What Remarketing Audiences Cannot Be Defined by Default

The short answer: three different kinds of 'can't'

Three separate problems get called 'can't,' and only one of them means the audience is truly impossible. The first is a real data gap — you're asking for a signal nobody collected, so there is nothing to define. The second is a size gap — the audience is defined correctly but has too few users to serve, so the platform holds it back. The third is an accuracy gap — the audience exists, serves, and reports numbers that don't mean what the interface implies.

Most 'why can't I build this' questions are actually the first problem: the platform never saw the behavior, because no tag, pixel, or event was firing when the user did the thing you want to target. Fix the collection, and the definition problem often disappears on its own.

What Google Ads and Analytics won't build without extra setup

A handful of audience types are unavailable the day you open a Google account, and they stay unavailable until something else is switched on:

• Demographics and interest categories — blocked until Google signals is enabled in Analytics.
• CRM and offline customer lists — blocked until you import data or set up Customer Match; there is no default source for emails or phone numbers you already hold.
• Cross-device stitched audiences — also need Google signals, since linking a phone session to a desktop purchase requires an identity graph the platform doesn't build automatically.
• Anything based on an event you never tagged — if 'added to wishlist' was never fired as an event, no audience can be built on it retroactively.

The oft-quoted minimum sizes — 100 cookies for the Display Network, 1,000 for Search — aren't a definition limit at all. You can define that audience with one user in it. The platform simply won't spend against it until it clears the floor, which is a serving rule, not a build rule.

The same three gaps inside Amazon DSP

Amazon DSP has its own version of each gap, and they matter more here because the spend behind them is usually larger. A pixel-based retargeting audience only ever includes activity from the day the Amazon Advertising tag actually started firing — it cannot be backdated to capture the month before installation, no matter how the lookback window is set. Amazon also enforces a minimum audience size before serving; the exact floor moves by audience type and marketplace, so check the console rather than assume a number.

The real gap is off-Amazon signal. A DSP audience built purely on-platform can't see a customer's newsletter click or a CRM record unless that data is matched through Amazon Marketing Cloud first. That isn't a missing feature — it's the deliberate boundary between what Amazon observed directly and what an advertiser brings in from outside.

Scale doesn't remove any of this. Across 30 advertisers in July 2026, the book delivered 6.04x return on ad spend, 78.4 million impressions at a $4.00 CPM and a blended $1.42 cost-per-click — and every one of those audiences still had to clear the same pixel-coverage and minimum-size checks a first-time advertiser hits on day one. Volume buys more data, faster. It doesn't buy an exemption from how an audience gets defined.

A worked example: the audience that looked broken and wasn't

Here's a mistake we've made ourselves. We launched a 30-day cart-abandoner audience the same week a client's pixel went live, expecting a full month of history to populate it immediately. The audience came back tiny — six days deep, not thirty — because the tag had only been firing for six days. The definition was correct. The data behind it didn't exist yet. The fix wasn't a setting change; it was waiting three and a half more weeks for the lookback window to fill.

That distinction — a wrong definition versus a right definition starved of history — is the most common support ticket in remarketing, on any platform. Across the wider book, blended cost per acquisition ran $5.49 across 57,137 attributed purchases, with 20.1% of them new to the brand. Audiences that looked 'broken' in week one were often just early. The ones still broken in week four had a real tagging problem underneath.

The number is wrong: a troubleshooting order, not a guess

Work through these in order before assuming the audience definition itself is at fault:

• Confirm the pixel or tag is actually firing — a broken tag looks identical to an empty audience.
• Check consent status; a user who declined cookies or ad tracking never enters the pool, by design.
• Check the lookback window against how long the tag has actually been live — you can't have thirty days of history from six days of tagging.
• Check the minimum size threshold before assuming the criteria are wrong.
• Check that the event you're filtering on is the event actually being sent, not a similarly-named one from an old implementation.
• Check you're looking at the right account or seat — audiences don't travel between accounts by default.

If all six check out and the number is still off, the definition is genuinely wrong — but that's the last thing to check, not the first.

Where this fits into a managed program

Dr. DSP is Fable 5's name for Amazon DSP — the demand-side platform for display, video and audio, not the Delivery Service Partner courier program — run as a managed product for advertisers who'd rather hand the pixel audits, lookback-window math and Amazon Marketing Cloud matching to someone who does it across many seats. Every audience change that goes live carries the evidence behind it, a measurement plan, and a rollback trigger before it runs. You don't need any of that to define an audience correctly today. You need it when the definition is right, the data isn't, and you want someone who's already seen that exact failure mode.