- A working pilot using the approved source set.
- Answers that point back to the source material.
- A focused evaluation against sponsor-provided questions.
- An access and governance brief.
- A findings walkthrough with Blake Smith.
- A clear Stage 2 recommendation or a documented stop.
For retail AP and operations teams
Turn one approved policy library into an answerable, cited pilot.
MarginHero runs a fixed-scope AP Knowledge Agent pilot that lets your team test a governed policy experience without committing to a broad AI program. It is read-only, approval-first, and designed to produce an honest go or no-go in about three weeks.
A contained first decision
One repository. One operating use case. One named sponsor.
Stage 1 is deliberately narrow. Your sponsor selects one approved policy or procedure library and a small set of real questions the team should be able to answer. MarginHero builds and evaluates the pilot against those sources before anyone discusses a larger rollout.
- Read-only source access.
- One approved repository and one use case.
- Human approval at consequential steps.
- No confidential former-employer material.
- No production promise before the pilot is evaluated.
- Change requests move to a separate Stage 2 scope.
Governed by design
Your team stays in control of the source, access, and decision.
The pilot starts only after the sponsor approves the source set, access model, and evaluation questions. Unsupported answers are treated as failures, not polished around. Stage 1 ends with an evidence-based decision about whether the use case deserves a larger implementation.
Operator-led
Retail AP judgment on one side. Practical AI building experience on the other.
Blake Smith spent 20 years across big-box and specialty retail asset protection and built an internal AI knowledge agent in his most recent corporate role. MarginHero combines that operating context with a small, decision-focused pilot rather than a generic AI demonstration.
The decision standard
A useful stop is better than an expensive maybe.
A Stage 1 pilot can end with a recommendation not to proceed. The work is successful when the sponsor has a working example, traceable evidence, and enough operating context to make the next decision without pretending the technology is ready for more than it is.