Did a launch bring extra conversions, and how sure can you be?
A launch is one thing you put into the world: a podcast ad, a webinar, a newsletter sponsorship, a press hit, a TV spot. Open Attribution, click Lift and then New launch. Give it a name, a channel, a start date, an optional end date and the conversion event to measure (your main event by default). GeoSwap compares the days after the launch with the 8 weeks before it.
Each launch shows two kinds of numbers, side by side and labelled.
utm_campaign, inside the launch's window. A conversion that came through both is counted once. Link up to 20 existing short links, or create a new link and QR code from the launch form. This number is a floor: people who heard the ad and typed your address aren't in it.If more conversions came through the launch's links than the launch seems to have added, the page says so: some of those people would likely have converted anyway (a returning customer who used the podcast link, say).
The window runs from the start date through 13 days after the last day (the end date, or the start date when there is none), so late responses count. A one-day webinar is measured over 14 days; a four-week sponsorship over 41 days. The end date can be at most 42 days after the start, so a window is at most 8 weeks.
| Status | When | What you see |
|---|---|---|
| Scheduled | Before the start date | The setup and what the launch can detect. |
| Collecting | During the window | The exact direct evidence, “Day n of N” and neutral “So far” numbers. No verdict yet. |
| Measuring | For 3 days after the window | Finalizing: late conversions and Google data are still arriving. |
| Measured | After that | The final ranges. They are updated daily for 14 more days to pick up late API conversions, then they stop changing. |
Once the start date has passed, the start date and the conversion event are locked, so the result can't be tuned after the fact. The end date can change until the window ends; the name, cost and linked evidence can change any time.
An estimate reads like “+60 extra orders (40–80)”: the launch most likely added about 60, and the true number is likely between 40 and 80 (a 90% range). When the range includes zero, the page says No clear lift and what it rules out as unlikely, like “Lifts above +45 are unlikely”. That is still an answer: a small launch that couldn't have moved much is worth knowing about too.
Each number needs at least 4 weeks of data before the start date and uses up to 8. History starts on your workspace's first day of data for that number: your first conversion of the event, or your first tracked page visit for brand visits. With fewer than 8 weeks the page calls it an early estimate: a trend in your data can't be told apart from the launch yet.
Too little history? The number shows “Results available after <date> — tracking only started on <date>” and the rest of the launch still works. To measure sooner, send past conversions through the API with their real occurred_at (see Backfilling history). Send complete days only: a day with some of its conversions missing reads as a quiet day.
A number with too few conversions to measure says so, too. We never block a launch: we tell you when the numbers can't answer.
While you fill in the form, GeoSwap lists what each window length could detect from your own history, like “2 weeks · 80% chance to detect ≥ 25%”, for 2, 3, 4, 6, 8 weeks, with your window highlighted. A quiet event can only detect big lifts; when even 8 weeks can only catch a large one, the form says to expect a wide range. It is advice, not a gate: you can save any launch.
Many people who hear about a launch never click a link: they search your name or type your address. Two more numbers catch them.
Launches appear as flags on the Analytics overview chart: a dashed line at the start and a faint band over the window. Click a flag for the launch's name, channel and dates, and See lift to open it. When many launches fall in the range, the later ones collapse into one “+n launches” flag.
Overview, Channels, Journeys and looking up a conversion, page by page.