Before the Demo: An AI Market Research Vendor Track Record

An AI market research vendor track record should be something you can check, not something you take on faith. Ours is: two companies’ operating history, real studies with real numbers attached, and one candid account of what changed after we got something wrong.

CodexMR runs on the combined operating history of two companies: Bright MR and Kalever, both based in Sofia, Bulgaria. One ran research operations, the other built platform engineering, years before CodexMR existed as a single product. It’s the actual practice the platform came out of. Every figure below is pulled from real projects run inside it, not a benchmark built to look good in a deck.

What We’ve Shipped

Start with ResearchReady, the newest tool in the stack. Before it launched publicly, we ran it retrospectively across 30 completed studies at a mid-size full-service agency. The test: would it have caught what only surfaced once fieldwork started? On one project, the client’s own length-of-interview estimate was 15 minutes. ResearchReady flagged 10. Fieldwork came in at 9. On another, the client estimated 15, ResearchReady flagged 21, fieldwork landed at 20. Every risk the tool surfaced before programming was later confirmed by what fieldwork produced. The agency kept it as a standing step in their process instead of quietly dropping it once the trial ended.

Since teams started running it that way, the pattern has held. Pre-fielding review effort dropped roughly 70%. Questionnaire finalization sped up 30%. Fieldwork timelines came in about 30% shorter.

The platform side tells a related story. One client running full quant studies through CodexMR reached 60% faster operational readiness. Programming effort in DIY mode dropped 90%, a number high enough that we checked it twice before we felt comfortable printing it here. Across the wider practice, survey programming through the platform runs up to 80% faster than a manual build. Projects that move through the full workflow have taken the average study timeline from 12.5 weeks down to 10.5.

There’s a migration story behind those numbers too. Bright MR’s survey transfer service, built on the same technology, moves a client’s existing questionnaire code onto a new platform in hours instead of the days a manual rebuild takes, across Qualtrics, Forsta, Decipher, UNICOM, and more. None of this is a lab result. It’s what happened when real deadlines, real client edits, and real fieldwork got involved.

What Changed Because We Got Something Wrong

Early engagements asked the client to adapt to us. A study spec arrived in whatever format the client’s team already used. Someone on our side converted it into the structure our systems could read before anything moved forward. That worked, but the conversion step sat entirely on our side of the relationship in name only. In practice, a client’s own shorthand, their routing conventions, and their template quirks all had to get relearned in our terms. Only then could we act on any of it. Slower project starts and more back-and-forth were the direct results.

We reversed it. The platform now reads what a client already has: a Word or Excel questionnaire written in their own convention. No reformatting. No house template required first. The adaptation cost moved from the client’s side to ours, which is where it belonged from the start. That shift is also why a new project starts with a quote and a straight run into build. No back-and-forth over file formats before anyone can even scope the work.

What an AI Market Research Vendor Track Record Should Show You

A demo shows a platform working under conditions the vendor controls. A track record shows what happened when the conditions weren’t controlled: real deadlines, mid-project scope changes, edge cases nobody scripted for in advance. Ask any AI market research vendor for that second kind of evidence. The quality of the answer will tell you more than an hour of screen-sharing.

We’d rather you ask us for specifics before you shortlist anyone, including us. Ask what’s gone wrong, and what changed because of it. A shrug, or a redirect back to features, is worth noting before you sign anything. That’s the difference between a logo slide and an AI market research vendor track record you can verify.

This is close to what ESOMAR’s due diligence checklist asks buyers to verify before trusting any AI vendor’s claims. Real deployments. Honest lessons. Credentials that hold up past the pitch. We’ve published our full answer to all 20 of those questions, section by section, in the second piece of this series.

This is the sixth article in the Before the Demo series:

Readers also enjoy: