AI funnel analysis: where signups drop off, and who dropped

A funnel tells you which step loses people. An AI agent can tell you why it might be, which form fields failed, and who you can still reach. Here is how funnel analysis works in Behavly, with goals that count on their own and an agent that reads the funnel for you.

Funnel analysis follows people through the steps toward something that matters, such as a signup, and shows where they drop off. AI funnel analysis adds an agent that reads the funnel for you, says which step loses the most people and which form fields failed, and points you to the people you can still reach.

In Behavly, the funnel and the agent read the same data from the same snippet, so you can build a funnel yourself in Insights or simply ask.

Ask the agent first

Ask Behavly Agent “where do signups drop off this week?” and it reads the signup funnel step by step:

  1. 1Visitors: everyone who came to the site in the period.
  2. 2Engaged: the visitors who clicked, scrolled or opened a second page, rather than just looking and leaving.
  3. 3Saw signup: the ones who reached the signup page.
  4. 4Started the form: the ones who began filling it in.
  5. 5Submitted: the ones who sent it.
  6. 6Newly identified: the ones who became signed-in users.

With it come the pages people left from without engaging, how many abandoned the form, and which form fields failed and how often. A big drop between Started the form and Submitted, with one field failing again and again, points at the form, not at your traffic.

Build your own funnel in Insights

For any other path, such as pricing to signup to first project, build a funnel from goals. Several goals count on their own, with nothing to set up: form submits, outbound link clicks, file downloads, and email and phone link clicks. You can add three more kinds:

Pageview goal
A visit to a page, such as /signup. Use * as a wildcard: /blog* matches every page under /blog.
Scroll goal
Scrolling a page to a depth, such as 50% of /pricing.
Custom event goal
An event your site sends, such as behavly.track("project_created").
  1. 1In Insights → Site activity, open the ⋮ menu on the Goals card and choose Manage goals.
  2. 2Choose Add goal for each step you need.
  3. 3Choose Add funnel and pick two to eight goals, in order.

Goals apply to your past data as soon as you add them, so a new funnel shows results right away. A funnel counts a visit at a step only if it completed every step before it, in order, in the same visit. Explore shows the paths people actually take: pick a starting page to see where they went next.

8

goals a Behavly funnel can follow in order, from two up, counted only when every step before was completed in the same visit.

From the step that leaks to the reason why

A funnel tells you where. To see why, open the step that loses the most people in Heatmaps. The map is painted on your live page, with clicks, rage clicks, movement and scroll. Or ask Behavly Agent what frustrates people on that page: it reads where people rage-clicked, which clicks did nothing and how far down they scrolled. More in Rage clicks and dead clicks.

From the rate to the people

Who you can reach depends on where they dropped:

  • Before signing up, visitors are anonymous. There is no one to email, so the fix is the page or the form. That is what the heatmap and the failed fields are for.
  • After signing up, people are identified, one row per person across devices. The ones who stopped at a later step, such as never creating a first project, are people you can follow up with.

Ask Behavly Agent for them in plain words, “signed up this week but never created a project”, and it answers with a pattern: the rules, the exact count and a sample, before anything is saved. Save it, and a campaign can write each of them an email from what they did. See Find user behavior patterns with AI.

Read the date range honestly

On a busy site, a long range is a sample, not a census. For anything you will make a decision on, prefer a shorter, complete range, and compare two equal short ranges rather than reading one long one. Ask the agent the same way: “this week against last week” beats “the last six months”.

The short version

  1. 1Ask Behavly Agent where signups drop off.
  2. 2Open the leaking step’s heatmap and fix what people fight with.
  3. 3Build a funnel in Insights for the steps after signup.
  4. 4Ask for the people who stopped after signing in, and follow up with each of them.

Funnels are one piece of agentic analytics. For the rest of what you can ask about your site, see AI website analytics.