Try the sheet here, on the example
The sheet is open below with sample numbers. Change any white cell and watch the results update, or press the blank-copy button to type your own data from scratch. Nothing you type leaves your browser.
Example from the sheet
| Business | Harbor Outdoor Gear | Change the confidence level and minimum conversions on Settings. | |||||
|---|---|---|---|---|---|---|---|
| Experiments logged | 14 | Variant wins | 3 | ||||
| Planned | 2 | Control wins | 1 | ||||
| Running | 3 | No clear difference | 4 | ||||
| Completed | 8 | Below planned sample | 3 | ||||
| Stopped | 1 | Not enough data | 1 | ||||
| Decision: ship variant | 3 | Win rate (variant wins / judged tests) | 37.5% | ||||
| Decision: keep control | 4 | Average uplift of winning variants | 22.5% | ||||
| Decision: iterate and retest | 2 | Confidence level used (%) | 95 | ||||
| Area | Tests | Variant wins | Control wins | No clear difference | Not judged yet | Win rate | Avg uplift of winners |
| Homepage | 2 | 1 | 0 | 1 | 0 | 50.0% | 15.7% |
| Product page | 1 | 0 | 0 | 1 | 0 | 0.0% | |
| Pricing | 2 | 0 | 0 | 1 | 0 | 0.0% | |
| Checkout | 2 | 1 | 0 | 0 | 1 | 100.0% | 26.7% |
| 2 | 0 | 1 | 0 | 1 | 0.0% | ||
| Paid ads | 2 | 0 | 0 | 1 | 0 | 0.0% | |
| Onboarding | 1 | 0 | 0 | 0 | 1 | ||
| Landing page | 2 | 1 | 0 | 0 | 1 | 100.0% | 25.2% |
| All areas | 14 | 3 | 1 | 4 | 4 | 37.5% |
Example from the “Summary” tab, with the numbers and formulas that are in the file.
What is inside
- Summary: Experiments by status and decision, results (variant wins, control wins, no clear difference, below planned sample, not enough data), win rate, average uplift of winners, a table by area and three charts.
- Experiments: The log, 200 rows: ID, name, area, owner, hypothesis, metric, status, dates, visitors and conversions for A and B, and planned visitors. It adds days run, the two rates, change, relative uplift, z, p-value, confidence and Result, then your Decision and learnings, with a Check column.
- Sample size: Enter your baseline rate, the smallest lift worth detecting and daily visitors to see the visitors needed per variant and the days to run, plus a table for five different lifts.
- Settings: Confidence level (90, 95 or 99), power, minimum conversions per variant, the as-of date and your list of areas.
The file opens on a start tab (“Start here”) that explains how to use it.
What makes it different
A log that only stores results lets a test count as a win because it looked good on day three. This one will not call a result until each side has the conversions you set and, if you entered a plan, the visitors you planned. In the sample, three of the twelve tests with data are called wins, three more are labelled Below planned sample and one Not enough data, so they are kept out of the win rate, which is 3 of 8 judged tests (37.5%).
How do I track A/B tests in a spreadsheet?
Give every experiment a row and write the hypothesis before the test starts, as a sentence that can be wrong: for example, that letting people check out without an account will cut abandonment. Record the area, the owner and the one primary metric you will judge it by, and set the status to Planned, then Running.
When the test ends, enter visitors and conversions for the control (A) and the variant (B). The sheet does the rest, and the Decision and What we learned columns turn the log into a record the team can search later, so the same idea is not tested twice.
How do I know if my A/B test result is significant?
The sheet runs a two-sided, pooled two-proportion z-test on conversions per visitor. If the absolute z reaches the threshold for your confidence level (1.96 for 95%), the result is called a win for the variant or the control. In the sample, the guest checkout test has a z of 4.24 and converts at 5.70% against 4.50%, an uplift of 26.7%, so the variant wins. The sticky add-to-cart test has a z of 1.83, which is an uplift of 15.5% but not enough to call, so the result is No clear difference.
Significance is evidence, not proof. The test assumes a sample you fixed in advance, one primary metric and two variants. If you stop when the numbers look good, test many variants or run many tests, some wins will be chance, so retest before a big decision.
How big does my A/B test sample need to be?
It depends on your baseline conversion rate, the smallest lift you care about, your confidence level and the power. With a 4% baseline, a 10% relative lift, 95% confidence and 80% power, the Sample size sheet shows 39,473 visitors per variant. Detecting a 20% lift needs 10,314 and a 5% lift needs 154,303, which is why small improvements are hard to prove on a low-traffic site.
Work this out before you start and type it into Planned visitors per variant: the log will not judge the test until both sides have reached it. If the test is on ads, the ad performance tracker holds the results by campaign.
