Custom Audiences for eCommerce: How to Build Them
- A custom audience is defined by something a person did, which is what separates it from interest targeting built on what a platform guesses about them.
- Meta accepts five Custom Audience sources: customer list, website traffic, app activity, engagement, and offline activity. Google Customer Match accepts contact data only.
- Six audiences cover almost every eCommerce use case: subscribers, first-time buyers, quiet one-time buyers, repeat customers, lapsed customers, and top-value customers.
- RFM segmentation is the shortest path from raw order data to an ad audience, because each segment already implies the offer it should receive.
- Since Meta Advantage+ and Google Performance Max automate targeting, a custom audience is now mainly a seed, a suggestion, or an exclusion rather than a hard targeting rule.
A custom audience is a group of people you define with your own data and then target with ads, instead of describing them with the ad platform's interest and demographic filters. If you sell walkers for the elderly, interest targeting guesses who might need one. A custom audience knows: these are the people who bought one, asked about one, or subscribed to your list after reading about one.
That difference is the whole point. Interest targeting is built on what a platform infers about a person. A custom audience is built on something the person did, in a system you own. This guide covers the data sources Meta and Google accept, the six audiences almost every eCommerce store should maintain, how to build them from RFM segments, and what has changed in the last three years about how ad platforms actually use them.
What are custom audiences?
The naming differs by platform but the mechanism does not. You supply identifiers or events. The platform matches them against its own users. What comes back is an addressable group, plus a match rate that tells you how many of your records the platform recognized.
Two things follow from that. First, the audience is only as good as the rule that produced it. "Everyone who ever bought" is a list, not an audience, because it contains people with nothing in common except a transaction. Second, the audience decays. A customer who was in your "one-time buyer" list in March may have bought twice by June, and an audience that does not refresh will keep showing them the wrong ad.
Which data sources can you build custom audiences from?
Meta's five sources are worth understanding individually, because they produce audiences of very different quality:
- Customer list. You upload emails, phone numbers, and other identifiers from your CRM or store. This is the highest-intent source, because every record represents a real relationship rather than a session. Colorado State University's social media team has a practical guide to formatting a customer file for a Facebook custom audience.
- Website traffic. Built from Meta Pixel events. You can combine event type, time window, and frequency: people who viewed a product but did not add to cart in the last 14 days, people who visited five times in the last week. Server-side event forwarding through the Conversions API improves how much of this signal survives browser and app restrictions.
- App activity. In-app behavior captured by the Meta SDK: installs, opens, in-app purchases. Useful when the app is where loyalty actually lives.
- Engagement. People who interacted with your Facebook or Instagram content, including video views, lead form opens, and profile visits. Retention windows differ by type, so an audience built on lead form engagement covers a much shorter period than one built on Page engagement.
- Offline activity. In-store purchases, phone orders, and other events you import. This is how a store with physical locations stops treating an in-store buyer as a prospect.
Video engagement is the clearest example of why these sources work. If people watched most of a product video, they have already spent attention on you. A second video with a different call to action reaches a group that has demonstrated interest, not a group that matches an interest label.
Google Customer Match is narrower on purpose. It takes only contact information you already hold, matched against signed-in Google accounts, and runs across Search, the Shopping tab, YouTube, Gmail, and Display. Google requires at least 100 matched members for a list to be usable, and list membership expires after 540 days without an update, which is another reason to automate the refresh rather than upload a CSV once.
The 6 best custom audiences for eCommerce
These are ordered by where the customer sits in the relationship, not by importance. Every one of them is defined by a rule you can run against your own order data today.
1. Subscribers who have never bought
Email subscribers and leads with zero transactions. The job is the first purchase. Email alone rarely gets there, because most of what you send is never opened, so a paid layer against the same list closes part of the gap. Split it further if the list is large: brand-new subscribers, subscribers who open and click, and subscribers who have gone quiet all need different creative.
