UserProof / Free tools / Fix check calculator
Fix check calculator: did a site change work? A before and after conversion rate calculator
For founders who build with a coding agent and have before and after visit counts: in the example, the rate went from 5.0% to 8.9%, up 3.9 points (95% interval 0.4 to 7.7), so it worked. Our own check on Sales.co (1 check, 203 visits, October 26, 2026) read 90 of 118 visits bouncing before and 45 of 85 after: down 23 points, interval 10 to 36 (worked example).
Your numbers
Example result
Worked. The rate went from 5.0% to 8.9%, a change of +3.9 points (95% interval 0.4 to 7.7). The interval stays on the good side of zero, so this looks like it worked. With no control, anything else that changed in the same weeks is mixed in.
UserProof's fix check would call this: improved.
| Group | Before | After | Change | 95% interval |
|---|---|---|---|---|
| This page or change | 5.0% (20 of 400) | 8.9% (34 of 380) | +3.9 points | 0.4 to 7.7 |
How to read the result
- Worked / got worse / inconclusive: the headline. It says worked only when the whole 95% interval is on the good side of zero (and, with a control, the interval against the control is too). It says got worse when the whole interval is on the bad side. Everything else is inconclusive.
- Change and 95% interval: the after rate minus the before rate, in percentage points, and the range of changes your counts are consistent with. A wide interval means too few visits, not a small effect.
- Net of the control: the page's change minus the control's change. If the whole site moved up, only the part above that counts as your fix.
- When the result is not a verdict, it also shows the visits on each side you would need to see a 20% relative change in the rate. To plan that before you ship, use the sample size calculator.
A verdict is about the rate, not the cause. Read it with what else changed on the page and on the site in the same weeks.
When a lift becomes clear
The same move, a 5% rate going to 9%, at four traffic levels. The lift is identical in each row; only the number of visits changes what you can say about it.
| Visits | Before | After | 95% interval, points | Verdict |
|---|---|---|---|---|
| 100 before, 100 after | 5 of 100 (5.0%) | 9 of 100 (9.0%) | −3.5 to 11.8 | Inconclusive (no clear change) |
| 500 before, 500 after | 25 of 500 (5.0%) | 45 of 500 (9.0%) | 0.8 to 7.2 | Worked (improved) |
| 2,000 before, 2,000 after | 100 of 2,000 (5.0%) | 180 of 2,000 (9.0%) | 2.4 to 5.6 | Worked (improved) |
| 100 before, 25 after | 5 of 100 (5.0%) | 3 of 25 (12.0%) | no interval: under 30 visits after | Inconclusive (needs 5 more visits) |
With 100 visits on each side the interval runs from −3.5 to 11.8 points and includes zero, so a lift of 4 points is inconclusive. With 500 it is 0.8 to 7.2, which clears zero. With 2,000 it narrows to 2.4 to 5.6. A rate that jumps from 5% to 12% on 25 visits is still inconclusive, because the floor is 30.
How it works
The method is the one in UserProof's fix check (src/lib/engine/verify.mjs and the interval helpers in findings.mjs), applied once to the four to eight numbers you type.
- Rates are
conversions ÷ visits, before and after. - Interval on each rate: the range the true rate plausibly sits in, which stays sensible even at 0 conversions. The formula is
(p + z²/2n ± z·√(p(1−p)/n + z²/4n²)) ÷ (1 + z²/n)with z = 1.96, limited to 0 to 100%. - Change and its interval: after rate minus before rate. The two rates' distances to their own Wilson bounds combine in quadrature: low = change − √((p₁−L₁)² + (U₂−p₂)²), high = change + √((U₁−p₁)² + (p₂−L₂)²).
- Floor: under 30 visits after the change there is no interval and no verdict, only how many more visits to wait for. No visits before is also inconclusive.
- Without a control: worked if the interval's low end is on the good side of zero; got worse if its high end is on the bad side; otherwise inconclusive. UserProof words: improved, worse, no clear change.
- With a control: the net change is the page's change minus the control's, and its interval combines the two in quadrature the same way. Worked (UserProof: lift confirmed) needs both the net interval and the page's own interval on the good side. Got worse needs the net interval wholly on the bad side. A page that moved the right way while the net interval reaches zero is inconclusive (UserProof: rose, not confirmed against other pages). The control needs 30 visits before and 30 after, or it is left out.
- Visits needed:
⌈2 × (1.96 + 0.84)² × p(1−p) ÷ δ²⌉per side, where p is the before rate (1% if it was 0) and δ is 20% of p (at least 0.2 points). That is 95% confidence and 80% power.
Where the product reads a before window of 14 days and an after window of up to 14 days, here you choose the periods. Use the same number of days on each side, and for a control the same days too.
Worked example
The example above: 20 of 400 visits converted before (5.0%) and 34 of 380 after (8.9%), so the change is 3.9 points. Wilson 95% intervals: 5.0% before is about 3.3% to 7.6%, and 8.9% after is about 6.5% to 12.2%. Combined, the change interval is 0.4 to 7.7 points. Its low end is above zero, so with no control it reads worked. To see a 20% relative change from 5.0% (a move to 6.0%) you would need about 7,448 visits on each side, so this result is a big lift that the data happen to support, not a precise size.