How to use it
- On Settings pick the confidence level (90, 95 or 99) and the power, set the minimum conversions each side needs, the "as of" date, and edit the list of areas you test.
- On Sample size enter your baseline conversion rate, the smallest lift worth detecting and your daily visitors to see how many visitors each side needs and how long the test must run, before you start.
- On Experiments log each test in a row: ID, name, area, owner, hypothesis, primary metric and status, then the start date and the planned visitors per variant.
- When the test ends enter the visitors and conversions of the control (A) and of the variant (B), and the end date.
- Read the grey columns: conversion rates, uplift, z, p-value, confidence and the Result. Then record your Decision and what you learned. The Check column flags impossible numbers.
- Open Summary for experiments by status, results, win rate and average uplift of winners, with a breakdown by area and three charts.
Before you start:
- Capacity: 200 experiments, 10 areas.
- Conversion rate = conversions / visitors for each side. Relative uplift = variant rate / control rate - 1. The test is a two-sided pooled two-proportion z-test on visitors.
- Result says "Variant wins" or "Control wins" when |z| reaches the threshold for your confidence level (1.645, 1.96 and 2.576 for 90%, 95% and 99%), "No clear difference" when it does not.
- "Below planned sample" means a side has fewer visitors than you planned, and "Not enough data" means a side has fewer conversions than the minimum on Settings. Neither is judged, which guards against stopping a test the moment it looks good.
- The test assumes a fixed sample decided in advance, one primary metric and two variants. It does not correct for peeking, several variants or several metrics, and running many tests means some will win by chance. Treat a win as evidence and retest before a big bet.
- The sample data are made-up examples for a fictional shop. Use whole numbers for visitors and conversions.
Still running this on Excel?
This sheet suits a start and a limited amount of data. If your way of working grows, or you need more than one person working on it at the same time, Softwiro can build you a program that fits the way you work.
| In this sheet | In Custom Software from Softwiro |
|---|---|
| One file that one person edits at a time, and copies reach you on WhatsApp. | A program with a login for each user and permissions you define, and everyone sees the same data. |
| You type everything and check the numbers by hand. | We design the screens around your work steps, so each person records only what concerns them. |
| Columns and formulas are fixed as the sheet was designed. | We build what you need: reports, alerts and screens your way. |
| Backups and security are your own responsibility. | The program runs on a secured server with regular backups. |
Frequently asked questions
Is this A/B test log free?
Yes. Download it, use it for your own tests and change anything you like.
Does it work in Google Sheets?
Yes. It uses ordinary formulas, including SUMIFS and COUNTIFS, so it works in Excel 2016 and later, LibreOffice and Google Sheets.
What test does it use?
A two-sided pooled two-proportion z-test comparing the conversion rate of the control and the variant, with the p-value and confidence worked out in the sheet.
Why does a test say Below planned sample or Not enough data?
Below planned sample means a side has fewer visitors than you planned. Not enough data means a side has fewer conversions than the minimum on Settings. Neither is judged, which stops you calling a test early.
How many tests and areas can it hold?
200 experiments and 10 areas. Start a new copy when the log is full.
Is the template free?
Yes, with no sign-up and no email address. Use it, edit it and share it as you like.
Does it work in Google Sheets and on mobile?
The sheet uses formulas supported by Excel 2016 and later, Google Sheets and LibreOffice, so you can upload it to Google Drive and open it with Google Sheets on a computer or a phone.
What is the difference between the white and the gray cells?
You type in the white cells; the gray cells are calculated automatically. Don't type in a gray cell, or you will erase its formula.
Can I use the sheet from an AI assistant?
Yes. Add our MCP server (URL https://softwiro.com/api/mcp, no key needed) to your assistant and ask it to do the calculation in this template. The numbers come from the sheet's own formulas, and we do not store your data. The steps and a ready-made prompt are in this article: https://softwiro.com/blog/mcp-excel-templates-ai-agent-guide.html
All the templates in this group: Excel templates for paid ads: budget, performance and reports
More free templates for: Marketing and advertising
More free templates
- Ad Performance Tracker for Excel (ROAS, CPA)
- Social Media Content Calendar Template for Excel
- Breakeven ROAS Calculator for Excel
- Ecommerce Profit Margin Calculator for Excel
- Installment Payment Tracker for Excel
- Class Schedule Maker for Excel
- GPA Calculator for Excel (Weighted and Unweighted, by Term)
- Lesson Plan Template for Excel with Curriculum Tracker
- Donation Tracker Spreadsheet for Nonprofits (Excel)
- Expiry Date Tracker for Excel (Inventory)
- HOA Dues Tracking Spreadsheet for Excel