Stop chasing traffic. Close the visitors already there.

Cromanion reads every live session, spots the visitors already close to buying, and steps in at the exact moment it matters.

NORTHFIELD

PlansJournalAccount

Subscription

Coffee that arrives the week you run out.

Roasted to order on Mondays. We learn your pace after two deliveries and adjust the schedule on our own.

$24/ month · ships free

Start my plan

Compare roasts

Pause or cancel any time — no contract, no setup fee.

You’ve read the plan details twice. That’s usually the question left.

Northfield’s page. The blue block is what Cromanion wrote into it.

Live session

00:0000:4101:5803:2204:1204:41

ExposedConverted

Intent score

0%18%34%62%81%96%

100% = certain
this visitor buys

What it read

00:41Landed from Google — “coffee subscription cancel any time”

01:58Read the plan details twice, never scrolled past them

03:22Hovered “Start my plan”, did not click — twice

04:05Went looking for a cancellation page

04:41Read the answer, then clicked “Start my plan”

Decision

Commitment objection. Answer it inline, before exit-intent.

Converted — 29 seconds after the nudge.

cromanion_assisted_conversion → your analytics

Left: a demo brand’s page. Right: Cromanion’s live read of the session, on its own surface. Illustrative example — your own site produces this panel with your own numbers.

Cromanion reads a live session and knows, before the visitor does, which way it goes.

better than chance when it calls a sale coming

vs chance

6.4×

of the time it ranks the buyer above the browser

across 4,100 sessions

91%

more conversions than the visitors it left alone

vs control group

+13%

Illustrative figures, with a control group running throughout — the shape of the result, not a promise of its size. Your own numbers land in your analytics from week one, whichever way they come out. How the measurement works →

How it works

One tag. Three things it does on its own.

It learns your site by itself.

Paste one tag — under 20 kB. The agent crawls your funnel, works out what counts as a conversion, and pulls your palette, type and voice so anything it shows looks like you built it.

It watches before it acts.

Every new site starts in Learn mode: it reads sessions, predicts what happens next, and grades itself — while showing your visitors nothing. You flip it to Act when the predictions convince you.

It acts, then proves it.

At most one nudge per page, in your brand, and doing nothing is the default. A permanent control group — 10% by default, yours to set — measures what the nudges were actually worth, in your own analytics, not ours.

How it knows

Cromanion knows live who’s about to buy.

Visitors who convert behave nothing like visitors who leave. The agent reads that difference live, on signals it worked out from your own traffic, and scores every session as it happens.

Nine times out of ten, the visitor who ends up buying is scoring higher than the one who leaves — minutes before either of them does anything about it. When the agent calls a sale coming, it is right several times more often than a coin toss would be. That part has to work before any nudge is worth showing.

Signals it reads live

Example data

bouncerconverter

Live intent scoreAUC 0.91

bouncers 29%converters 71%

Started filling the formAUC 0.89

bouncers 11%converters 88%

Focused a form fieldAUC 0.88

bouncers 13%converters 90%

Time on the CTAAUC 0.82

bouncers 26.1sconverters 71.4s

4,100

Sessions scored

218

Converters

5.3%

Base rate

AUC — how often a converter outscores a bouncer on this signal. 0.50 is a coin flip; 0.99 is near-perfect separation.

Where it beats chance

Chance = always betting on the most common outcome. Everything past 1× is skill.

convert6.4×

engage_form3.1×

click_cta1.5×

exit1.2×

Illustrative example — your own site produces this panel with your own numbers.

The proof

And Cromanion publishes the holdout, line by line.

A share of your visitors never sees an intervention — 10% by default, and yours to change. Every week the agent compares them to the rest, and writes both sides into the analytics tool you already open.

Seeing the moment is one thing; being worth acting on is another. The panel beside this puts the exposed group against the control and says which engagement signals actually separated — and which did not.

Two of the four separated; two did not, and the panel says so rather than hiding them. That is the shape of an honest read: some of it lands, some of it is still noise, and you see both.

