Funnel Conversion Calculator

The underlying calculation

Funnel conversion is the product of the step-level rates: overall = step1 rate × step2 rate × step3 rate, and so on.

Because the rates multiply, the weakest step constrains everything downstream. Improving a strong step moves the total very little.

Step rate = users reaching the next step ÷ users at this step, which requires the funnel to be instrumented rather than inferred.

Worked through

40,000 sessions → 6,400 product views (16%) → 3,333 carts (52%) → 1,333 checkouts (40%) → 1,000 orders (75%). Overall: 2.5%.

The weakest step is checkout initiation at 40%. Raising it to 50% gives 1,667 checkouts and 1,250 orders, a 25% increase in orders from one step.

Raising the strongest step, checkout completion, from 75% to 85% gives 1,133 orders, a 13% increase for a similar amount of work.

The multiplicative structure means effort should go to the lowest rate, and the funnel view is what identifies it. A site-wide conversion figure never would.

Where it goes wrong

Real journeys are not linear. People re-enter at different points, browse across sessions and skip steps entirely, so a strict funnel understates conversion by losing anyone who deviates.

Cross-device journeys break it further, since a phone session and a laptop purchase are recorded as two unrelated funnels with one abandonment and one orphan order.

Making it useful

Instrument the steps that matter rather than trying to model the whole journey, and accept the funnel as a simplification that identifies the weak point rather than as a description of behaviour.

Then re-check after each improvement, since fixing the weakest step moves the constraint somewhere else and the priority changes with it.

Why the weakest step is not always the one to fix

The multiplicative logic identifies the largest opportunity and not the cheapest one. A 40% step might be structurally hard to improve while a 75% step has an obvious, quick fix.

The useful ranking is expected gain divided by effort, which requires estimating both rather than reading the funnel alone.

It is also worth checking whether a low rate is genuinely a problem. Sixteen percent of sessions reaching a product page may be correct for a site where much of the traffic arrives on content pages with no purchase intent, and forcing that number up by pushing product at researchers usually damages the pages doing their actual job.

Segmenting the funnel by device usually shows the constraint in a different place for mobile and desktop, which means a single set of priorities will be wrong for whichever segment is larger.

Building the funnel twice is little extra work and it prevents effort going to a step that was only ever a problem on the device carrying a third of the traffic.

Rebuilding the funnel after any significant change is worth the effort, since improving one step moves the constraint and the previous priority list stops being correct.

Comparing the funnel for new against returning visitors identifies which steps are barriers to unfamiliarity and which are barriers for everyone.