What a funnel calculation tells you that a single conversion rate cannot
An end-to-end conversion rate tells you the size of your problem. A stage-by-stage funnel tells you where it is. Both come from the same data, and the second is strictly more informative, because a single rate of 0.18% is consistent with thousands of different funnels - some leaking almost everything at the first step, some bleeding steadily at every step.
A funnel is a chain of conditional rates. Each stage rate answers a narrow question: of the records that reached this stage, what share reached the next one? Because each rate is conditional on surviving the stage above, the rates multiply rather than add. Four stage rates of 50% each give 0.5⁴ = 6.25% end to end, not 200%. This is the single most important structural fact about funnels and the source of most of the errors people make when they report them.
The multiplicative structure has a consequence that changes how you prioritise work. Because the stages multiply, a given relative improvement is worth exactly the same wherever you apply it. Lifting a 6% stage rate to 6.6% and lifting a 25% stage rate to 27.5% are both 10% relative gains, and both raise the end-to-end rate by precisely 10%. The one limit is the ceiling: a stage already converting at 95% cannot deliver a 10% relative gain, because 104.5% is not a rate. Below that ceiling the equivalence is exact, so the stage with the ugliest number is not automatically the stage to fix. Choose instead by cost: which 10% is cheapest to buy, and which stage still has the headroom to give it?
The other thing a funnel gives you is the inverse calculation. Divide your customer target by the end-to-end rate and you get the top-of-funnel volume required to hit it. That is the number that turns a revenue goal into a traffic or lead-flow goal, and it is the honest way to answer a demand for more customers when nobody wants to hear that rates would have to change.
The arithmetic: stage rates, drop-off, and why the product telescopes
Each stage rate is rk = nk+1 ÷ nk, and the drop-off is simply 1 − rk. Multiply the four stage rates of a five-stage funnel and you get the end-to-end rate.
Write the product out and you can see why it must equal the last stage over the first: (n₂/n₁) × (n₃/n₂) × (n₄/n₃) × (n₅/n₄). Every intermediate count appears once in a numerator and once in a denominator, so they cancel, leaving n₅/n₁. This telescoping property is a free audit: if the product of your stage rates does not reproduce your headline rate, your stage counts are not measuring what you think they are - usually because one stage is counted over a different date range or counts a different object, such as records rather than sessions. This calculator prints both the direct division and the product so you can check they agree.
Drop-off cannot be added across stages. A funnel with drop-offs of 94%, 60%, 70% and 75% has not lost 299% of anything. Each percentage is conditional on a different base. In absolute terms the losses do add, and they are the figures worth reporting to people outside the funnel: 18,800 records lost at the first step, 720 at the second, 336 at the third, 108 at the fourth, totalling 19,964 of the original 20,000.
Neither view is a priority metric on its own. The top stage is always the biggest absolute loser, because that is where nearly all the volume sits, and it is often the lowest rate as well; the relative-gain rule from the previous section is what breaks the tie.
The inverse calculation divides the target by the end-to-end rate: n₁ = T ÷ R. Round up, because you cannot buy a fraction of a visit. If the required volume comes out below what you already have, the target is reachable at today's rates and the extra requirement is zero - which is worth knowing before you ask for budget.
Worked example: a 20,000-session B2B funnel closing 36 deals
Last quarter a demand-gen team recorded 20,000 sessions, 1,200 form fills, 480 marketing-qualified leads, 144 accepted opportunities and 36 closed-won deals. Average first-year value is $9,500 and the target is 60 deals a quarter.
- Stage 1 → 2. 1,200 ÷ 20,000 = 0.06 = 6.00%, so the drop-off is 94.00% and 18,800 sessions are lost.
- Stage 2 → 3. 480 ÷ 1,200 = 0.40 = 40.00%, drop-off 60.00%, 720 leads lost.
- Stage 3 → 4. 144 ÷ 480 = 0.30 = 30.00%, drop-off 70.00%, 336 MQLs lost.
- Stage 4 → 5. 36 ÷ 144 = 0.25 = 25.00%, drop-off 75.00%, 108 opportunities lost.
