# Likelihood to convert: find the sessions that nearly converted > Cuttl now scores every session High, Medium or Low for likelihood to convert, from your own site’s data. Filter to likely buyers who left and see why. Source: https://cuttl.io/blog/likelihood-to-convert/ · Last updated: 2026-10-29 Cuttl now scores every session High, Medium or Low for likelihood to convert, from your own site’s data. Filter to likely buyers who left and see why. ## What’s new Cuttl now gives every session a likelihood to convert, High, Medium or Low, and an interest, the kind of page the visitor spent most time on. Both work as filters in Journey Insights, so you can open recordings or heatmaps of likely buyers who left without converting. They are live now in early access on every plan, Free included. - Likelihood to convert: a model trained on your site’s own sessions scores each session after it ends. - Interest: the page group a visitor spent most time on, such as pricing, a product range or the docs. - Site settings, Likelihood model: the model’s state, how it did on a week it never trained on, and the signals that move the score most. ## Why it matters Most sites record more sessions than anyone can watch, and a random sample is mostly visitors who were never going to convert. The useful ones are sessions that looked like a conversion and then stopped. A session with a high likelihood that didn’t convert is a visitor who behaved like the people who convert on your site. What stopped them can be a fixable problem, such as an error at checkout, a confusing delivery note or a form that asks too much. Filtering to those sessions points recordings, heatmaps and reports at the people who nearly converted. Interest answers a separate question without building a report: what visitors came for. A segment of visitors interested in pricing pages behaves differently from one reading the docs. ## How it works Each site gets its own model, trained only on that site’s sessions. No data is pooled across sites, and nothing changes on your website. - It learns from what visitors did: pages viewed, time on page, scroll depth, frustration signals (rage, dead and error clicks, quick backs, refreshes), channel, device, forms started and earlier sessions. - It starts once the site has conversions set up as goals and at least 500 converted sessions in the last 28 days. Until then the likelihood filter matches no sessions and the settings tab shows the progress. - Before it is used, the model is graded on the latest week it never trained on. If its ranking score (AUC) is below 0.70, the tab says “Not reliable yet” and the filter stays empty. - Bands are relative to your site: High is its top 20% of sessions by score, Low the bottom half, and Medium the rest. There is no per-visitor percentage. - Sessions are scored after they end (30 minutes with no activity), so it isn’t real time. The model retrains every day. Interest works from the first day. It needs at least 3 seconds on a page group, never counts the home page, and for sessions that converted only looks at pages before the conversion. Journey Insights. See how Cuttl handles likelihood to convert. ## Who it’s for - CRO leads: filter recordings to High likelihood and Didn’t convert, watch ten, and turn the most common blocker into the next A/B test. - UX researchers: compare heatmaps of the checkout for High likelihood sessions that converted and those that didn’t. - Growth marketers: check which channels bring High likelihood visitors who then leave, before moving budget. - Product managers: use interest to see whether visitors who read the docs convert at a different rate from those who go straight to pricing. - Founders: on Free, combine Likelihood High with the Didn’t convert quick filter to pick which sessions to watch first. ## By industry - Online stores: High likelihood shoppers who added to cart and left show whether delivery costs, a coupon field or a payment error stops them. - SaaS: visitors with a pricing interest and High likelihood who didn’t start a trial show which plan details cause doubt. - Lead generation: High likelihood sessions that opened the quote form and stopped show which field makes people quit. - Education: visitors interested in course pages who nearly enrolled show where the enrolment steps lose them. ## How to start in Cuttl In app.cuttl.io, set up at least one goal for the conversion you care about, if you haven’t yet. Then open Site settings, Likelihood model to see whether the model is learning, not reliable yet or ready. When it is ready, the button “Watch high-likelihood sessions that didn’t convert” opens Recordings with that filter set. On any Journey Insights page with the filter bar, choose Add filter, then Likelihood & interest. Both filters are marked Early access, and which plans include them may change later. AI assistants connected through the Cuttl MCP server can use likelihood and interest in the same tools that take a segment, and read the model’s state. The Journey Insights page shows the rest of the product, and the guide to session recordings covers what to look for once you start watching. ## Common questions ### What is likelihood to convert? It is an estimate of how likely a visitor’s session was to end in a conversion, based on how it compares with the sessions that converted before. Cuttl shows it as High, Medium or Low. ### How is purchase propensity calculated? A model learns which behaviours, such as pages viewed, time on page and channel, tend to come before a conversion, then scores new sessions on the same signals. Cuttl trains a separate model for each site on its last 28 days. ### How many conversions does the model need? Cuttl’s model starts once a site has at least 500 converted sessions in the last 28 days, and it is used only if it scores at least 0.70 on a week it never trained on. ### Is likelihood to convert on the free plan? Yes. It is in early access on every plan, Free included, and which plans include it may change later. ### What’s new in Cuttl: active session time, AI setup and 16 other updates ### What is form analytics? Find the field where visitors give up ### What is a holdout group, and why keep one after you ship a test?