A research firm that distrusted AI now runs 85% of its projects on custom AI pipelines we built.

AnswerLab is a user experience research firm serving Fortune 500 technology, financial services, and retail clients. Its competitors had begun launching branded AI research platforms, and clients were increasingly asking what AI meant for the research they were buying. Inside the firm, AI use was close to zero.

The resistance was reasonable. Researchers are trained to distrust claims they can’t trace to a source, and the tools available at the time hallucinated often enough that verifying the output cost about as much time as the output saved. Meanwhile the mechanical work went on untouched: every research session produced a recording that a researcher downloaded by hand, cleaned up, re-uploaded to the correct client Drive folder, and summarized before the end of the day, which across dozens of concurrent studies consumed hours that should have gone to analysis.

Our solution was to build a proprietary AI ecosystem that we called AnswerLab Intelligence, or ALI for short. Our first approach was to automate the research workflow end to end, and we built a handful of assistants to do it. Researchers rejected them. Two assumptions had been wrong: there was no shared definition of good across the research team, because every researcher approached the work in their own way and no single assistant could satisfy all of them; and client situations varied enough that the context we could capture and pass to a model never matched what a researcher already knew about the client.

Instead of automating how a study gets done, we automated the components every study depends on regardless of method, starting with transcript pre-processing and session summaries. Both run on the Claude API, which cleans up each raw transcript and drafts the summary a researcher would otherwise be writing by hand at the end of the day. A researcher connects a project once, and from there the pipeline runs on its own, delivering finished documents into the folder the team already works from. Each summary is written against the study’s own stated research objectives, which gives a skeptical researcher something to check the output against. None of it required a researcher to change how they approached the work.

We deprecated the assistants menu in the same release. Researchers had independently adopted NotebookLM and the Claude app for their own analysis work, so we made ALI’s output feed those tools cleanly after seeing how enriched transcripts make a general-purpose model substantially better at research analysis, and are worth more to a researcher than another bespoke agent. We also cut out-of-the-box processing times from roughly thirty minutes to under a minute.

ALI now runs on more than 85% of all eligible research projects at a firm where AI use had been close to zero, among a user population professionally disposed to distrust it. An early impact analysis across the firm’s eligible projects indicated that ALI contributed to a nearly 20% reduction in average time to delivery and to projects running with roughly twice the buffer against their scoped hours, while customer satisfaction stayed high. Researchers stopped downloading from Zoom and re-uploading to client drives entirely.

The hours that had gone to file handling and transcript cleanup went back into analysis and strategy. Researchers who had been the most skeptical of AI became the people best equipped to judge it, and that hands-on experience now shapes how they advise clients on where AI belongs in research and where human judgment still matters. It has pulled AnswerLab into conversations about rethinking research workflows and AI adoption at several of its largest accounts.