Case study 02 · Station 03 · The Build
From internal tool to platform flagship
An automation I built for our consultants became the company’s flagship AI product. That graduation path was the strategy.
01 ·The problem.
Every enterprise learning deal starts with a solutioning cycle: a senior consultant reads the client’s brief, maps it to frameworks and a product catalog, designs a journey, costs it, and produces a branded proposal. It takes a day or more of scarce senior time, and quality varies with who’s available. Junior consultants couldn’t produce senior-grade output; senior consultants couldn’t scale. Nobody had budgeted for it.
02 ·Discovery and the decision.
The decision was not to pitch a platform. Instead I built an internal tool: free-form learning challenge or call transcript in → parsed requirements → framework mapping → costed, branded brief out (PDF/PPT/DOCX), batch-capable. Real consultants used it on real opportunities. Every complaint was free product discovery: where it hallucinated, where the costing logic broke, where consultants stopped trusting it.
03 ·The build.
The internal tool earned its promotion by proving three things: the workflow was learnable by a machine, the output was usable with minimal editing, and the users’ trust hinged on control, not autonomy. That became the product bar for the flagship copilot: a generated proposal must be usable by a senior consultant with fifteen minutes of editing or less.
The copilot runs on a retrieval architecture grounded in a proprietary skills ontology (224 skills across 9 domains, 658 behavioural indicators, 19 journey templates) — the moat is the knowledge, not the model. Forced structured output ends free-text drift.
04 ·Trust.
An “open questions” confirmation bar surfaces every assumption the system made and makes the consultant approve or correct each one before the proposal finalizes. Roughly 230 automated tests, including adversarial suites that attack the system’s confidentiality boundaries, run before anything ships.
05 ·Launch and learn.
Launch here meant graduation: the internal tool became the company’s flagship AI product, sold as part of the platform rather than shown as a demo. In production, on live enterprise proposals. A day-plus design cycle compressed to minutes; junior consultants producing senior-grade first drafts; two artifacts from one conversation — the client-facing proposal and the machine-readable spec that downstream build systems consume.
The fastest route to an enterprise AI product is an internal tool with real users and real stakes.