Try it on your own data
Every report in this article is in the free plan. One snippet for web and apps, up to 10 sites.
Start freeConversion funnel optimization usually starts with tactics and ends with an argument about whether they worked. This guide puts the steps in the order that avoids the argument: measure the funnel correctly (five common mistakes), find the step that matters, size what fixing it is worth, find out why people leave, then change one thing and verify it with the same definition.
What you will learn
- Five measurement mistakes to fix before optimizing anything
- Equal relative gains add the same orders at any step; effort decides where to start
- Verify a fix with the same funnel, window and filters, and watch the next step
What changed · Nov 3, 2026
Rewritten as a funnel optimization guide: finding the step that matters, a worked gain table, common mistakes and an FAQ.
Conversion funnel optimization starts with a funnel you can trustStep 1: avoid the five measurement mistakesStep 2: find the step that mattersStep 3: size the gain before you fix anythingStep 4: find out why people leaveStep 5: fix one step and verifyConversion funnel metrics to trackFAQ
Conversion funnel optimization starts with a funnel you can trust
New section
Conversion funnel optimization is the work of finding the step where most people give up and making that step easier, then proving the change worked. Most guides jump straight to tactics: shorter forms, trust badges, faster pages. The tactics are fine, but they only help if you picked the right step, and you can only pick the right step if the funnel itself is measured correctly. In the funnels we are asked to debug, the measurement is usually where things go wrong first.
This guide follows five steps, in order:
| Step | Question | Output |
|---|---|---|
| 1. Measure correctly | Does the funnel count what you think it counts? | A funnel you can defend |
| 2. Find the step | Which step loses the most people that matter? | One step to work on |
| 3. Size the gain | What is fixing it worth, compared with the others? | A ranked list of steps |
| 4. Diagnose | Why do people leave at that step? | A hypothesis to test |
| 5. Fix and verify | Did the change move the step, and the total? | A decision to keep or roll back |
If you need the definitions and formulas first, start with what funnel analysis is.
Step 1: avoid the five measurement mistakes
New section
Mistake 1: counting pageviews instead of people
A visitor who refreshes the checkout page three times becomes three checkouts. The tell is a funnel where a later step is larger than an earlier one, which is impossible when you count people. Build funnels on visitors, or on identified users when people log in. TapCub funnels count each person once per step inside the window.
Mistake 2: no time window
Without a window, someone who saw pricing in January and bought in June converts in the same funnel as someone who did both in ten minutes. Pick a window that matches the decision: a session or 24 hours for a checkout, 14 or 30 days for a trial-to-paid funnel. TapCub windows go up to 180 days. Put the window in the report name so nobody compares a 1-day funnel with a 30-day one.
Mistake 3: forcing strict order
Real paths are messy: pricing, docs, pricing again, checkout. A funnel that requires steps with nothing in between drops everyone who detoured. Use "in order, anything in between" by default, and keep strict order for flows the product enforces, like a multi-step form.
Mistake 4: mixing segments and bots
A funnel over all traffic averages groups that behave nothing alike, and bot traffic is the worst of them: a crawler that loads every pricing URL but never starts checkout adds a fat top and a scary first drop. Filter to humans (TapCub's humans-only switch applies to funnels), then break down one dimension at a time.
Mistake 5: reading the total instead of the biggest drop
The overall conversion rate is a product of every step, so it moves when anything moves and points at nothing. Read the step-to-step drops first.
Before you trust a funnel, run the five checks below. They take five minutes.
Step 2: find the step that matters
New section
Take the sample checkout funnel: 2,400 visitors reach pricing, 1,248 start checkout, 674 reach payment and 508 complete. Overall conversion is 21%. The step conversions are 52%, 54% and 75%, so the two early steps each lose about half of the people who reach them.
Then break the worst step down by device, source and new versus returning. The breakdown is usually where the finding is: in the sample, the checkout-to-payment step loses 20% on desktop and 70% on mobile, which turns "checkout is leaky" into "mobile checkout has a problem". TapCub keeps up to 10 groups in a funnel breakdown, enough to compare devices, top channels or countries in one view.
Step 3: size the gain before you fix anything
New section
Overall conversion is the product of the step rates, so the same relative improvement at any step adds the same number of orders. Absolute points are different: a five-point gain is worth more at a step with a low rate. Using the sample funnel:
| Change | New step rate | Completed | Extra orders |
|---|---|---|---|
| Baseline | — | 508 | — |
| Pricing → checkout +10% relative | 52% → 57.2% | 559 | +51 |
| Checkout → payment +10% relative | 54% → 59.4% | 559 | +51 |
| Payment → complete +10% relative | 75% → 82.9% | 559 | +51 |
| Pricing → checkout +5 points | 52% → 57% | 557 | +49 |
| Checkout → payment +5 points | 54% → 59% | 555 | +47 |
| Payment → complete +5 points | 75% → 80% | 542 | +34 |
The lesson: no step is automatically the most valuable. Rank steps by the gain you can realistically get there and what it costs to get it. A copy change on the pricing page may be cheaper than rebuilding payment, even if payment looks worse. The conversion rate calculator helps you run the same arithmetic on your numbers.
