What is intent prediction?
Intent prediction is inferring what a visitor is trying to accomplish, and what they are most likely to do next, from the sequence of their behaviour during a live session. It is distinguished from analytics by timing: a prediction that arrives after the visitor has acted is a description, and nothing can be done with it.
Last updated
How does intent prediction work?
By reading behaviour as a sequence rather than as a set of counts. Individually, a scroll and a pause are noise; in order — read the pricing page, opened the FAQ, returned to the plan selector — they describe someone deciding.
The signals are ordinary: pages visited and in what order, scroll depth, dwell time on particular elements, hesitation before a call to action, backtracking, form fields started and abandoned, whether this is a first visit or a return.
What turns them into a prediction is the ordering and the context. The same thirty seconds on a pricing page means one thing after reading a comparison and another as the first act of a session, and no model that receives those signals as an unordered bag can tell the difference.
What can actually be predicted?
The decisive moves: converting, engaging with a form, clicking a primary call to action, or leaving. Vaguer questions — what someone will buy in a month, whether they liked the page — are not answerable from a session and should not be claimed.
Restricting the scoreboard to decisive moves is also what keeps the accuracy figure honest. A system scored on whether it correctly predicted that a visitor would keep scrolling is being graded on non-events, and will look far more accurate than it is.
A prediction is worth something only if it beats the base rate. Knowing that 2% of visitors convert is not a prediction; identifying which 2% is.
How is intent prediction different from lead scoring?
Lead scoring ranks known people using history and attributes, usually after the fact and for a sales team. Intent prediction judges an anonymous session as it happens, so something can be done during that session.
The two answer different questions and often coexist: one decides who a salesperson calls tomorrow, the other decides what this visitor sees in four seconds. Only the second requires the inference to be live, and that requirement is what shapes everything about how such a system has to be built.
See it on your own site
Paste one tag. Cromanion crawls your site, watches real sessions in Learn mode, and only acts when you switch it on — with a permanent 10% holdout proving what it caused. Free to start, no credit card: a 14-day or 1,000-session live trial, then keep measuring for free.