Funnel Analysis Without Event Tracking or a Tracking Plan

You do not need a tracking plan to see where people leave. Build funnels from the page views, clicks, form submits and API requests you already record, and avoid the pitfalls that make them lie.

You can build a useful funnel without defining a single custom event: use the page views, clicks, form submits and successful API requests you already record, pick the steps from the pages most of your converters actually passed, count them in loose order within one visit, and filter out bots, missing SPA route changes and ids in URLs before you trust the numbers.

Tracking plans are fine in theory. In a small team they go stale the first time someone renames a button, and nobody adds checkout_started to the new page. The data you get for free (URLs, element texts, form posts, request logs) is less tidy but it never stops being sent.

Step 1: Know What You Already Have

Before writing anything, list the raw signals:

  • Page views: from your analytics script, RUM tool or server access logs (GET requests for HTML pages).
  • Clicks: element text or selector, if your analytics or session tool records them.
  • Form submits: the form's action or id and the page it was on.
  • API requests: method, route and status code. A POST /api/orders that answered 201 is a better "order placed" step than any event someone remembered to add.

Any of these can be a step. A good goal is usually a successful write request or a finish page like /order/confirmed.

Step 2: Clean the Page Paths

Raw URLs break funnels in three ways.

Ids in URLs. /products/123 and /products/456 are the same step. Collapse ids before grouping (PostgreSQL):

SELECT session_id,
       ts,
       regexp_replace(
         path,
         '/([0-9]+|[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12})(?=/|$)',
         '/:id',
         'gi'
       ) AS page
FROM page_views
WHERE user_agent !~* '(bot|crawl|spider|headless|lighthouse|pingdom|uptime)';

Bots. Crawlers, uptime monitors and headless browsers visit your home page constantly and never buy, which drags step one up and conversion down. The WHERE clause above removes the obvious ones.

SPA route changes. In a single-page app, moving from /cart to /checkout often does not load a new page, so a script that only records the initial load misses every step after the first. Check that your analytics records pushState navigations. If it does not, this is the whole fix:

let lastPath = location.pathname

function onRoute() {
  if (location.pathname === lastPath) return
  lastPath = location.pathname
  recordPageView(lastPath) // your analytics call
}

for (const method of ['pushState', 'replaceState']) {
  const original = history[method]
  history[method] = function (...args) {
    const result = original.apply(this, args)
    onRoute()
    return result
  }
}
window.addEventListener('popstate', onRoute)

Step 3: Pick Steps From What Converters Did

Do not guess the steps. Ask the data which pages most people who reached the goal passed before it. Assuming a pv view built from the query in Step 2:

WITH goal AS (
  SELECT session_id, min(ts) AS goal_at
  FROM pv
  WHERE page = '/order/confirmed'
  GROUP BY session_id
),
first_seen AS (
  SELECT session_id, page, min(ts) AS ts
  FROM pv
  GROUP BY session_id, page
)
SELECT f.page,
       count(*) AS converters,
       round(100.0 * count(*) / (SELECT count(*) FROM goal), 1) AS pct_of_converters,
       percentile_cont(0.5) WITHIN GROUP (
         ORDER BY extract(epoch FROM g.goal_at - f.ts)
       ) AS median_seconds_before_goal
FROM first_seen f
JOIN goal g USING (session_id)
WHERE f.ts < g.goal_at
GROUP BY f.page
HAVING count(*) >= 0.6 * (SELECT count(*) FROM goal)
ORDER BY median_seconds_before_goal DESC;

Pages most converters passed, sorted from furthest before the goal to closest, are your steps in order. If no single page qualifies but two together do (people reach checkout from /cart or from a "Buy now" on /products/:id), make that step an "any of" step.

Step 4: Choose Loose or Strict Order

Loose order means each step happens after the previous one, with anything in between. Someone who goes Cart, Shipping info page, back to Cart, then Checkout still counts. Use this by default.

Strict order means the next step must be the very next matching event. Use it when you want to know whether people go straight from one step to the next, for example whether the "Continue" button on the cart page is doing its job.

Counting a loose funnel in plain SQL is a chain of "first time after the previous step":

WITH s1 AS (
  SELECT session_id, min(ts) AS t FROM pv WHERE page = '/' GROUP BY session_id
),
s2 AS (
  SELECT pv.session_id, min(pv.ts) AS t
  FROM pv JOIN s1 USING (session_id)
  WHERE pv.page = '/products' AND pv.ts > s1.t
  GROUP BY pv.session_id
),
s3 AS (
  SELECT pv.session_id, min(pv.ts) AS t
  FROM pv JOIN s2 USING (session_id)
  WHERE pv.page = '/cart' AND pv.ts > s2.t
  GROUP BY pv.session_id
)
SELECT (SELECT count(*) FROM s1) AS home,
       (SELECT count(*) FROM s2) AS products,
       (SELECT count(*) FROM s3) AS cart;

On ClickHouse, windowFunnel does it in one pass and returns the deepest step each visit reached (pass 'strict_order' as a second parameter for strict order):

SELECT level, count() AS sessions
FROM (
  SELECT session_id,
         windowFunnel(3600)(ts, page = '/', page = '/products', page = '/cart') AS level
  FROM pv
  GROUP BY session_id
)
GROUP BY level
ORDER BY level;

People at step N are the visits with level >= N.

Step 5: Choose One Visit or One Day

One visit asks "did they do it in one sitting?" and fits checkout and sign-up. One day per visitor fits journeys people pause and resume, like comparing plans in the morning and upgrading after lunch. With a visit window, that person looks like a drop-off twice. Pick the window that matches how your customers actually behave, and keep it fixed when you compare weeks.

Funnels Without Setup in Relia

Relia builds funnels from what its SDK already records, with no event names to define. A step can be a page, a button text, a form, or a request that worked, with up to three "any of" alternatives; order is loose or strict; the window is one visit or one day. Bots are left out and ids in paths are collapsed.

It also finds your main journeys itself. Goals are successful write requests, form submits, and critical or finish pages; steps are the pages most people who reached the goal passed. In a demo shop, that produced a checkout journey of Home 4,928, Products 4,928, Product page 3,707, Cart 2,043, Checkout 1,042 visits (21.1% conversion), with the biggest drop, 1,664 visits, after the product page.

Open a step to see why people left: where they went next, failed requests and JS errors after it, refused form fields, struggle, the page's LCP, segments with a drop rate 1.5 times the step's or more, open incidents, and example visits. A sudden fall of 40% or more against the same hour on previous days becomes an incident. Relia never changes your code or opens pull requests.

Free includes the funnels Relia finds (3 watched). Your own funnels and the "why" are on Growth ($19 a month) and Pro ($49 a month). For the checkout case specifically, see why users drop off at checkout.

FAQ

Is a funnel from page views accurate enough to act on?

For finding where people leave, yes, once bots, SPA navigations and ids are handled. Use a successful API request as the final step when you can, because a confirmation page can be reloaded or skipped.

How many steps should a funnel have?

Three to six. More steps mostly add noise between steps that nobody acts on.

What if converters take very different paths?

Use "any of" steps for the common alternatives, and keep only pages that most converters passed. If no path covers most of them, you probably have two journeys and should build two funnels.

All guides