2. First-time buyers
People whose first order landed recently, typically in the last 30 days. The job is the second purchase, which is the single most important transition in eCommerce because it is where a buyer becomes a customer. Exclude them from acquisition campaigns and show them replenishment, complementary products, or the part of your brand story a first order never covers.
3. One-time buyers who have gone quiet
Exactly one order, placed long enough ago that a second one is overdue by your normal purchase cycle. Filter on a window that matches your category, for example 180 days for a consumable and much longer for furniture. This audience is where Customer Lifetime Value is won or lost, and it is usually the largest untapped group in the account.
4. Repeat customers
Two or more orders. They have already trusted you more than once, and the fastest way to lose that is to show them an ad that says "20 percent off for new customers only." Advertise to them as people you recognize: early access, restocks, the next tier of the product line. Recognition outperforms discounting here, and it does not erode margin.
5. Lapsed customers
Customers whose last order is now well beyond two normal purchase cycles. Something stopped them. Win-back campaigns work on part of this group, but the recovery rate falls the longer you wait, which is why the "quiet one-time buyer" audience above matters more than this one. Budget accordingly and do not chase people who have simply left the category.
6. Highest-value customers
Rank customers three ways, by total amount spent, by number of orders, and by average order value, then take the top of each. The overlap is your VIP audience. Use it twice: as a targeting audience for cross-sell and early access, and as the seed for a value-based lookalike, which is where a small, clean, high-value list beats a large, mixed one. Seed quality is what a lookalike is actually made of.
How RFM segments become custom audiences
The six audiences above are really six RFM rules written in plain English, with one exception: subscribers who never bought have no purchase history to score, so they sit outside RFM and are built from your list instead. That overlap is not a coincidence. RFM segmentation exists precisely to turn order history into groups that share a next action, which is the same thing an ad audience needs to be useful.
| RFM segment | What the scores say | Audience to build | How to read the result |
|---|---|---|---|
| Soulmates | High recency, high frequency, high spend | VIP targeting audience, plus the lookalike seed | Judge on repeat revenue and lookalike quality, not on cost per click |
| Apprentices | Recent, but only one order | Second-purchase audience | Watch the share of the cohort reaching order two, not the immediate ROAS |
| About-to-dump-you | Recency falling after strong frequency or spend | Win-back audience, highest priority | Measure reactivation rate against a holdout that receives nothing |
| Break-ups | Very low recency, prolonged inactivity | Low-budget reactivation, or an exclusion | Compare recovery cost against the cost of acquiring a new customer |
| Lovers | Recent and frequent, mid-range spend | Repeat-customer audience, excluded from discounts | Watch order frequency and AOV, and keep them out of new-customer offers |
The advantage of driving audiences from RFM rather than from a manual export is that membership updates itself. A customer who places a second order leaves the second-purchase audience and enters the repeat audience on their own, which is exactly the maintenance work nobody does by hand.
Score every customer on recency, frequency, and monetary value, then push the segments straight into Meta Ads and Google Ads.
Learn more about Customer Intelligence in Nexus →How to build an eCommerce custom audience, step by step
-
Write the rule before you touch the ad platformOne sentence, in your own data's terms: "customers with exactly one order, placed between 180 and 540 days ago." If you cannot write the rule, the audience will not be actionable no matter how many people it contains.
-
Check the rule produces enough peopleMeta and Google both need roughly 100 matched people as a floor, and both perform better well above it. If a segment is too small, widen the time window rather than loosening the behavior, because the behavior is what makes the audience worth targeting.
-
Export every identifier you are allowed to useEmail, phone number, first and last name, country. Match rates rise with each additional field, and a list that matches poorly wastes the segmentation work that went into it.
-
Confirm you have a lawful basisBoth platforms require that you have the right to use the data for advertising. Under GDPR and similar laws that usually means consent or a clearly disclosed existing customer relationship. Handle this before the upload, not after.
-
Upload, or connect a tool that syncsA manual upload is fine for a one-off test. For anything you intend to keep running, connect a segmentation platform so the audience updates on a schedule and members leave it when they no longer match.
