Have a tricky problem that you think requires a rewrite of your software? Join as a guest on our podcast and we’ll talk you out of it.

How it went

Discovery from the transcripts

Requirements came from recorded stakeholder meetings, not a ticket queue. They became written personas for the pricing lead, the data contact and the operations staff, plus a short list of invariants no change may regress, with overridden requirements kept in a changelog.

Build in the open

A browser app with no build step, Vercel functions and a Postgres store. Clean ERP exports are mapped deterministically. Messy workbooks, with title rows, two-level header bands and multi-sheet masters, fall back to an agentic step where Claude proposes a column mapping that a person confirms before anything is scored.

Rollout

The four must-haves shipped in the first weeks: persistent classification overrides, multi-factor weighting, pack-size roll-up and customer roll-up. Then a shared team dataset, a formula-first view, and Slack and email alerts.

What they got

  • An explainable action queue in the five words the pricing team already used: raise, optimize, evaluate exit, protected hold, and no action
  • Protection for low-margin SKUs tied to high-value customers, so a cut never costs a key account
  • Overrides that persist, expire and land in a review queue
  • Pure, tested logic for ingest, classification and overrides, separate from the UI, with a drift test that fails if the in-browser worker and the library disagree
  • Schema-only telemetry and an opt-in shared dataset, so SKU and customer data stays theirs
  • A user guide built on synthetic data, pinned by a test so no real figures leak into it

Why this is a forward-deployed story

Nobody handed over a spec. The engineer sat in the meetings, wrote down what the people doing the work needed, built the tool in front of them, and changed it as they used it. The AI parts are small and bounded: Claude maps a messy spreadsheet, a person confirms, and deterministic code does the scoring. That’s the shape we aim for on every engagement.

Isaac did this work as an embedded engineer through Operating Co., whose model is senior operators paired with production engineers inside PE-backed and mid-market companies. About six people used the console in its first run: three sales leaders, the product manager and the commercial controller, with the sales reps next. The client isn’t named here, and the method is the same one we bring to every YGG engagement.

Have a team doing this by hand?

Book a call. We’ll talk through what a two-week discovery with your stakeholders would produce.

Book a call How we work & pricing

Prefer email? start@ygg.llc