- End to end, directly. 36 ÷ 20,000 = 0.0018 = 0.180%.
- End to end, as a product. 0.06 × 0.40 × 0.30 × 0.25 = 0.0018. The two agree, so the stage counts reconcile.
- Pipeline value. 36 × $9,500 = $342,000 of closed business.
- Value of one session. $342,000 ÷ 20,000 = $17.10. This is the number that tells you what you can afford to pay for traffic.
- Sessions needed for 60 deals. 60 ÷ 0.0018 = 33,333.3, so 33,334 sessions - which is 13,334 more than you have, a 67% increase in top-of-funnel volume.
- Revenue at the target. 60 × $9,500 = $570,000, against $342,000 today.
Now test the relative-gain rule. Suppose instead of buying traffic you improve one stage by 10% relative. Lift stage 1 → 2 from 6.00% to 6.60%: the end-to-end rate becomes 0.066 × 0.40 × 0.30 × 0.25 = 0.00198, or 0.198%, and 20,000 sessions now yield 39.6 deals. Lift stage 4 → 5 from 25.00% to 27.50% instead: 0.06 × 0.40 × 0.30 × 0.275 = 0.00198 as well, and again 39.6 deals. Identical, because multiplication does not care which factor you scale. What differs is the work: raising a 6% form-fill rate by 0.6 points is a landing-page and offer problem, while raising a 25% close rate by 2.5 points is a sales-enablement and pricing problem, and one of those is almost certainly cheaper at your company than the other.
Notice also how far short the improvement route falls on its own. 39.6 deals against a target of 60 means no single 10% stage gain gets you there. To hit 60 on 20,000 sessions you need an end-to-end rate of 60 ÷ 20,000 = 0.30%, which is 1.667 times today's 0.18% - a 66.7% relative gain, spread across stages however you like.
How to read the stage rates once you have them
Start by checking that the funnel reconciles, then compare each stage against its own history rather than against a benchmark. Stage definitions are local: one company's MQL is another's raw lead, and one store's “product view” fires on a category page. Cross-company stage benchmarks are therefore close to meaningless, whereas your own stage rate last quarter is a fair comparison as long as the definitions have not changed.
Three patterns are worth naming. A funnel that loses nearly everything at the first step and holds up afterwards is usually a traffic-quality or offer problem: you are attracting people who were never in market, or asking for too much too early. A funnel with a healthy first step and a collapse in the middle is usually a qualification problem: the definitions between stages are doing the filtering that targeting should have done upstream. A funnel that leaks evenly at every stage is usually a process problem - slow follow-up, unclear next steps, or handoffs that drop records.
Watch the volumes as well as the rates. A stage rate computed on a small base moves violently for no reason: 36 closed deals out of 144 opportunities carries a 95% margin of error of 1.96 × √(0.25 × 0.75 ÷ 144) = 0.0707, or roughly ±7.1 percentage points, so a move from 25% to 28% next quarter is well inside the noise. The method for that error bar is in the conversion rate calculator, and if you want to declare that a change caused a difference, size the comparison first with the A/B test sample size calculator.
Finally, price the funnel. Value per top-of-funnel record - $17.10 in the worked example - is your ceiling for paid traffic before any margin. Compare it against what you actually pay using the cost per acquisition calculator and check the resulting unit economics with the LTV to CAC ratio calculator. If a lead costs more than the funnel returns, the answer is a funnel change, not a bigger media budget.
End-to-end conversion when every stage converts at the same rate
| Rate at every stage | 4 stages (3 steps) | 5 stages (4 steps) | 6 stages (5 steps) |
|---|---|---|---|
| 20% | 0.800% | 0.160% | 0.032% |
| 30% | 2.700% | 0.810% | 0.243% |
| 40% | 6.400% | 2.560% | 1.024% |
| 50% | 12.500% | 6.250% | 3.125% |
| 60% | 21.600% | 12.960% | 7.776% |
| 70% | 34.300% | 24.010% | 16.807% |
Read down a column to see the cost of compounding, and across a row to see the cost of adding a stage. Splitting a funnel into more measured stages does not change the end-to-end rate - but adding a genuine extra gate that records must pass does.
