EXECUTIVE BRIEFING
PSG EXECUTIVE BRIEFING · NO. 05
The 90-day AI proof-of-value plan.
Exactly what a first AI engagement should deliver by day 30, 60, and 90 — the evidence gates that separate momentum from sunk cost, and the kill criteria that make a pilot safe to fund.
PLATINUM STRATEGY GROUP · 2026
WWW.PSG-INC.COM
Month one: measure the problem, then build against it.
The defining act of a disciplined first engagement happens before any AI is configured: the manual process gets measured. Hours consumed, cycle time, error rate, and the fully-loaded cost of the work — captured in week one, signed by the process owner. This baseline is what day 90 will be judged against, which means value will be demonstrated, not asserted. Skip it, and the end-of-pilot meeting becomes a matter of opinion.
WHAT EXISTS BY DAY 30
A signed baseline. One page of measured numbers for the target workflow — the “before” photo nobody can dispute later.
One scoped use case. A single workflow with one owner, clear inputs, and measurable output — not a “transformation program.” Breadth is the enemy of proof.
A working first version. Rough, on real data, in front of the people who own the process. Feedback on a working tool in week four beats a perfect tool in week twelve.
Written success and kill criteria. The accuracy bar, the adoption bar, and the evidence that would stop the project — agreed while everyone is still objective.
Gate one (day 30). Does the first version, on real data, show enough signal that the process owner wants it to continue? If the answer needs a caveat-laden slide to explain, the honest answer is no — rescope now, at 20% of budget spent, not at day 90 with all of it.
Month two: run it beside the real process and keep score.
The tool now runs in shadow mode — doing the real work in parallel while the manual process remains the system of record. Every output is graded by the people who own the workflow: correct, usable-with-edits, or wrong. This is where trust is either earned or the project fails cheaply. Nothing goes live on enthusiasm; it goes live on a scorecard.
WHAT EXISTS BY DAY 60
A graded scorecard. Several weeks of accuracy data on real volume, scored by operators — not a vendor demo on curated examples.
A tuned system. The gap list from grading — retrieval misses, edge cases, format failures — worked down weekly, with the trend line visible.
A control design. The review step, exception queue, and audit evidence defined — so going live doesn’t create the governance debt described in Briefing No. 04.
Early value evidence. Hours already saved in shadow (drafts used, exceptions caught) — the first entries in the value ledger, logged against the baseline.
Gate two (day 60) — the kill criteria in action. Accuracy flat for three consecutive weeks below the bar; operators routing around the tool; or data quality problems whose fix costs more than the value at stake. Any one of these, documented, ends or rescopes the project — and saying so out loud is what makes the pilot fundable in the first place.
Month three: go live, prove the number, decide what’s next.
The tool becomes the system of record for its workflow — with its review step operating, its exceptions queued to a named owner, and its usage logged. The final two weeks produce the only deliverable executives remember: the before-and-after, in the baseline’s own units. Hours per cycle, days per close, error rate, dollars — measured, not estimated.
THE DAY-90 DECISION PACKAGE
The measured result
Baseline vs. day-90 actuals, one page. If the 2:1 year-one math from Briefing No. 03 holds, scaling is an easy yes.
The operating runbook
Owner, cadence, retraining schedule, and the audit-ready control docs — proof the system survives the consultants leaving.
The ranked next list
Three-to-five candidate workflows scored on the same value-and-effort basis, so the second project starts from evidence.
WHY THIS STRUCTURE CONVERTS SKEPTICS
Bounded downside. Two evidence gates mean the worst case is a documented, early, inexpensive stop — not a year of sunk cost.
Operator ownership from day one. The process owner grades every output, so go-live lands on a team that already trusts the tool it helped shape.
A number the CFO can repeat. The engagement ends with a measured result in business units — the sentence that funds everything that follows.
Want this plan scoped to one of your workflows?
A 30-minute diagnostic consultation — the right first use case, the baseline to capture, and a costed 90-day plan within 48 hours.