Learn from past campaigns
Upload past results in any format or connect your tools, so Geofy.ai starts your first campaign from what already worked for you.
Geofy.ai is smart from the first campaign and keeps learning while campaigns run. Your own history makes the start smarter: open Learn from past campaigns (/learning).
Add your history
- Upload files: any export or report, in any format. CSV and Excel sheets, JSON, PDF reports, screenshots and even a pasted note. Geofy.ai finds the header row, the columns (spend, impressions, clicks, conversions, installs, leads, revenue, dates, campaigns, channels, platforms, countries) and the currency by itself, in English, Hebrew and other languages. Totals rows are skipped. Unusual layouts, PDFs and images are read by Geofy.ai.
- Connect your tools: analytics and app attribution tools you connect under Settings → Measurement can pull their history. Google Ads and Meta connect in one click, read only: sign in, and every ad account the sign-in can see can pull its past results. Other ad accounts show how to export a report.
- Campaigns you ran here are added with one button.
The best file has the last 6 to 12 months by day and campaign, with spend and conversions. The same file uploaded twice is read once. Remove a file and what was learned from it at any time.
What Geofy.ai learns
- Your cost per result overall and by channel, platform and country, with small samples pulled toward your average so a lucky week is not over-read.
- What worked best and what cost too much, your cheapest weekday, and whether costs were improving.
- How well Geofy.ai knows your results: a score that grows with more results, more months and more sources.
How it is used
- The campaign builder fills in your target and channel mix from your history, and explains why. You can change anything.
- The optimizer starts from your own costs instead of a guess, then keeps learning from live results.
- Sample data is shown but never used to steer campaigns.
Watch a campaign learn
Every campaign has a What Geofy.ai learned button on its page (and in the campaigns list). It opens one view with:
- the learning status (learning, learned, or learning slowly when there are too few results for the budget) and how far along it is;
- Learned from this campaign so far: how the cost per result stands against your target, which channels, countries and creatives do better or worse than the campaign average, whether the campaign confirms or differs from your history, and how sure Geofy.ai is of each line;
- Before it started: what your past results said about this campaign's goal, channels and countries;
- What Geofy.ai did with it: the latest changes and the reason for each.
At the top, a short summary in your language and a chart of the cost per result so far, day by day, against your target. Lines that appeared since your last visit are marked New. The view refreshes by itself while it is open, and Learn from past campaigns lists your latest campaigns with their learning status.
Lifetime value
Lifetime value (/learning/ltv) groups the people your measurement reports by the week of their first install (or first purchase) and shows, for each week, what they spent on average by day 0, 1, 3, 7, 14 and 30 and the share who came back on those days. From the groups that are already 30 days old Geofy.ai learns how first-week revenue grows into 30-day revenue, and predicts the 30-day value of newer groups and of a new person from each campaign. Until two of your weekly groups are 30 days old it predicts nothing: it never fills the gap with an industry rule of thumb. It needs purchase or install events with their value from your pixel or attribution tool.
The page also draws what an average new person has spent by each day, and for each campaign shows what a new person cost, how many times that they are worth, and the day they pay it back. The grey line on the chart is that cost: where the spend line crosses it, new people have paid for themselves. Tap a campaign to see only the people it brought.
All your channels
All your channels (/learning/channels) puts your campaigns here next to what you spend elsewhere: upload a file with a date, channel, spend and conversions column, or connect Google Ads or Meta. You see the total cost per result across every channel and each channel's share of spend and results. With enough weeks of data, a simple media-mix model compares weekly spend on each channel with weekly results and estimates how many results each channel really brought, with a likely range, the cost per extra result, how long its effect lasts, and how many results would have come anyway. It is an estimate, not an experiment, and gets sharper with more weeks and when spend changes.
When the model is fairly sure one channel brings results much more cheaply than another, Where to move budget suggests moving a fifth of the costlier channel's weekly spend to the cheaper one, with how many extra results a week to expect. It keeps the move small because the model assumes extra spend works as well as the average, which holds only near what you already spend.
Every number on these pages shows where it came from: Your file for what you uploaded, From Google Ads (or another tool) for what was read from your own connected account, Measured for what your campaigns here delivered, and Estimate for predictions and the media-mix model, with how it was worked out when you hover. It is your account's data only and is never mixed with other accounts.
Each file you upload has How Geofy.ai read it: which column became spend, conversions, dates and so on. If a column was read wrong, rename it in the file and upload it again.
Related
- Create your first campaignThe three steps from goal to launch, and what Geofy.ai fills in for you.
- InsightsAlerts, opportunities and campaign ideas from Geofy.ai, each with its impact and one button to act.
- Geofy.ai notes and undoWhat Geofy.ai changes on its own, how it explains it, and how to reverse it.