Try it on your own numbers
No account, no integration, no access to anything of yours. Ask your own analytics tool for a small daily export, paste it in, and see what a forward view of your cohorts looks like — including how it scores against what actually happened.
Nothing you paste is uploaded. The forecast runs in this browser — there is no request to a server, so nothing to review with your security team before trying it. Close the tab and it is gone.
1. Get the data out of your own tools
Daily rates by cohort, roughly 90 days. No user-level rows, no identifiers, and no volumes at all if those are confidential — rates alone forecast just as well.
- 1.Ask Amplitude, or the Global Agent: "daily conversion rate from <first event> to <second event>, last 90 days, split by <country / platform / plan>".
- 2.Check it built a Segmentation chart with DAILY granularity and a breakdown — not a weekly total.
- 3.Open the chart, then use the export control and choose CSV.
- 4.Paste the CSV below. Column names do not matter; the parser looks for a date, a cohort and a number.
Or hand this prompt to an assistant over your warehouse
I need a small daily export from our product analytics. Please return data, not commentary. Pick our 3 to 5 most important user cohorts — split by country, platform, plan, or acquisition channel, whichever matters most to us. For each cohort, give me ONE conversion or retention rate per day. HOW MUCH HISTORY Last 90 days if you have it. 8 weeks is the minimum that produces anything useful, and under 2 weeks per cohort I would rather have nothing than a number I cannot trust. One row per cohort per day — do not roll it up to weekly. FORMAT — return ONLY this CSV and nothing else: date,cohort,numerator,denominator - date: YYYY-MM-DD - cohort: a short label, spelled identically every day - numerator: users in that cohort who completed the action that day - denominator: users in that cohort eligible to complete it that day IF USER COUNTS ARE SENSITIVE, send rates instead and skip the volumes entirely: date,cohort,rate where rate is a percentage between 0 and 100. The forecast works either way. Counts are only used to judge whether a cohort is big enough to be worth forecasting. WHAT NOT TO INCLUDE No user IDs, emails, names or any other identifier — I need daily totals only, never a row per person. If a cohort's own label is sensitive, rename it: "Cohort A", "Cohort B" and so on work perfectly well, and nothing downstream depends on what they are called. No revenue figures and no absolute business volumes beyond the denominators above, and not even those if they are confidential. Do not round to whole numbers, do not merge the cohorts together, and do not add totals or averages.
2. Upload the file
Whatever your tool exported — dates down the side or across the top, CSV or TSV. It is read here in the page and never sent anywhere.