A/B testing and feature flags you own forever.
Agencies pay per-seat and per-client fees that scale with their success. Self-hosted, pay-once tools replace per-seat pricing with a flat cost that margins actually survive.
The tool this job usually means renting is VWO, at $199/mo — roughly $2,388 a year, every year. Splitpoint does the job for a one-time $49 — running on infrastructure you control.
Control plus N variants, weighted traffic split, holdout percentage, and URL/device targeting.
For agencies, experiments pays off across every client account, with no per-seat math — and it's covered by the same one-time $49, not metered per month.
A JS track() call, or automatic conversion when a visitor hits a matching pageview URL.
For agencies, goals two ways pays off across every client account, with no per-seat math — and it's covered by the same one-time $49, not metered per month.
Vanilla JS under 3KB, no dependencies, never touches the DOM; assignments stick via deterministic hashing and localStorage so there is no flicker.
For agencies, tiny SDK pays off across every client account, with no per-seat math — and it's covered by the same one-time $49, not metered per month.
One payment of $49 gets you the packaged Windows installer with updates, or clone the MIT source at github.com/bensblueprints/ab-testing-tool-mvp and run it free.
Run it as a Windows desktop app, or docker compose onto a $5 VPS when you need it public.
Add the sp.js tag, call getVariant to branch your UI, and track() on conversion.
Splitpoint is $49, once. VWO at $199/mo — roughly $2,388 a year, every year. Less than one month of VWO covers Splitpoint forever — and it never tiers you by traffic.
Less than one month of VWO covers Splitpoint forever — and it never tiers you by traffic. Splitpoint costs $49 once; VWO charges $199/mo — about $2388 a year. For agencies, that fixed cost is the whole point: the tool is a one-time line item, not a monthly one.
Yes — Splitpoint is MIT-licensed at github.com/bensblueprints/ab-testing-tool-mvp. The $49 buys the packaged Windows installer and updates; building from source is free.
It runs a pooled two-proportion z-test and reports the two-tailed p-value, confidence and Wald 95% CI. Standard caveats apply: fix your sample size up front, do not peek-and-stop the moment a result looks significant, and remember 1 in 20 winners at 95% is noise.