Funnel mistakes that produce confident wrong answers
- Adding stage rates or drop-offs. They are conditional on different bases. Multiply the survival rates instead, and check the product against the direct division.
- Counting different objects at different stages. Sessions at the top, people in the middle, accounts at the bottom. Each transition then measures something other than survival, and the product stops telescoping.
- Measuring all stages over the same calendar period. With a long sales cycle, this quarter's closed deals came from last quarter's leads, so a same-period rate understates conversion during growth and overstates it during decline. Cohort the records by entry date instead.
- Ranking fixes by the ugliest stage rate. Equal relative gains anywhere in a multiplicative chain are worth the same end to end. Rank by cost and by headroom, not by which number looks worst.
- Reading a stage rate off a tiny base. 36 of 144 carries a 95% margin of error near ±7 percentage points, so quarter-to-quarter movement at that stage is mostly noise.
- Letting records skip stages. An inbound deal that goes straight to opportunity makes a later stage larger than the one above it, which breaks the arithmetic. Backfill the skipped stage or exclude that path.
- Assuming more top-of-funnel volume converts at the same rate. The inverse calculation holds rates constant by construction; in practice the marginal traffic you buy to reach 33,334 sessions is usually worse than the traffic you already have.
Same-period funnels lie when the sales cycle is long
If your cycle is 90 days and you divide this quarter's wins by this quarter's leads, you are dividing outcomes from one cohort by inputs from another. During any period of growth in lead volume the denominator is larger than the group that produced the numerator, so the measured rate is too low; when lead volume falls, it is too high. The fix is cohort accounting: tag every record with the date it entered stage 1, then measure that cohort's progression as it matures. Report the cohort rate for decision-making and keep the same-period rate only as an operational dashboard number. The sales pipeline coverage calculator handles the timing side of the same problem from the quota direction.
Related methods and where to go next
A funnel model is one of three ways to describe the same commercial reality, and each answers a different question.
Rates and stages - this calculator - answer “where do records stop moving?” It is the right tool for diagnosing a process and for converting a target into required volume. For a single stage in isolation, or for the error bar around any one rate, use the conversion rate calculator. For the last two stages of a store's funnel specifically, the cart abandonment rate calculator uses exactly the same multiplicative arithmetic with the conventional e-commerce labels.
Unit economics answer “is this funnel worth running?” Value per top-of-funnel record, cost per acquisition and lifetime value together decide whether more volume is a good idea at all. Work them with the customer acquisition cost calculator and the customer lifetime value calculator.
Experiments answer “did my change actually work?” A stage rate that improves after a redesign is not evidence on its own, because mix and seasonality move at the same time. Read the change with the A/B test significance calculator.
Two limits are worth stating plainly. The model assumes a strict sequence, when real journeys loop. And it assumes the stages are independent, when loosening a qualification gate raises the rate into the next stage and usually lowers the rate out of it - often leaving the end-to-end rate unchanged. Recompute the whole chain after any change to a stage definition before comparing anything to history.
Key terms
- Stage conversion rate
- The share of records at one stage that reach the next. Conditional on having reached the earlier stage, which is why stage rates multiply.
- Drop-off
- One minus the stage conversion rate. Expressible as a percentage of the stage above, or as an absolute count of records lost.
- End-to-end conversion rate
- Final stage divided by first stage, equal to the product of all the stage rates. Often a fraction of a percent in B2B funnels.
- MQL and SQL
- Marketing-qualified and sales-qualified lead: two internal gates whose definitions are local to each company, which is why cross-company stage benchmarks do not transfer.
- Cohort funnel
- A funnel measured by tagging records with their entry date and following that group forward, rather than dividing one period's outputs by the same period's inputs.
- Value per top-of-funnel record
- Closed value divided by stage-1 volume. The ceiling on what a visit or enquiry is worth to you before margin and cost of sale.