What a conversion rate measures, and what it hides
A conversion rate is the share of a defined audience that completes a defined action inside a defined window. All three definitions do real work, and changing any one of them changes the number without anything changing on your site.
Because the rate is a ratio, it moves for two independent reasons: the numerator can change, or the denominator can change. A fall from 3.5% to 3.0% might mean your checkout got worse, or it might mean you bought a wave of cold display traffic that converts at 0.4% and diluted a healthy base. Those two diagnoses point at opposite actions, and the rate on its own cannot tell them apart. Read a conversion rate next to its two components and next to its traffic mix, always.
What the rate is genuinely good for is comparison against itself. Your own trailing rate for the same page, the same source and the same season is the most honest benchmark available, because it holds constant everything a cross-industry average cannot: your price point, your delivery promise, your brand awareness, your device split. The rate is also the right unit for pricing effort. Once you know that 28.6 visitors currently buy one conversion, you can compare the cost of finding 28.6 more visitors against the cost of persuading one more of the ones you already have.
The formula, and the two decisions hidden inside it
The arithmetic is one division: conversions divided by visitors, times 100 to state it as a percent. The difficulty is never the division. It is choosing the denominator and the attribution window, and both choices are yours to make and to write down.
Sessions or users? A session-scoped rate answers “what share of visits ended in a purchase”. A user-scoped rate answers “what share of people bought”. Because most buyers visit more than once before ordering, the user-scoped rate is the larger of the two, often substantially. Google Analytics 4 reports both, and they are not interchangeable. Pick one, name it in the report title, and never compare a session rate in one month against a user rate in another.
How long does a visitor have to convert? If someone lands on Monday and orders on Thursday, whose Monday does that order belong to? Last-click attribution credits the Thursday session, so a Monday-scoped denominator holds a visitor with no conversion attached. Over a stable period this washes out. Over a launch week it does not, and it is the usual reason a big traffic day appears to have destroyed the conversion rate.
Two derived quantities are worth computing every time. Visitors per conversion is the reciprocal of the rate and is far easier to reason about in a planning meeting: at 3.5%, one conversion costs 28.6 visitors, so a 2,000-visitor campaign is worth about 70 conversions. And the sampling interval — roughly 1.96 times the square root of p(1 − p)/n — tells you how much of a period-on-period move is noise. It is the difference between a metric and a mood.
Worked example: 868 conversions from 24,800 sessions
A shop records 24,800 sessions and 868 orders last month, against a 3.0% rate the month before. The team's target is 4.0% and average order value is $68.
- Divide. 868 ÷ 24,800 = 0.035. Multiply by 100: 3.5%.
- Invert it. 24,800 ÷ 868 = 28.57 visitors per conversion.
- Compare with last month. The absolute change is 3.5% − 3.0% = 0.5 percentage points. The relative change is 0.5 ÷ 3.0 = 0.1667, so +16.7%. Both statements describe the same movement; only the second one is a “lift”.
- Check whether the move is real. With p = 0.035 and n = 24,800, the standard error is √(0.035 × 0.965 ÷ 24,800) = √0.000001362 = 0.001167, which is 0.117 percentage points. The 95% interval is 3.5% ± 1.96 × 0.117, or 3.27% to 3.73%. Last month's 3.0% sits outside that interval, so the improvement is more than small-count noise.
- Price the target. At 4.0%, the same 24,800 sessions would produce 24,800 × 0.04 = 992 orders. You have 868, so the gap is 124 orders, worth 124 × $68 = $8,432 a month.
- Or buy the gap with traffic. To reach 992 orders at today's 3.5% you need 992 ÷ 0.035 = 28,343 sessions — 3,543 more visitors. If media costs more than $8,432 for 3,543 sessions, fixing the page is the cheaper route. If it costs less, buy the traffic.
That last comparison is the whole reason to run this calculation. Step 5 and step 6 buy the same 124 orders by different means, and the calculator prices both so you can choose.
How to read the result without fooling yourself
Read the interval before the point estimate. A rate built on 12 conversions and one built on 12,000 look identical on a dashboard and mean entirely different things. At 3.33% from 3,000 visitors the 95% interval is roughly ±0.64 percentage points — wide enough that a move to 3.9% proves nothing. At 868 conversions from 24,800 visitors it narrows to about ±0.23 points. Small denominators do not merely add uncertainty; they invite you to act on noise, which costs more than doing nothing.
Then segment before you celebrate. A site-wide rate is an average over wildly different populations: branded search converts far better than cold display, returning users better than new, desktop typically better than mobile on high-consideration purchases. If your blended rate rose while every segment fell, your mix changed rather than your site. This is Simpson's paradox in its most expensive commercial form, and it is common enough that segmenting by source and device should be automatic.
Finally, resist cross-industry benchmarks. Published averages mix session-scoped and user-scoped definitions, mix categories with a ten-minute purchase decision against ones with a six-week decision, and rarely disclose the traffic mix behind the figure. Your own trailing twelve months, segmented, is worth more than all of them. And when you genuinely need to know whether a change helped, stop comparing periods: run a controlled test and read it with the A/B test significance calculator, sized in advance with the sample size calculator.
