Unfair advantages for your CRO.

State of the art Machine Learning running 24/7 for you at agentic scale.

What RoutingCat does

Every variant stays live, each visitor gets the one most likely to earn the most, and one number says what that added.

Adaptive personalized routing
RoutingCat adapts to your content and visitors in real time, optimizing itself in the background. Traffic gets progressively routed to where it earns you the most. Add or remove variants at any time: routing is always ON.
Routing is a contextual bandit: each variant gets traffic in line with its chance of being the best, for each kind of visitor, by device, source, ad, first or repeat visit and time of day.
Ask your agent
Try three headlines on the home page.
User guide: routing and A/B testing
Ad-aware routing
Each ad brings a different visitor. Group your campaigns into audiences, and RoutingCat learns which variant works for each one.
It tells Meta, Google, TikTok, Microsoft, LinkedIn and Pinterest ads apart on its own, and reads your campaign tags for the rest.
Ask your agent
Our ad sets are named by audience. Let the hero learn which variant works for each one.
User guide: personalization and audiences
Optimize profit and conversions
Define the real value to optimize: profit per visit, LTV, custom weights, or plain conversions. RoutingCat solves the tradeoff for you and routes visitors to maximize your actual bottom line. Crucial for your offer.
Ask your agent
Optimize the product page for profit, not orders.
User guide: conversions and what they are worth
Multi-slot experiments
Stop A/B testing every detail separately: you can optimize multiple slots (Hero, visuals, etc) at the same time, RoutingCat handles everything together efficiently.
Every slot of a page learns from the same visits, at the same time, so seven questions take the time of one.
Ask your agent
Add slots for the hero photo and the call to action on the pricing page, three variants each.
User guide: many slots at once
A measured gain
One visitor in ten always sees your current site. The difference between them and everyone else is what routing added: measured, not modeled, with the range the evidence supports.
Every figure says what kind it is. Counted is what happened. Predicted is the model’s belief, with a range. Measured comes from a random split, and is the only number that counts as a gain.
Ask your agent
How much has routing added on the home page, and how sure are we?
User guide: results and the measured gain
Analyze conversions and attribution
Get insights into your business and each variant with comprehensive analytics and attribution. Learn what works, for whom, and which channels bring you the most qualified traffic.
Every conversion is credited to the visit that led to it, then counted by channel, source, campaign and landing page: where your conversions come from, not a guess at what caused them. Stripe and Whop payments are credited with no webhook code.
Ask your agent
Where did last month’s sign-ups come from, and which campaigns brought the best ones?
User guide: run it every week

Agentic infrastructure

Winning variants are one prompt away. Your agents are capable of creating, analyzing and iterating through thousands of items in parallel while you focus on your product.

› Try two prices on the product page: $57 a candle,
  and three for two. Optimize for profit, not orders.

◆ Skill  routingcat
● Edit  routingcat/pdp.ts   +3
  + objective: profit
  + prices:    single-57
  +            three-for-two
● Bash  npx routingcat plan
    valid  push will accept this
● Bash  vercel --prod
    routingcat  LIVE pdp: 3 variants in prices, routing within about 2 minutes

Both prices are live. Give it a week.

                            — one week later —

› How are the new prices doing?

● Bash  npx routingcat status pdp
    gain    +$1,870 profit [−7,240, +10,980] · measured
      single-57      favored   +5% [−9, +19] vs default
      three-for-two  learning  −16% [−31, −1] vs default

$57 leads, but its range still spans zero. Three for two
is losing clearly, its whole range below zero. The gain
is too early to tell.
Want me to retire three-for-two?

npx routingcat init installs a skill that teaches your agent the whole workflow. It works with Claude Code, Codex, Cursor, Gemini CLI and GitHub Copilot. Every version of every variant is kept with its results, so each session starts from what was tried.

Ask your agent

Run the home page’s experiments, and check in every Monday.

User guide: work with your agent

Also built in

The controls, the connections and the guarantees around it.

Rules and beliefs
Show a variant only to some visitors, or never to others. Or say where you expect one to do better: it starts with more of those visits, and the evidence has the last word.
Ask your agent
I think the short hero works best on phones. Start from that, and tell me if I’m wrong.
User guide: rules and beliefs
Any shape of variant
A headline, an image, a price, the order of the sections, a whole page: a variant is anything your code can render, not only what a visual editor can reach.
- A value: a headline, a price
- An image
- A section
- The order of the sections
- The whole page
Ask your agent
Try the testimonials before the plans on the pricing page.
User guide: what a variant can be
Change what is live
A new variant ships with your deploy. Pausing, retiring, promoting or reverting one does not: it is live within about two minutes, from the command line or the dashboard, and can be previewed first.
Ask your agent
Pause the new hero until I’ve fixed its image.
User guide: change what is live
Classic A/B tests
When you want a fixed split, it is one setting away. Give a slot a fixed share and its visitors are split evenly, each keeping the same variant on every visit.
Ask your agent
Test $24 against $19 for Pro as a clean A/B test.
User guide: when a fixed split is the better tool
Fast, up and private
Your site comes first: it stays fast, it stays up, and your visitors stay anonymous.
- Fast. On Next.js, Nuxt and Astro, variants are picked on your own server before the page renders; elsewhere, by a small script before it paints. The browser code is about 11 KB.
- Up. If RoutingCat cannot be reached, routing goes on with what it last knew, or the page shows its defaults.
- Private. No raw IP addresses, no names or emails, no tracking across sites, first-party cookies only.
User guide: speed, failure, privacy