the first campaign at a new facility is treated as baseline data collection — early out-of-distribution flags are expected and are used to calibrate detection, not surfaced as alerts. Weeks 5–12 represent steady-state performance.
Success criteria (the standard set — amendable by agreement)
| # | Criterion | Standard threshold |
|---|---|---|
| SC-1 | Anomaly detection recall | ≥ 80% of known defects from historical reports identified without prompting |
| SC-2 | False positive rate | ≤ 20% of flagged anomalies assessed as false positives by your pilot lead |
| SC-3 | Query response quality | ≥ 70% of natural-language queries rated “useful” or better |
| SC-4 | Time to first finding | First anomaly surfaced within 48 hours of completed ingestion |
| SC-5 | Response latency | ≤ 5 seconds (P95) for natural-language queries |
| SC-6 | SCADA correlation (if connected) | ≥ 1 confirmed correlation between an IOW exceedance and a visual anomaly at the same asset |
What each side commits
| Kav AI | You |
|---|---|
| A dedicated customer-success engineer for the full pilot | A named pilot lead (integrity engineer or operations manager) |
| Written progress update every week | Historical inspection reports for cross-validation |
| Encryption in transit and at rest; access logging | Availability for three structured review sessions |
| Full data deletion within 5 business days, with certificate, if you do not proceed | A named decision-maker at the evaluation meeting |
Commercial terms
Fixed fee agreed before commencement, credited against the Year 1 subscription if you proceed. All inspection data remains your property throughout and after; Kav AI never uses pilot data for model training without explicit written consent; mutual NDA before any transfer. You may terminate at any time with 5 business days’ notice. Kav AI’s liability during the pilot is capped at the pilot fee; all outputs are recommendations to your qualified operators. If the success criteria are met, you receive a production proposal within 10 business days of the evaluation meeting — including, where preferred, a partner-integrated option where a qualified mechanical-integrity firm staffs the engineer-review seat under a single procurement vehicle.
Kav AI Platform · PRD v3.12 · 2026-09-03 · Full framework: PRD Appendix G · contact: sales@kavai.net