EXIT INTENT
Exit intent is a signal. Not a strategy.
Exit-intent technology detects that someone is about to leave. It does not detect why they are leaving, whether anything on your site could have changed that, or whether interrupting them will make you money.
Exit-intent software watches for behaviour that predicts departure — cursor movement towards the browser chrome on desktop, rapid upward scroll or back-gesture on mobile — and fires a message before the visitor goes. It is cheap to install, easy to measure, and genuinely effective at list building and last-second capture on low-value baskets.
Its limit is structural, not a matter of quality: a departure predictor is a timing device. It tells you when, and it has nothing to say about why or whether.
CREDIT WHERE IT IS DUE
Four places exit intent genuinely earns its keep.
This is not an argument that exit-intent software is bad. It is an argument that it answers a narrower question than the way it is usually deployed assumes.
01
List building
An email in exchange for something useful, at the moment the visitor was leaving anyway. Low stakes, clearly voluntary, honest trade.
02
Impulse baskets
Under about €50, where the decision was never careful and a small incentive legitimately tips it. The cost of being wrong is a shrug.
03
Genuine deadlines
A sale that really ends, stock that really is limited, a delivery cut-off that really exists. Real urgency communicated once is information, not pressure.
04
Saving progress
“Save this configuration” on a long form is useful at exactly the moment of departure, and it costs the buyer nothing to decline.
Notice what all four have in common: the message does not depend on knowing why the visitor is leaving. They work precisely because they are indifferent to the reason. The trouble starts when the same mechanism is asked to rescue a decision it cannot diagnose.
THE LIMIT
Three questions a departure predictor cannot answer.
QUESTION ONE
Why are they leaving?
Two buyers exit the same page for opposite reasons — one cannot separate two products, the other has discovered they are not eligible. A cursor heading for the tab bar looks identical in both cases.
Consequence: the message has to be generic, which usually means an incentive, because an incentive is the only thing that works without understanding the buyer.
QUESTION TWO
Is this loss recoverable at all?
Some departures have nothing to do with your site: a phone call, a partner who has not replied, a measurement that needs taking, a genuine no on eligibility. Nothing you render changes any of them.
Consequence: a large share of impressions are spent on sessions that were never winnable, and the report cannot tell you which ones.
QUESTION THREE
Would speaking make me more money?
The buyer who was going to return tomorrow and pay full price is inside the same trigger population as the buyer who was genuinely lost. Show both a discount and you have paid one of them to do what they were already doing.
Consequence: without a holdout, margin transfer is invisible. Conversion rate goes up. Profit does not.
THE ANSWER FIRST
When should you not show an exit-intent popup? In six situations, and they cover most of a considered-purchase journey.
01
The blocker is off your site
A second decision-maker, a measurement, a bank appointment. Share-link and traveller-count edits are the tell. Nothing you say resolves it.
02
The visitor was progressing normally
Leaving is not always losing. On a multi-session purchase, a visitor who read three pages and left is on schedule, not in trouble.
03
They already dismissed something
They have told you explicitly. A second surface from a different rule is not persistence, it is not listening — and it costs more than the session is worth.
04
The answer is genuinely no
Not eligible, not in coverage, not available on those dates. Persuasion here wastes both parties' time and buys you a complaint later.
05
Your only message is a discount
If the incentive is the whole idea, you are paying full-price buyers to identify themselves. At €40 that is a rounding error. At €1,800 it is the P&L.
06
You cannot measure it against a holdout
Without a permanently withheld group you cannot separate recovered sales from sales that would have happened anyway. Then every result flatters the popup.
On the considered-purchase journeys we work in, some combination of these six is true for the large majority of exits. That is the whole argument: the trigger is firing correctly and the decision behind it is missing.
THE ALTERNATIVE
Decide earlier, act smaller, and be willing to say nothing.
Exit is late. By the time the cursor moves, the buyer has already concluded. The signals that identify why — comparison loops, term toggles, a fee FAQ, a repeated coverage check — happened minutes earlier, while there was still something to resolve.
A decision layer reads those instead, forms a scored hypothesis about the blocker, and then asks the question exit-intent software never gets to: is intervening here worth more than leaving this person alone? On illustrative pilot traffic the answer is no about nine times in ten.
How AI decisioning works → · Why no action belongs in the ranking →
| Exit-intent popup | Decision layer | |
|---|---|---|
| Fires on | Predicted departure | A scored friction hypothesis, minutes earlier |
| Knows why | No | A named friction with a confidence score |
| Can choose silence | Only by not matching a rule | Yes — ranked, chosen, and recorded |
| Typical response | Overlay with an incentive | A line of information in page flow, or one question |
| Reported as | Impressions, captures, conversion rate | Incremental gross profit vs a permanent holdout |
| Cost of being wrong | Unmeasured | Priced as trust cost, before the decision is made |
PRACTICAL
Keeping your exit-intent tool? Fix these four things this week.
None of these require buying anything. Three of them are configuration, and the fourth is an argument with whoever owns the reporting.
01
Hold out 10% permanently
Randomise inside the triggered population, not across all traffic. Anything else compares people who exited with people who never did.
02
Report margin, not captures
Net out the discount you gave away and the orders that would have happened anyway. Put dismissals on the same screen as the lift.
03
Suppress after one dismissal
For the whole session, across every surface — not just the one that was closed, and not until a scroll resets it.
04
Delete the fake urgency
Invented viewer counts and countdowns that reset on reload. Buyers have learned to read them, and on a large purchase they cost trust.
If the holdout shows your exit-intent programme is producing incremental margin on a considered purchase, keep it and ignore this page. That result would be genuinely interesting, and we would want to see it.
FREQUENTLY ASKED
Exit intent, briefly.
What is exit-intent technology?
Software that predicts imminent departure from behavioural cues — cursor trajectory towards the browser chrome, rapid upward scroll, back-gesture — and fires a message before the visitor leaves.
Do exit-intent popups work?
For list building and low-value baskets, often yes. On considered purchases the reported wins are usually measured without a holdout, which means recovered sales and sales that would have happened anyway are counted together.
Does exit intent work on mobile?
Less reliably. There is no cursor, so implementations infer from scroll velocity, idle time or the back gesture — all of which fire on ordinary reading. The false-positive rate is the reason mobile exit intent annoys more people than it recovers.
What is the alternative to exit-intent popups?
Diagnose the friction earlier from in-session behaviour, resolve it with the smallest response that works — often a line of information in page flow — and decide against acting at all when the evidence is weak.
Do exit-intent popups hurt SEO or Core Web Vitals?
An interstitial that blocks content on entry is a known problem; an exit-triggered overlay generally is not. The bigger risk is layout shift and the script weight of a tag that fires on every session to serve a few.
Can I use exit intent as one input to a decision?
Yes, and that is the right use. It is a legitimate timing signal that raises the urgency of a decision already formed from richer evidence. It just should not be the whole decision.
Find out how many of your exits were never winnable.
Two weeks in shadow mode names the friction behind every stalled session on one journey — and renders nothing to a single visitor.