Step 4: find out why people leave
New section
A funnel shows where people stop, not why. Use the tools that show behavior at that step:
- Path analysis: where did the people who dropped go instead? Back to pricing, to the FAQ, off the site?
- Heatmaps and session replay: what did they click, and where did they hesitate or rage-click?
- Form and error events: which field was the last one touched; did a validation error fire?
- Segments: does the drop concentrate on one device, browser, country or traffic source?
- Conversations: what do people ask in chat on that page? Questions are friction you can read.
For checkout steps, public research gives a useful starting list. Baymard Institute puts average documented online cart abandonment at 70.22%, from 50 studies, and lists extra costs such as shipping, taxes and fees as the most common reason in its survey of shoppers (Baymard Institute). Use it as a checklist of suspects, then confirm with your own data.
Step 5: fix one step and verify
New section
| Funnel stage | Common cause | Fix to test |
|---|---|---|
| Landing → product or pricing | Message does not match the ad or search query | Match the headline to the source; one clear next step |
| Product → cart | Missing information (price, size, delivery date) | Put the answer next to the button |
| Cart → checkout | Surprise costs, forced account creation | Show full costs early; allow guest checkout |
| Checkout → payment | Long forms, errors, slow mobile pages | Remove fields, inline validation, test on a phone |
| Payment → complete | Payment method missing, trust | Add the methods your countries use; show security and returns |
Change one thing at a time, then read the same funnel with the same window and filters before and after. Compare against a segment that did not get the change if you can, and watch the steps after the one you fixed: a fix that moves people forward can simply move the drop one step down. Save the funnel with its window and filters in the name ("Checkout · humans · 24 h") so next month's comparison uses the same definition.
| Check | Question | If the answer is no |
|---|---|---|
| People, not events | Is each step counted once per visitor? | Switch the funnel to count visitors |
| Window | Is a conversion window set and shown in the name? | Set one that matches the decision |
| Order | Is the order "in sequence, anything between"? | Relax strict order unless the product enforces it |
| Humans only | Are bots excluded? | Turn on the humans-only filter |
| Biggest drop | Did you read the step drops before the total? | Start there, then break down by device and source |
Conversion funnel metrics to track
New section
| Metric | Formula | Use it to |
|---|---|---|
| Step conversion | People at step N ÷ people at step N−1 | Judge a single step |
| Overall conversion | People at the last step ÷ people at step 1 | Track the whole process |
| Drop-off | People at step N−1 − people at step N | See where the people are lost |
| Time to convert | Median time from first to last step | Spot slow steps and set the window |
| Conversion by segment | Any of the above, split by device, source or country | Find where the problem lives |
If the open question is which stage of the customer lifecycle to work on at all, start with AARRR metrics and then build the funnel for that stage. A funnel is only as clean as the event names behind it; event naming conventions covers the rules.
FAQ
New section
What is conversion funnel optimization?
Finding the funnel step where the most valuable visitors drop off, diagnosing why, changing that step and verifying the result with the same funnel definition before and after.
How do I find the biggest drop-off in a funnel?
Read step-to-step conversion rather than the overall rate, filter out bots, and break the weakest steps down by device, source and new versus returning visitors.
Which funnel step should I optimize first?
The one with the best ratio of realistic gain to effort. Equal relative improvements add the same orders at any step, so effort and confidence usually decide.
How long should I run a funnel change before judging it?
Long enough to cover at least one full weekly cycle with the same window and filters, and longer for funnels with multi-day decisions.
What are common funnel analysis mistakes?
Counting pageviews instead of people, no conversion window, strict step order, mixing bots and segments, and reading only the overall rate.
See also: Drop-off rate.
See also: Funnel Analytics Tools: Criteria That Change Your Funnel Numbers.
See also: Funnel Analysis: How to Build, Read and Fix a Funnel.
Update history1
- Major updateRewritten as a funnel optimization guide: finding the step that matters, a worked gain table, common mistakes and an FAQ.
- First published
Was this article helpful?
We use your answers to fix and expand this page.
Couldn’t send. Please try again in a minute.
Thanks for the feedback. We’ll use it to improve this page.You’ve already sent feedback on this page. Thanks.

Product Analyst
Focuses on event analysis, conversion funnels, and retention, using clear metric definitions and structured analysis to explain product performance.



