Did this AI pilot earn production —or is it time to kill it?
Leadership is forcing the call: scale the pilot or cut it. Score each running pilot on six signals from your own marks and get one honest verdict — SCALE, EXTEND PILOT, or KILL — with a gate that won't let an unmeasured or money-losing pilot pass on polish alone.
Most AI pilots stall in purgatory — and get cut for the wrong reasons.
of agentic-AI projects are forecast to be canceled by 2027 — over cost, unclear value, or weak controls. The cancellation wave starts now.
the year's tell: a chatbot or RPA rebranded as an “agent.” It looks like progress and delivers chatbot results, so the whole program gets blamed.
the single most common reason a pilot gets cut: it was never measured against the target it launched to hit, so no one can prove it worked.
The honest decision isn't “is the AI cool” — it's “has this pilot earned the cost of scaling.” This grades exactly that, and refuses to let a polished-but-unproven pilot read as a win.
Score a pilot and watch the gate do its job.
Mark each signal. The verdict updates live - same math as the workbook.
Was it measured against the goal it launched to hit?
Does delivered value exceed the all-in run cost?
Embedded where people work, or a bolt-on?
Holds up beyond the clean demo path?
One person accountable for the metric?
Plans, executes, and adapts - or just responds?
Scores in the SCALE band, but a gate signal is at 0 - so the verdict is forced to KILL. A pilot you never measured, or one that costs more than it returns, can't scale on polish elsewhere.
Fix first: Measured outcome vs. the target
Your marks only · no benchmark · grades the pilot, not people
One command, every pilot, an auditable verdict.
The zero-dependency Python engine reads your pilot list and prints the same verdict the workbook and the demo produce. Two pilots below score in a passing band and still read KILL — one was never measured, one costs more than it returns.
AI Pilot Proof-of-Value & Scale-or-Kill Verdict
====================================================
Refund triage agent 76/100 KILL [GATE -> KILL]
fix first: Measured outcome vs. the pre-set target
Sales-email drafter 65/100 KILL [GATE -> KILL]
fix first: Cost per outcome vs. value delivered
Invoice-coding assistant 94/100 SCALE
fix first: Genuine autonomy (not agent-washed)
Support-KB summarizer 50/100 EXTEND PILOT
fix first: Measured outcome vs. the pre-set target
Contract-review bot 56/100 EXTEND PILOT
fix first: Genuine autonomy (not agent-washed)
Onboarding email agent 100/100 SCALE
----------------------------------------------------
Portfolio: KILL-FIRST
2 of 6 pilot(s) read KILL.Six signals, weighted to 100 — two of them gates.
Measured outcome = 0 OR cost-vs-value = 0 forces KILL regardless of score. Either alone is fatal — and the gate only worsens a verdict, never lifts one.
round-half-up weighted total, no industry multiplier, no AI scoring. A pilot earns SCALE on its own marks, not on a flattering average.
Lift a triggered signal off 0 and the gate releases — the verdict goes back to whatever the score earned. The fix is always named.
A scale-or-kill decision aid, not a dashboard.
- A deterministic verdict from your own marks on a running pilot.
- The post-pilot companion to the pre-build Agent Use-Case Fit Gate.
- A portfolio rollup that names the worst pilot to act on first.
- Offline — engine, workbook, and demo agree to the number.
- Not connected to your agent, logs, or billing — nothing is read live.
- Not an AI that grades you; your marks drive the verdict.
- Not a guarantee of any outcome, and it scores no person.
- Not a benchmark — no industry multiplier is baked in.
Not financial, investment, or legal advice. This is a planning decision aid that grades a pilot's proof of value from your own marks. It applies no industry multiplier, connects to nothing, and scores no person — the scale-or-kill decision and its consequences are yours.
Anyone being asked to defend an AI pilot.
Decide to build, watch it run, then gate the scale.
The pre-build sibling: should you build this agent at all, before you spend.
ViewReveals the end-to-end success rate a per-step metric hides while the pilot runs.
ViewEvaluates an agent's trajectory and gates CI once you decide to scale.
ViewThe honest answers.
That gate is the pre-build decision: should you build this agent at all, before you spend. This is the post-pilot decision: the thing is built and running, and leadership wants proof it works or it gets cut. One scores a proposal; this scores a running pilot's earned value and returns SCALE, EXTEND PILOT, or KILL.
Because two signals are gates. If a pilot was never measured against the target it launched to hit, or if its all-in run cost exceeds the value it delivers, the verdict is KILL no matter how polished everything else is - a pilot you can't prove and a pilot that loses money are each, alone, a reason to stop. The shipped Refund triage sample scores 76 and KILLs because it was never measured.
Measured outcome vs. the pre-set target (24, gate), cost per outcome vs. value delivered (22, gate), integration depth (16), exception handling at scale (14), a named owner on the number (12), and genuine autonomy rather than an agent-washed chatbot (12). You mark each 0, 1, or 2 from your own read; the weights sum to 100.
No. It is deterministic and offline. You enter your own marks and it computes the verdict - it never connects to a system, reads live usage, or runs an agent. The same logic runs in the workbook, the Python engine, and the on-page demo, so all three agree to the number.
Yes. One row per running pilot. The dashboard rolls them up to ALL SCALE, SOME TO EXTEND, or KILL-FIRST, and names the worst pilot to act on first - so a leadership review has a single honest read across every pilot in flight.
No. It is a planning decision aid that grades a project's proof of value from your own marks - no benchmark, no industry multiplier, and it scores no person. Use it to structure the scale-or-kill conversation; the decision and its consequences are yours.
Stop guessing.
Get the verdict.
One purchase, lifetime access, 12 months of updates. $79, once.
Not financial, investment, or legal advice. This is a planning decision aid that grades a pilot's proof of value from your own marks. It applies no industry multiplier, connects to nothing, and scores no person — the scale-or-kill decision and its consequences are yours.
Sold by RedHub AI LLC · Secured by Stripe · redhub.ai