Customer Match lets you upload a list of your own customers to Google Ads and use it to include or exclude those people in campaigns. The list is only as good as the cohort definition behind it, and most stores get the definition wrong in the same few ways. This playbook covers how to define cohorts from order data, which ones to use for exclusion, and what Magistry handles for you today.
Start with exclusions
The cheapest win in Customer Match is usually subtractive. Prospecting campaigns that keep reaching people who already bought, or who just asked for their money back, spend acquisition budget on the wrong people. Three cohorts cover most of it:
- Recent refunders: every customer with a refunded order in the last 30 days. Keep them out of prospecting while the refund is fresh.
- Buyers of a product with a complaint cluster: customers who bought one specific product in the last 90 days, when support tickets show a pattern of complaints about it. Keep them out of prospecting for that product until the issue is fixed.
- High-value buyers: customers whose lifetime spend passes a threshold. Exclude them from first-purchase campaigns so you stop paying acquisition prices for repeat customers.
Magistry takes the high-value threshold from the VIP LTV threshold in the CS agent's settings, the same number that makes a high-value customer's ticket escalate in the helpdesk. One definition of a valuable customer, read by both.
Define cohorts from orders, carefully
A careless cohort teaches Google the wrong thing, because Google uses the list to find people like the ones on it. Magistry's cohort definitions follow these rules, and they are worth copying even if you build lists by hand:
- Test orders do not count. Uploading your own address as a customer tells Google to find more people like you.
- Deleted orders do not count.
- Refunded orders do not count toward lifetime value. A buyer who returned everything is not a high-value customer.
- Lifetime value is summed only in the store's own currency. Orders in another currency are left out instead of converted at a guessed rate.
- Orders imported from a previous shop stay visible for reporting but are never used to build a live audience.
Hashing, consent and membership
Google requires customer data to be normalised and hashed before upload. Magistry trims and lowercases each email and hashes it with SHA-256 before it leaves the system, and the lists it creates keep members for 540 days. You remain responsible for having consent to use customer data this way and for saying so in your privacy policy. For customers in the EEA, Google also asks for consent signals with the upload.
Where Magistry stands today
Magistry recomputes the 30-day refunder cohort every day and creates the matching Customer Match list in the store's Google Ads account. Uploading members to that list is on hold. Google no longer accepts Customer Match uploads from Magistry through the Google Ads API and points to its newer Data Manager API instead, and Magistry's audience upload has not moved to that API yet. Until it does, the lists exist but stay empty, and each refused upload is recorded as a skip with Google's reason instead of being retried. The complaint and high-value cohorts are defined the same way and have not been pushed to any account.
An upload is also a write to your ad account, so it only runs where the merchant has handed the Google Ads account to Magistry (Brain autopilot with the Google Ads domain on and a daily cap) or where a store member approves the change.
Exclude before you target. The people who just refunded and the people who already buy from you are the first two lists worth building.
Refresh and review
Cohorts go stale quickly: a refund window closes, a complaint gets fixed, a new buyer crosses the value threshold. Rebuild exclusion lists at least daily from order data, and look at the cohort size before each upload. A list that suddenly doubles or empties usually points to a data problem upstream.