First, confirm the leak
Two sources of evidence should agree before anything changes. Inside Ads Manager, where the campaign type offers it, the audience segment breakdown splits spend and results among the people you defined as existing customers, engaged audiences and new audiences; read it over a few weeks, not a few days. Outside it, the shop or CRM can mark each order as from a new or returning customer, and the share of returning orders on days when Meta prospecting carried more spend is a rough cross-check.
If both show a large share of spend and purchases among existing customers, the leak is real. If the breakdown says customers but the shop says new buyers, the customer definition is probably wrong. If neither shows much, the complaint may come from a single sighting of an ad by a colleague who is also a customer. The mechanics of how Advantage+ audience treats each input are in the Advantage+ audiences guide; this page uses them only to locate the leak.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Segment breakdown shows a large customer share in prospecting | Audience controls section for excluded custom audiences | No customer or recent-purchaser exclusion where exclusions are allowed | Add a customer list and purchase-event audience as exclusions |
| Leak began after audience suggestions were added | Suggestions section for included custom audiences or buyer lookalikes | Suggestions built from past buyers or engagers | Remove buyer-based suggestions from prospecting, or rely on broad delivery |
| Exclusion is set but customers still appear in the shop's returning orders | Exclusion audience size and last refresh date | Customer list stale or poorly matched | Refresh the list with more identifiers and a purchase-event audience |
| Returning orders dominate while reported return looks excellent | Optimisation event, offer and creative message | Goal and creative rewarding easy repeat purchases | Use a new-customer goal where offered and creative aimed at first-time buyers |
| Breakdown labels many people as engaged and the team reads it as customers | Customer definition in the account settings; shop's new versus returning split | Breakdown read wrongly | Define existing customers properly and read the segments separately |
Fix, verify and prevent each cause
No exclusion where one is allowed
Meta documents excluded custom audiences as a control, the one audience input that stays hard under Advantage+ audience. An ad set without one has nothing stopping it from reaching buyers.
Fix: add a current customer list and a website purchase audience as exclusions on prospecting ad sets where the campaign type permits.
Verify: the customer segment's share of spend falls over the next few weeks, and returning orders attributed to prospecting fall in the shop's data.
Prevent: make the converter exclusion part of the prospecting template, using the rules in the exclusions and overlap guide.
Suggestions seeded with buyers
Included custom audiences and lookalikes are only starting points under Advantage+ audience, but they are where delivery begins. A prospecting ad set whose suggestions include purchasers, engagers or a lookalike of buyers will start among people close to existing customers.
Fix: remove buyer and engager audiences from prospecting suggestions; keep them for RemarketingAdvertising to previous visitors or engagers. ad sets that are meant to reach them.
Verify: the segment breakdown shows a larger new-audience share.
Prevent: keep a written rule for which inputs belong in which ad set, as discussed in the suggestions section of the Advantage+ audiences guide.
A stale or thin exclusion list
An exclusion is only as complete as the audience behind it. A list uploaded once, or one that matches a small share of customers, lets most buyers through.
Fix: replace the list with a fresh export carrying every identifier you lawfully hold, and pair it with a purchase-event audience over a long window.
Verify: the exclusion audience size grows, and returning orders from prospecting fall.
Prevent: schedule the refresh; the custom audiences guide explains how freshness and matching limit lists, and the custom audience too small page covers lists that will not grow.
Goal and creative favouring repeat buyers
If the campaign optimises for purchases and existing customers buy more readily, delivery will find them, because nothing in the goal says a first purchase is worth more. Creative full of loyalty references or product-update news makes it worse.
Fix: where the campaign type offers a new-customer goal or setting, use it, and brief creative that explains the product to someone who has never bought.
Verify: new-customer share of purchases rises in the shop's data, and cost per new customer is reported separately.
Prevent: set objectives from what the business values, using the campaign objectives guide to match goal to outcome.
Engaged people read as customers
The segment breakdown depends on the audiences you defined. Engaged people, such as page followers or video viewers, are not necessarily buyers, and without a customer definition, buyers can sit in the engaged or new rows.
Fix: set the existing customer definition from current lists and purchase events, then read each segment separately.
Verify: the breakdown and the shop's new versus returning split tell the same story.
Prevent: report prospecting results on the same window and definitions every month, as recommended in the attribution settings guide.
Prevention: new-customer guardrails
The leak is easier to prevent than to unwind, because a prospecting ad set that has learned to find cheap repeat buyers keeps finding them. Three guardrails keep acquisition honest.
The first is a metric. Report customer acquisition cost (CACCost to acquire one new customer.) for prospecting from the shop's new-customer orders, alongside platform return on ad spend (ROASRevenue generated per unit of ad spend.), and treat a rising platform return with flat new-customer numbers as a warning rather than a win. The ROAS calculator helps recompute return from new-customer revenue only.
The second is a set of standing exclusions. Every prospecting ad set inherits the same customer list and purchase audience, refreshed on a fixed schedule, with an owner. When a new product line launches, decide explicitly whether buyers of other lines should be excluded or welcomed as cross-sell prospects, rather than letting the default decide. The same customer file can feed Google Ads, where the guide to excluding existing customers from prospecting sets the exclusions by campaign type, including Performance Max's customer acquisition goal.
The third is a monthly read of the segment breakdown and the shop's split, recorded so that drift is visible. If the customer share creeps up after a creative refresh or a new suggestion, the cause is usually that change. Where the business wants repeat purchases, give them their own retention ad set with its own budget, so the question of who should see which ads is answered by structure rather than by delivery.
A coffee subscription brand paying to reach subscribers
A coffee subscription brand runs prospecting with Advantage+ audience at £150 a day and reports a ROAS of 3.8. Finance notes that new subscriptions have been flat for two months. The segment breakdown over 28 days shows 46% of prospecting purchases among existing customers. The ad set has no exclusions; its suggestions include a lookalike of all buyers and a list of newsletter subscribers. The team adds the active-subscriber list and a 180-day purchase audience as exclusions, removes the newsletter suggestion, and swaps a "new roast this month" ad for one aimed at first-time buyers. Over the next 28 days reported ROAS falls to 2.6, but the existing-customer share drops to 11% and new subscriptions rise by 38%.
Where the leak persists and a defined boundary looks safer, the comparison of lookalike audiences and Advantage+ audience shows how each handles existing customers and when to switch.
Common mistakes
Putting the customer list in the suggestions instead of the exclusions. Excluding all buyers from every ad set in a multi-line business. Celebrating a higher return that comes from repeat buyers. Judging the leak from one colleague's feed. Uploading an exclusion list once and never refreshing it.
