Turning a thirty-year research methodology into a working AI product.

The situation #

A consumer-insights firm with a proprietary qualitative research methodology had spent three decades refining how they deliver behavioral insights to enterprise marketing teams. Their core deliverable was a written "blueprint" describing how a target audience subconsciously responds to a product or category. Every blueprint required a senior researcher to produce and a partner to deliver. The methodology was sharp. The delivery model didn't scale.

What we found #

The bottleneck wasn't the research, it was the format. Marketers wanted to ask follow-up questions, test creative against a blueprint, run different audience cuts. None of that was possible without booking another consulting hour. The blueprint was a static document where it should have been an interface.

A second problem sat underneath the first. Any AI layer dropped on top of the methodology had to preserve the firm's intellectual property. A vanilla LLM with the blueprint pasted in as context would expose proprietary structure the first time someone tried a prompt injection.

What we built #

We started inside the firm, not outside it. The pipeline researchers already used to produce blueprints had manual steps that didn't require human judgment, and automating those gave the team back hours per blueprint and produced a clean dataset to build a customer-facing layer on top of.

The B2B product is a chat interface where marketers can query a blueprint directly. Underneath, a vector store handles retrieval and a custom instruction stack enforces the firm's methodology at the system level, isolated from anything the customer sees. Each client engagement can run multiple agents with distinct voices: an audience-persona agent, a brand-voice agent, and an evaluation agent that stays out of character. The architecture is platform-agnostic, with paths to run on different LLM providers as cost and customer requirements change.

What changed #

Insights that used to require a consulting engagement now happen in a chat window. The firm's methodology became a product that scales, while the IP that took thirty years to build stays protected at the layer customers never see.

This is the kind of work the AI Operations Audit is built to scope. Find the part of the business that doesn't scale, then build the system that fixes it. Start an audit →