$ Prasun Anand
I write frameworks for product teams shipping AI into places where being wrong is asymmetrically expensive, and I build the same products I write about.
02 / Frameworks & Writing
Six named tools for the AI shift — where being wrong is asymmetrically expensive. Each is distilled to its parts; the full argument is in the essay.
03 / Work
Three pieces I'd defend at a whiteboard. Dezerv · 2021–2026.
The biggest trust risk at Dezerv wasn't the UX, it was reconciliation. I rebuilt the pipeline with reconciliation as a first-class object; the models (Gemini, Claude) came in last.
↳ Most of 99.8% was structural. The model just cleaned up the tail.
→Read case studyThe funnel said the feature worked. Behavioral data said otherwise: our highest-value users weren't individuals, they were households, and we'd designed for individuals.
↳ When the funnel and the behavior disagree, trust the behavior.
→Read case studyLed the build with 10+ engineers and 3+ designers, using high-frequency moments, not feature count, as the design constraint. Today ~80% of clients open monthly, ~30% daily.
↳ The first screen earned the daily open. The rest was a search problem.
→Read case study04 / About
Most “AI features” are interface theater. The interesting work is structural: it changes who does what, when humans stay in the loop, and which decisions a model is allowed to commit to.
Five years in wealth-tech at Dezerv, from a 2021 startup through Series C: business analyst, then analytics, growth, and product. The work I've cared about has never been on the surface. It was in the reconciliation pipeline ops patched by hand, the funnel that read healthy while the behavioral data said otherwise, and the boring corner where a model quietly absorbs twenty hours of manual classification.
05 / Contact
Frameworks to argue with, products to pressure-test, or work to do together — the inbox is open.