Tells you who to ring first
Every customer still with you gets a risk score and a place in the queue. Start at the top and work down for as long as your week allows.
Free churn prediction tool
Some of your customers are going to leave and you would rather hear it now than at renewal. Upload a CSV of your customers with a column saying who has already left, and get back a ranked list of who looks most like them — plus what the pattern hinges on. It runs in your browser, so the file never leaves your machine.
Drop a customer CSV here, or
One row per customer, up to 20,000 rows, with a column saying who churned. The rest is worked out for you.
The file is read, encoded and modelled in this browser tab. It never reaches our server, so there is nothing for us to keep.
- No account
- Nothing uploaded
- Ranked at-risk list
- Free, no sign-up
- CSV export
What this churn analysis tool does
You know your churn rate. You do not know which accounts make it up, or which of the things you believe about churn are actually true.
Honest about how well it works
The tool checks itself against customers it was not shown, so the accuracy you see is the accuracy you would get in real life — not a flattering number.
Shows what the pattern hinges on
Which of your columns actually separate the customers who left from the ones who stayed, and in which direction. Useful for deciding what to fix, not just who to call.
Warns you when the answer is too good
If one of your columns quietly gives the game away — a refund date, a closed-account flag — the results are worthless. You get told before you see the number, not after.
Says no when your data is too thin
If too few of your customers have ever left, there is no pattern to find. The tool tells you that instead of making something up.
Download the list, nothing uploaded
Take the ranked customers away as a CSV. Your file is read inside your browser tab and never reaches us. Free, no sign-up.
How to predict customer churn from a CSV
- Step 1
Export your customers to CSV
One row per customer, and a column that says whether they left — churned, cancelled, true/false, anything of that shape. Bring along whatever else you record about them: plan, signup date, spend, usage, support tickets.
- Step 2
Drop the file on this page
It is read inside your browser tab and nothing is uploaded. No account, no waiting.
- Step 3
Work down the list
Read any warnings first, then the ranked at-risk customers and what the pattern hinges on. Download the list and give the top of it to whoever makes the calls.
A file this shape is enough to start:
customer_id,signup_date,last_active,monthly_sessions,support_tickets,plan,seats,churned
CUST_001,2024-01-15,2025-08-20,15,2,pro,4,1
CUST_002,2024-03-10,2026-09-01,45,0,team,12,0
CUST_003,2024-06-01,2025-08-15,8,5,starter,1,1Churn rate calculator, or churn prediction model?
People search for both and mean different things. A churn rate answers how many left. Churn prediction answers which ones are about to. You need the first to make sense of the second, so you get both here.
Churn rate
churned in the period ÷ customers at the start of the period
One division, no tool needed. Every result here shows yours anyway — how many left, how many stayed, and the share — because losing 3% of your customers and losing 40% are read completely differently.
Churn prediction
learned from the customers who already left
Needs history, and needs a decent number of customers who have actually left to learn from. What you get back is a ranked list of accounts, which a rate can never give you.
Customer retention prediction is the same thing said the other way round: a low churn score is a retention score. The list is ranked by risk because that is where a retention team’s week actually goes.
What it does not do
- It does not tell you what causes churn. It tells you what the leavers had in common. Support tickets go with churn nearly everywhere, but stopping people raising tickets would save nobody.
- One file, one table. It cannot join your orders to your customers to your support tickets. Anything you want it to use has to already be a column on the row.
- It says whether, not when. You get a risk ranking, not a date. If you need to know how long someone has left, this is the wrong tool.
- It does not know about your pricing change. If what drives churn shifted partway through your history, it quietly averages across both periods.
- Nothing is saved. Close the tab and the results are gone. Download the CSV if you want to keep the scores.
Frequently asked questions
What is churn prediction?
Churn prediction is working out, from what you already know about a customer, how likely they are to leave. A model learns the pattern from customers who have already gone and then scores the ones who are still here, so you can rank them by risk. It answers “who is about to leave”, which is a different question from “how many left last month” — that one is the churn rate, and it needs no model at all.
How do you calculate churn rate?
Divide the customers who left during a period by the customers you had at the start of it, then multiply by 100. If you began January with 500 customers and 15 cancelled, that is 15 ÷ 500 = 3% monthly churn. Customers who joined mid-period are usually left out of the denominator, because counting them makes the number look better than it is. Revenue churn is the same sum done on recurring revenue rather than logos.
What is a good churn rate for SaaS?
For a business selling to mid-market and enterprise, good is roughly 5-7% annual logo churn, which is under 1% a month. Self-serve and SMB products run much hotter: 3-5% a month is normal there and is not a sign of a broken product. Consumer subscriptions are hotter still. The benchmark that matters more than any of these is whether your churn is falling as your cohorts mature.
What is the difference between customer churn and revenue churn?
Customer churn counts accounts that left; revenue churn counts the money that left with them. They diverge whenever your customers are different sizes, and the gap is informative. Losing lots of small accounts gives high customer churn and mild revenue churn. Losing one large account does the reverse, and is the more dangerous of the two.
What is negative churn?
Negative churn is when expansion revenue from your existing customers is larger than the revenue you lose to cancellations and downgrades, so the base grows even with no new sales. It is measured as net revenue retention above 100%. Best-in-class B2B SaaS sits at 120% or more. It usually comes from seat-based or usage-based pricing rather than from customers being happier.
What causes customer churn?
Most churn traces back to a customer never reaching the point where the product became useful to them. The common drivers are a weak onboarding, a single champion who left the account, falling usage over the months before the cancellation, repeated support problems, and failed payments. That last one is bigger than people expect — involuntary churn from expired cards can be a quarter of all cancellations, and it is the cheapest kind to fix.
How much historical data do you need to predict churn?
Enough churned customers, not just enough customers. A usable model needs a few hundred rows and at least 30-50 examples in the churned group; below that it learns noise and the accuracy it reports back is fiction. If only 2% of your base has ever cancelled, you need a much bigger file to get those examples. Twelve months of history is a reasonable target because it covers a full renewal cycle.
How accurate can a churn prediction model be?
Judge it on AUC rather than accuracy, and expect 0.70-0.85 on real customer data. Plain accuracy is misleading here: if 5% of customers churn, a model that predicts “nobody churns” is 95% accurate and worthless. A model scoring well above 0.90 is usually a warning sign of leakage, meaning a column snuck in that already encodes the cancellation, such as a refund date or a closed-account flag.
The version that follows up
This page does one thing with one file, in your browser, and then stops. It cannot look at anything that is not already a column on the row, and it cannot answer the question you will have thirty seconds after reading the list.
For the follow-up questions, run the same analysis inside your AI assistant with meelu-analytics-mcp. It reads the files off your own disk, joins your customers to your orders to your support history, and answers the next question in the same breath as the first. Free, open source, and local for the same reason this page is. See the churn prediction use case.
The other free tools on the same customer file
A risk score tells you who to ring. These answer the next three questions from the same export, and all of them run in your browser too.
- Cohort analysis — whether the customers arriving now hold up better than the ones who arrived last year, which is the churn question asked backwards.
- RFM analysis — segments by recency, frequency and spend, for when you want groups to treat differently rather than a ranked list.
- Customer LTV calculator — what a saved account is actually worth, which is the number that decides how much a retention offer can cost.
All of them are pieces of Meelu, a desktop app where an AI marketing agent runs your marketing on your own machine. Join the waitlist.