
4 August 2026
A/B testing was built for a web that no longer exists
A/B testing is one of the most successful ideas in web development. It replaced opinion with...
Read more →Writing on adaptive UI, multi-armed bandits, and conversion optimisation. SentientUI learns what each visitor persona wants and shows them the interface most likely to convert — visit 1 learns, visit 2 converts. These posts are the reasoning behind that: the algorithms we chose, the ones we rejected, and what we got wrong on the way.
Most of it is written for the person who has to build and defend the thing — engineers shipping personalization into a real codebase, and the growth folk who have to justify it afterwards.

4 August 2026
A/B testing is one of the most successful ideas in web development. It replaced opinion with...
Read more →2 July 2026
There is an assumption buried so deep in how we build software that it rarely gets questioned. The...
Read more →Four themes come up again and again. If you only read one post, start with whichever of these is the argument you are currently losing.
Why waiting for statistical significance is the wrong tool for a site with modest traffic, and what Thompson Sampling does instead. We run a Beta posterior per variant per segment, so traffic keeps flowing to what is working while the tail is still being explored — no experiment queue, no analyst called in to declare a winner.
How a pseudonymous visit becomes a persona without fingerprinting or cross-site tracking. Posts here cover segment derivation, cold-start seeding when a new variant has no history, and why the layout policy is keyed on the persona while the component bandit stays keyed on the stable device-and-source segment.
Most personalization tools flash the default, then swap it client-side — which costs you layout shift and Core Web Vitals. We write about resolving the variant on the server so the right interface is in the initial HTML, what that costs in latency, and how the timeout falls back to default order cleanly.
Wilson intervals, two-proportion tests, and the difference between a real lift and a small sample. Conversion work is full of numbers that look decisive and are not, so posts here are as much about which results to distrust as about which to ship.
The live demo adapts in front of you, the docs cover the one-line snippet and the React SDK, and integrations walks through Webflow, WordPress, and Framer. Pricing is visitor-based, with a free plan to start.