Parallel run

How it works

The term is also used for other system migrations: the old system keeps doing the real work while the new one runs beside it and is checked against it. In an MMP migration the old MMP stays the system of record and the new one only observes. The run starts when a build with both SDKs is released and ends at cutover. The practical steps, and what to compare, are in running two SDKs in parallel.

When it matters

How long to run depends on what rides on the numbers. A very small app with one channel and no purchase events can keep it short. A team whose campaigns bid on revenue or another in-app event should not skip it, because a counting error would feed straight into bidding. The run also has a cost: two vendors are billed for the same period and the app carries extra code. Set an end date at the start.

How it relates to attribution

A parallel run tests counting more than attribution. Ads still use the old links, so the new MMP sees installs without the clicks behind them and shows most as organic. What the run does give you is a ratio between old and new numbers, the only bridge across the break in your historical data.