Conversions per 1,000 visitors, and visitors per conversion
| Conversion rate | Conversions per 1,000 visitors | Visitors per conversion | Conversions per 100,000 visitors |
|---|---|---|---|
| 0.25% | 2.5 | 400 | 250 |
| 0.5% | 5 | 200 | 500 |
| 1% | 10 | 100 | 1,000 |
| 2% | 20 | 50 | 2,000 |
| 3% | 30 | 33.3 | 3,000 |
| 3.5% | 35 | 28.6 | 3,500 |
| 5% | 50 | 20 | 5,000 |
| 8% | 80 | 12.5 | 8,000 |
| 10% | 100 | 10 | 10,000 |
Multiply the middle column by your value per conversion to get revenue per 1,000 visitors, which is the same quantity a revenue-per-visitor calculation reports.
Relative lift and absolute lift are different numbers
“We improved conversion by 1%” is the most expensive ambiguous sentence in marketing. On a 3.5% baseline it can mean 3.535%, which is a 1% relative lift, or 4.5%, which is a rise of 1 percentage point. Per 1,000 visitors the first wins 0.35 extra orders (35.35 instead of 35) and the second wins 10 (45 instead of 35), so the second is nearly twenty-nine times the first.
Use percentage points for the arithmetic difference between two rates and percent for the ratio between them. Going from 3.0% to 3.5% is a rise of 0.5 percentage points and a relative lift of 16.7%. Both are correct; only one of them is a “16.7% improvement”. Experiment platforms almost always report relative lift and finance teams almost always plan in absolute points, which is exactly how a test that “won 20%” becomes a forecast nobody can reconcile.
The distinction also decides how much traffic a test needs. Required sample size scales with the inverse square of the absolute effect, so halving the absolute gap you want to detect quadruples the visitors you need. On a low baseline a modest-sounding relative target is a very small absolute effect, and therefore a very large traffic bill. A 10% relative lift on a 1% baseline is 0.1 percentage points; the same 10% relative lift on a 20% baseline is 2 points, twenty times the absolute gap and a small fraction of the traffic. State the target in both units before anyone plans a test.
Mistakes that produce a wrong conversion rate
- Mismatched scopes. Session-scoped orders over a user-scoped denominator, or a monthly conversion count over a weekly visitor count. Both numbers must come out of the same query.
- Counting bot and internal traffic in the denominator. Filter your own office, your QA runs and known crawlers, or you are diluting the rate with visitors who were never going to buy.
- Letting one page's rate stand for the site. A landing page rate and a site-wide rate answer different questions. Name the scope in the metric itself.
- Comparing periods of different length or seasonality. A 28-day month against a 31-day month, or December against February, compares calendars rather than performance.
- Acting on a difference smaller than the sampling interval. If the 95% interval is ±0.6 points, a 0.3-point move is not a finding. Compute the interval before the meeting, not after it.
- Double-counting conversions. A thank-you page that fires on refresh, or a lead form that submits twice, inflates the numerator quietly and permanently.
- Averaging rates instead of pooling counts. The conversion rate of three pages is not the mean of their three rates. Add the conversions, add the visitors, then divide once.
- Optimising the rate rather than the profit. Removing shipping charges lifts conversion and can still lose money. Judge changes on margin per visitor, not on the rate alone.
Where the conversion rate sits among its neighbours
The conversion rate is the middle term of every commercial identity worth knowing. Revenue equals visitors times conversion rate times average order value, so a 10% gain in any one of the three multiplies out identically. That is why revenue per visitor is often the better headline metric: it moves when the rate moves and when order value moves, and it cannot be flattered by discounting your way to a higher rate.
On the cost side, the rate is what turns a media price into an acquisition cost. Cost per acquisition equals cost per click divided by conversion rate, so at unchanged media prices, doubling the rate halves your cost per acquisition. That identity is the strongest argument for conversion work in any budget conversation: it lowers the price of every click you have already bought.
When the site-wide rate falls and you need to know where, one ratio is too blunt. Break the journey into steps and read the drop-off at each one with the sales funnel drop-off calculator, or go straight to the checkout with the cart abandonment rate calculator, where the single largest leak in most retail funnels lives. If you are forecasting rather than diagnosing, work backwards from the goal with the website traffic goal calculator, and price the order side with the average order value calculator.
Key terms
- Conversion
- Any action you have chosen to count as success: an order, a signup, a qualified lead, a download. It is a definition rather than a fact, so write it down.
- Session-scoped rate
- Conversions divided by sessions. Answers what share of visits converted.
- User-scoped rate
- Conversions divided by distinct users. Answers what share of people converted, and is the larger figure whenever buyers visit more than once before ordering.
- Percentage point
- The unit of the arithmetic difference between two rates. 3.0% to 3.5% is 0.5 percentage points, not 0.5%.
- Relative lift
- The ratio change between two rates, (new - old) / old. The 3.0% to 3.5% move is a 16.7% relative lift.
- Visitors per conversion
- The reciprocal of the rate, 100 divided by the percentage. Easier to plan with than a small percentage.
- Sampling interval
- The range of true rates consistent with your observed counts: approximately p plus or minus 1.96 times the square root of p(1 - p) / n.