we let an AI loose on our messiest slack channel and it became our roadmap

when our head of product did a roadmap review, only 3 of the features we'd shipped traced back to anything in the interview library, while most traced to a slack channel called #cs-rage that our CS team had been venting in for years and nobody had thought to mine. we built what i thought was a serious customer interview program at our b2b Saas, with a proper recruitment funnel, paid incentives, a research ops lead, and an annual budget that made our CFO wince. we were doing a steady cadence of hour-long interviews and the synthesis docs were thick and footnoted. we put BuildBetter in front of all of it, i still don't fully understand how it ties messy slack threads to features the eng team can scope, but the proportion has flipped, roughly 70% of recently shipped features now trace back to something the system surfaced on its own. we're still doing the interviews because the depth helps with prioritization, but they're no longer where we discover what to build, and the org-political fallout of that has been the harder thing to work through. the budget i fought hardest for last year is now our lowest-ROI line item in product.