experiment across your whole stack
AI growth experiments across your whole stack.
AI growth experiments: Cogny turns your connected data into a hypothesis and one concrete change, ships it in the tool after you approve, and measures the result in your own data, 3 days later for ads and 7 for content and SEO. Each experiment closes as worked, partial, didn't work or inconclusive. It is a before-and-after reading, not a traffic split.
or from your own AI
claude mcp add --transport http cogny https://app.cogny.com/mcpHow can I run marketing tests across ads, landing pages and email and know which ones won?
Run one change at a time per channel and judge it in the data where the result shows up: ad metrics for ads, GA4 conversions for a landing page, opens and clicks for email. Cogny ships the change after you approve and reads the before and after 3 days (ads) or 7 days (content) later. It does not split traffic; for true A/B tests, keep GrowthBook, whose experiment results Cogny can read.
Does Cogny split traffic like an A/B testing tool?
No. Cogny changes the real thing, a budget, a page title, an email segment, and compares the period after the change with the period before it in your own connected data. It doesn't randomise visitors or compute statistical significance. When the signal is too weak to call after four checks, the change closes as inconclusive instead of being called a win.
Can Cogny read my GrowthBook experiments?
Yes. Connect GrowthBook and Cogny can list your experiments, feature flags and metrics and read each experiment's latest computed results. Your AI can then put a split test's outcome next to what moved in GA4, the ad platforms or Search Console. Cogny reads GrowthBook; it doesn't create experiments or change flags there.
what AI growth experiments are, and how cogny measures them
An AI growth experiment is one specific change, proposed from your data, shipped in a real tool and judged against what happened next. The AI part isn't a smarter hypothesis spreadsheet. It's an agent with read and write access to your stack that does the research, makes the change once you say yes, and comes back to check.
Cogny connects to 72 integrations: ad platforms, analytics, search, email, e-commerce, CRM and the CMS your site runs on. That is what makes the experiments cross-stack. A change in one tool is measured in the data of another: a Google Ads budget read against GA4 purchases, a rewritten title against Search Console clicks, a Klaviyo segment against Shopify orders.
Every change you approve in the Cogny app becomes a ticket; from your own AI, launches become tickets automatically and other changes your AI offers to register. Cogny measures it after three days for ads and seven days for content and SEO, comparing the period after the change with the period before it in your own connected data. Then it closes the ticket with exactly one of four verdicts: worked, partial, didn't work or inconclusive. If four checks still can't tell, it says so and closes it as inconclusive rather than guessing.
That is a before-and-after reading, not a randomised A/B test. Cogny doesn't split traffic and doesn't compute statistical significance. If you already run split tests in GrowthBook, Cogny reads their results; for everything else, the verdict says what moved and how sure it can be.
One change in one tool, measured in the data of another.
why most experiment programmes stall
The backlog outgrows the team
Fifty ideas, two tests a month. Writing the hypothesis, making the change and checking the result each cost someone an afternoon.
Ideas come from opinion
Without a quick read of your own data, the loudest idea gets tested first and the cheapest win waits.
Channels are tested in isolation
Ads are tested in the ad platform and pages in the CMS, so nobody sees what a budget change did to sign-ups or orders.
Results don't compound
Outcomes live in slides and Slack threads. Six months later nobody remembers what worked, and the misses get run again.
from raw data to a verdict
The same loop on every plan, Free included. Minutes from question to shipped change, not a sprint.
- 01
Ask
Start from a question or a graph. Cogny reads the connected data and what already shipped.
- 02
One next step
It proposes one concrete change, the number it should move and where that number will be read.
- 03
Approve
In the Cogny app, write tools ask first. From your own AI, its approval prompt is the gate.
- 04
Ship
The change is made in the real tool: Google Ads, Meta, WordPress, Webflow, Shopify, Klaviyo and more.
- 05
Measure
After 3 days for ads, 7 for content and SEO, before is compared with after in your data.
