Every PRISM program runs on telemetry from week one — usage, adoption, cost, and impact, tracked weekly against a written baseline. Our fees are priced against it.
Every band reports into the same four dials.
The framework
If a number matters to the board, it lives on one of these.
DIAL 01
weekly active
The leading indicator — impact, cost, and safety all trail it by weeks.
DIAL 02
hrs vs. baseline
Hours saved against the written baseline — the dial outcome pricing bills against.
DIAL 03
volume on fast tier
Spend attributed per workflow — unit economics, not a lump sum labeled “AI.”
DIAL 04
evals passing
Incidents, overrides, eval pass rates — the dial diligence asks about.
Week one
ROI is defined per use case before the pilot — one claim each, measured, never claimed after the fact.
Current state measured, methodology agreed in writing.
Skills go live inside the function.
Hours and cycle times against the signed baseline.
If the number didn’t move, you don’t pay.
Why it holds
The console
A real program week, reconstructed — all four dials on one screen.
ANNUALIZED · METHODOLOGY v1.0, SIGNED WK 1
Anomaly · Marketing premium-tier usage 3.1× last month — routed for review.
The same numbers go to the function owners, the committee, and the sponsor. Nobody gets a different version of the truth.
The silent margin killer
We treat inference cost as a margin line from day one — not a lump sum on the IT line that nobody owns.
Cost per invoice coded, per ticket deflected — a unit cost the CFO puts next to the unit value.
Frontier models where the stakes are high, fast models for volume — a routing table, not a culture campaign.
The spend report explains itself: “Finance moved down a tier — spend fell 40% while throughput rose.”
The adoption curve
Every rollout follows the same curve — the difference is what happens in the trough.
Week 1
Everyone tries it. Curiosity isn’t adoption — this week’s numbers predict nothing.
Weeks 4–8
Novelty fades and usage drops. This is where unmanaged programs die — one lapsed user at a time.
Month 6
Usage climbs past the spike and stays — 84% weekly active, sustained. It’s how the work gets done.
Leading indicators
Invocations per user, feedback volume, and time-to-second-use turn down weeks before the value line moves — so the intervention starts early.
The intervention playbook
A usage drop triggers office hours the same week. Killing a dead skill is telemetry working, not the program failing.
Honest benchmarks
The planning numbers we put in business cases — smaller than the headlines, measured in production, real.
Engineering
Once review and coordination are counted — plan on the lower band.
Customer support
Issues resolved per hour — gains concentrate where experience is thinnest.
Proposals & RFPs
Speed and quality move together on document-heavy work.
Program-level
Full P&L impact lands in 7–12 months — plan against the floor.
Buyer beware
The cadence
The same four dials roll up three altitudes — no re-cutting, no reconciliation meetings.
Weekly
Function owners
Cost, safety, adoption, function KPIs — refreshed every week, no meeting required to read it.
Monthly
Governance committee
Dials reviewed, escalations cleared, expansions approved — thirty minutes when telemetry is healthy.
Quarterly
Board & sponsor
Hours against baseline, spend against budget, the safety record, what scales next.
Repeatable
One reusable page — baseline, measured result, cost, projected impact at scale — justifies expanding any pilot that proves out.
The foundation
Telemetry runs under every band alongside two sister layers.
The point
The Rest of the Spectrum
Ready to move
We’ll come back with the current-state numbers, the methodology in writing, and the four dials we’d report against — before anything ships.
Talk to LightCI