Now add a control. Say the site's other pages went from 150 of 3,000 (5.0%) to 160 of 2,900 (5.5%): up 0.5 points on their own. The net change is 3.4 points with a 95% interval of −0.3 to 7.3. That reaches zero, so the page moved the right way but the verdict is inconclusive (UserProof: rose, not confirmed against other pages). Press "Try the example with a control" to see it.
Check against a real read: the Sales.co check on the fix check page had 90 of 118 visits bounce before and 45 of 85 after. Enter it with "lower is better" and the calculator gives a change of −23.3 points with a 95% interval of −35.8 to −10.1. The page prints it as a 23-point drop, 10 to 36.
What the numbers say
- The same 4-point lift on 100 visits a side is inconclusive (−3.5 to 11.8 points) and on 500 a side it clears zero (0.8 to 7.2).
- Seeing a 20% relative change from a 5% rate takes about 7,448 visits on each side.
- With a control of 2,900 to 3,000 visits a side, a 3.9-point lift on 380 to 400 visits became inconclusive (−0.3 to 7.3 points net), because the other pages rose too.
- Under 30 visits after the change, the calculator shows no verdict at any rate.
Limits: what it can't tell you
- It can't isolate the cause. Before and after is not an A/B test. A second change on the page, a new traffic source, a launch or a holiday in the same weeks all land in the result. A control absorbs site-wide drift; it does not remove a change that hit only your page.
- When to use an A/B testing tool instead: when you have not shipped yet and can build both versions, when you have enough traffic to split, or when the decision is large and hard to reverse. A split of simultaneous traffic is the only way to isolate the cause.
- Counts have to match. Visits and conversions should come from the same tool, the same definition of a visit and the same number of days on each side. Exclude bots and your own traffic.
- Visits that repeat. If the same person visits twice, the counts are not independent and the interval is too narrow. Use unique visitors if you can.
Run the check on recorded visits
The calculator reads the four numbers you type. UserProof does the same check on recorded visits. When your coding agent ships a fix and marks it applied, UserProof compares the page before and after against your other pages, with the same interval and the same 30-visit floor. Then it tells the agent whether the fix worked. Read how on the fix check page. Free for 3 sites and 5,000 recorded visits a month; Pro is $29 a month for 50,000; Scale is $99 for 250,000 (pricing).
Before you ship the next change, the conversion rate increase calculator says what the lift would be worth a month and whether your traffic can show it in 14 days.
Related: the funnel calculator, the dead click finder, all free tools.
Sources, read November 14, 2026
- UserProof fix check (the 30-visit floor, the interval, the control, the verdict words, the visits-needed formula): our code, src/lib/engine/verify.mjs and findings.mjs, read November 14, 2026.
- Sales.co check, 90 of 118 and 45 of 85: UserProof fix check page, read November 14, 2026.
- UserProof plans: pricing page and our facts file, read November 14, 2026.
- The example numbers are made up to show the math; they are not a benchmark. The calculator's code is the open file fix-check-calc.js.
Cite this
UserProof, "Fix check calculator," https://getuserproof.com/fix-check-calculator, updated November 14, 2026. Change in a rate between two periods with a 95% Wilson-based interval, no verdict under 30 visits after the change, and a difference in differences against a control.
FAQ
How do I calculate a before and after conversion rate?
Divide the conversions by the visits for each period, then subtract the before rate from the after rate. In the example, 20 of 400 visits is 5.0% and 34 of 380 is 8.9%, a change of 3.9 points. The change alone is not enough: the 95% interval says how much of it could be chance, which depends on how many visits each side has.
How many visits do I need after a change?
At least 30 for the calculator to say anything, which is UserProof's own floor. To see a 20% relative change from a 5% rate (5.0% to 6.0%) with 95% confidence and 80% power you need about 7,448 visits on each side. Small changes on low rates need far more traffic than most pages get in a week.
Is this the same as an A/B test?
No. An A/B test splits visitors between two versions at the same time, so the only difference between the groups is the change. A before and after read compares two different weeks, so a holiday, a new traffic source or a second change on the same page shows up as part of the result. If you can build both versions and split traffic, use an A/B testing tool. This calculator is for the change you already shipped.
What is the control, and do I need one?
The control is the rest of your site, measured over the same two periods. If it moved by the same amount, the change on your page was probably not your fix. With a control the calculator reports the difference in differences: the page's change minus the control's change, with its own 95% interval. You don't need one, but without it a seasonal swing can look like a win.
What does the 95% interval mean here?
It is the range of changes consistent with your counts. If the whole range is above zero, the rate very probably went up; if it spans zero, the data can't tell a real change from noise. It is built from a range on each rate, combined for the difference, the same method UserProof's fix check uses. It tells you how far off the real change could be.
Does the calculator store or send my numbers?
No. The calculator runs in your browser, makes no requests and saves nothing. The page itself has the same visit analytics as the rest of this site, and the form and the result are marked so that analytics never records what you type or what the result shows. Reload the page and it starts from the example again.