Root-Cause Analysis: New AI Platform Accounts Not Converting to Production Workloads

You own an AI platform. New sign-ups have been rising for a while, yet production usage over the same stretch has not grown. How would you work out why? ```hint Look between the two numbers Production usage sits several steps after account creation. Lay out what a new account has to do before it sends production traffic, then compare where recent cohorts stop with where earlier cohorts stopped. ``` ### Clarifying Questions - What counts as production usage: live traffic from deployed applications (for example, calls on production keys or paid projects), or any activity after sign-up, such as trying prompts in a playground? - Is the platform built for developers calling an API, for business users working in a no-code interface, or for both? - Is the flat line the number of accounts with production usage or the total production volume, and for how long have sign-ups and production usage diverged? - Did anything change around the time sign-ups started climbing, such as a campaign, a launch announcement, free credits, a new sign-up path or a pricing change? ### What a Strong Answer Covers - A precise definition of production usage, and whether the flat metric counts accounts or volume. - Checks that the gap is real rather than a tracking, attribution or definition artifact, with cohorts compared at equal age. - A breakdown of production usage into new accounts converting, existing accounts growing or shrinking, and accounts lost. - The path from sign-up to sustained production traffic, segmented by acquisition source, customer type and use case to find where and for whom it breaks. - Hypotheses checked in a sensible order, a most likely cause, and the fix and metric that would confirm it. ### Follow-up Questions - You find that nearly all the extra sign-ups came from a free-credits promotion. Is flat production usage still a problem, and what would you change? - New production accounts are growing, but production volume per account is falling. What does that suggest, and where do you look next? - How long after sign-up should you wait before judging whether a cohort reached production, and how would you choose that window?