GA4PlausibleMatomoFathomSegment

cromanion_decisioncromanion_action_showncromanion_action_clickcromanion_assisted_conversion

Every one of them also carries a cromanion_holdout field, so filtering on it rebuilds the exposed-vs-control comparison yourself, natively, in your own tool.

Start free

Your own version of this panel, from week one.

Exposed vs control

Example data

+13% lift

1,996 sessions measured · control group is 20% of traffic

Exposed

12.4%198 / 1,596

Control · holdout

11.0%44 / 400

Engagement

Separated on 2 of 4

1,996 sessions · share of each group above the median

SignalExp.Ctrl.Verdict

Stopped to read55.1%44.8%p .004

Reading depth52.6%46.1%p .038

Active time51.4%48.2%p .210

Saw more than one page30.2%28.6%p .580

Illustrative example — your own site produces this panel with your own numbers.

Control group

yours to set, any time

10%20%50%of your traffic never sees an intervention

Exposed — the agent may act

↑ 10% default

Starts at 10% — enough to measure, cheap to give up. Raise it while you are still deciding whether to trust the agent; lower it once the number stops surprising you. It never goes to zero.

Your forms live in HubSpot?

A HubSpot embedded form submits inside an iframe, so an ordinary tag never sees it. Cromanion does — and surfaces that form as a conversion goal for you to validate, because what counts as a conversion is your definition, never ours. Leads Cromanion captures itself can go straight into your contacts. How the HubSpot connection works →

Instead of A/B testing

Cromanion writes the page this visitor needs, while they’re still on it.

An A/B test splits your traffic between two fixed versions somebody had to build, waits weeks for significance, then ships one winner to everybody — including all the people it never suited.

Cromanion doesn’t author variants at all. It reads the session in front of it and composes the version that fits that visitor, out of what it has already learned about who converts on your site. There is no winner to pick, because there is no single version to win.

The control group stays — not to choose between two pages, but to tell you what the whole approach is worth.

The A/B way

Author two fixed versionsSplit the trafficWait for significanceShip one winner to everyoneStart over next quarter

The winner is the average visitor. Nobody is the average visitor.

With Cromanion

No versions to authorIt reads this sessionIt composes this versionNext visitor, different pageOr no page change at allThe control group prices it

One version per visitor, not one per quarter.

Rewrites the copyReorders what’s thereBuilds what isn’tFor one visitor at a time

Live session

Reading this sessionRewrote the headlineMoved the price under the buttonsAdded the reassuranceExample data

NORTHFIELD

Welcome to Northfield Coffee Co.

Coffee that arrives the week you run out.

Roasted to order on Mondays. We learn your pace after two deliveries.

$24/ month · ships free

Start my plan

Compare roasts

$24/ month · cancel any time

Skip any monthMonth to month, always

Cancel onlineNo call, no form

No contractOne tap, no email

Three kinds of edit, all reversible, none of them written into your code. The rewrite is applied to the visitor as it happens. Moving a block or adding one can break a layout the agent cannot see, so those two are grounded, logged and drawn on your own page for you to approve — the agent proposes, you decide. This visitor was worried about commitment, so this is the page they got.

What it changes

Six surfaces, and Cromanion picks the one that fits the moment.

Nothing here is a template you fill in. The agent chooses the surface from the session it just read, writes the copy from your own content, and reverses it the moment the visitor moves on.

Inline answer

Rewrites the paragraph the visitor stalled on, in place. No overlay, no interruption.

Tooltip

Answers one objection where it was raised — beside the price, the field, the term.

Modal

For the objection worth the whole screen: a guarantee, a comparison, a missing step. Exit-intent only, and default-off.

Bottom sheet

Mobile, thumb-height. Rises when the session says the visitor is about to leave.

Sticky bar

A single line that follows the scroll — shipping, returns, the deadline that matters.

Section swap

Reorders or replaces a block so the thing this visitor needs comes first. Moving a block can break a layout the agent can't see, so it arrives as a proposal on your own page and you approve it.

Guardrails

Cromanion only ever says what your own content can back up.

No fabrication.

Every fact it shows — a review, a count, a price — must trace to a real reference in your content, or to something you told it yourself. No reference, no intervention.