- 06
Verdict
Worked, partial, didn't work or inconclusive. The next experiment starts from the result.
one change, measured somewhere else
This is what the ledger looks like when experiments cross tools: the column that matters is where the result was read.
Example rows, not customer data.
what you get
Experiments proposed from your data
Ask in chat, or on Cloud press Experiments on any graph and get three aimed at what moved it, put straight on the ticket board.
A ticket per change
Every write is logged with the tool, the time and who authorised it. From your own AI, launches become tickets on their own and other changes are one yes away.
Measured in your own numbers
GA4, Search Console, the ad platforms, your shop. Budget, bid, status and targeting changes are measured too, not only new objects.
Misses on the record
A change that didn't work closes as didn't work, so nobody repeats it by accident.
Loops for what works
A finished chat can become a Loop that runs on a schedule on Solo, Cogny AI and Cogny Pro. Each run is its own measured ticket.
A card for every win
A measured win gets a share card you can copy as an image or text and send to whoever asked.
where cross-stack experiments pay off
Landing pages for paid traffic
Change the page in your CMS, read the result in the conversion rate of the paid traffic that lands on it. Cogny proposes the headline, offer or layout change, ships it on approval, and reads GA4 and the ad platform seven days later.
Funnel friction
Cogny reads where people drop between visit, sign-up and purchase, then proposes one change at the step that loses the most. Each change is its own ticket, so the funnel's history becomes a list of verdicts instead of a slide.
Budget and bidding
A reallocation in Google Ads or Meta is read three days later against purchases or leads in GA4, not against the platform's own conversions. A cheaper click that brings no customers shows up as a miss.
what it doesn't do
- It doesn't split traffic or run randomised A/B tests. Verdicts compare before with after in your own data.
- It doesn't calculate statistical significance. When the signal is too weak to call, the verdict is inconclusive.
- It doesn't change your accounts without an approval: in the Cogny app write tools ask first, and from your own AI its approval prompt is the gate.
- It doesn't promise a number of experiments a month. You run as many as you approve.
the record.
Published case studies, each one linked.
questions
Is this A/B testing?
No. Cogny measures before and after in your connected data and closes each change with a verdict. It doesn't split traffic or run significance tests. If you run A/B tests in GrowthBook, Cogny can read their results alongside everything else.
How long until an experiment gets a verdict?
The first check runs three days after an ads change and seven days after a content or SEO change. If the signal isn't clear yet, Cogny checks again; after four checks without a clear answer it closes the ticket as inconclusive.
Which tools can an experiment touch?
Any of the 72 integrations with write tools: ad platforms such as Google Ads, Meta, LinkedIn and TikTok, CMSs such as WordPress, Webflow and Sanity, Shopify, and email tools such as Klaviyo and Mailchimp. Results are read from whichever connected source holds the outcome.
Can I run experiments from Claude or ChatGPT?
Yes. Add Cogny as an MCP server in Claude, ChatGPT, Cursor or Codex. Your AI does the work; Cogny logs every write. Launches, such as a new campaign or ad, become measured tickets automatically, and your AI offers to register other changes so they're measured too. Free includes 15 MCP calls a month; Solo is unmetered under fair use.
Do misses really stay on the record?
Yes. A ticket that didn't work closes as didn't work, in the same list as the wins. Our own published truth ledger lists results that had no baseline to claim anything from.
what it costs.
- Free $015 MCP calls a month and 50 one-time chat credits, no card
- Solo $9unmetered in your own AI, fair use
- Cogny AI $18150 credits
- Cogny Pro $901,500 credits
- Cloud $4995,000 credits, for teams
Per month. Solo, Cogny AI and Cogny Pro have one seat each. Every change is recorded on every plan; Solo measures 3 a month, Cogny AI and Pro measure every one and add monthly credits. Teams need Cloud. Compare plans →
every experiment gets a verdict.