-
Pair the audience with the matching offerThe audience only creates value if the creative changes with it. A win-back audience and a VIP audience should never see the same ad, and neither should see a new-customer discount.
-
Set the exclusionsExclusions are half the value. Exclude recent buyers from acquisition campaigns, exclude VIPs from discount campaigns, and exclude anyone already converted from the retargeting pool. This is the cheapest efficiency gain available in most accounts.
-
Test the offer, not just the audienceOnce the audience lands on your site, the landing experience decides what happens next. Run the offer, the message, and the page as an experiment rather than assuming the targeting did the work.
Meta Custom Audiences vs. Google Customer Match
| Meta Custom Audiences | Google Customer Match | |
|---|---|---|
| Data sources | Customer list, website traffic, app activity, engagement, offline activity | Contact data only: email, phone, mailing address, mobile device ID |
| Where the ads run | Facebook, Instagram, Messenger, Audience Network | Search, Shopping tab, YouTube, Gmail, Display |
| Minimum audience | 100 matched people, materially better above 1,000 | 100 matched members, refreshed within 540 days |
| Finding more like them | Lookalike audiences, including value-based lookalikes | Similar segments seeded from a Customer Match list |
| Best at | Creating demand among people who are not searching | Capturing demand at the moment of the search |
One change matters more than the feature comparison. On both platforms, automated campaign types now make most of the targeting decisions themselves. Meta's Advantage+ and Google's Performance Max treat your lists less as a hard boundary and more as input: a seed for a lookalike, an audience suggestion, or an exclusion. That has made custom audiences less of a manual lever and more of a data-quality question. The cleaner and better-segmented the list you hand over, the better the automation performs, which is an argument for building audiences from customer value rather than from raw volume. Our guide to Facebook ads for eCommerce covers how this plays out inside the campaign structure, and audience analysis covers the research that should precede any of it.
Privacy, cookies, and consent in 2026
The cookieless deadline that dominated marketing planning for years did not arrive in the form anyone expected. Third-party cookies remain in Chrome, with no removal timeline, and the replacement APIs were retired. That is not a return to 2019, though. Safari and Firefox still block third-party cookies by default, Apple's App Tracking Transparency still gates app-level identifiers behind an explicit prompt, and consent requirements under GDPR and comparable laws apply regardless of what any browser does.
The practical consequence for custom audiences is unchanged: the audiences you build from consented first-party data, matched server-side, are the ones that keep working. Which is another way of saying that the customer data you already own is not a fallback for a cookieless future. It is the primary asset, and it always was.
Tools that build eCommerce custom audiences
-
Customer intelligence
Automates RFM segmentation, cohort analysis, and CLV tracking on your order data, then syncs the resulting segments to Meta Ads, Google Ads, and your ESP as audiences that stay current. It also shows which products create repeat buyers and which ones create one-time buyers, which changes what you should be advertising in the first place.
-
Experimentation
Tests the landing experience your custom audiences arrive on: the offer, the message, and the page. A VIP audience and a win-back audience should not see the same page, and experimentation is how you find out what each one should see.
-
DeliveryYour ESP or marketing automation platform
The same segments should drive email and ads together, so a customer in the win-back segment gets one coordinated message rather than two contradictory ones. The segmentation platform defines the audience, the ESP and the ad account deliver it.
Whatever you use, the sequence is the same: segment on real behavior, sync rather than export, and change the creative with the audience. For the tactics that follow the audience, see how to make customers buy again and brands that get the customer experience right.
Frequently Asked Questions
A custom audience is a group of people you define with your own data and then target with ads, instead of describing them with the ad platform's interest and demographic filters. The definition comes from something the person actually did: they bought from you, subscribed to your list, visited a product page, watched a video, or opened your app. On Meta the feature is called Custom Audiences. On Google Ads the equivalent, built from a customer list, is called Customer Match.
Meta builds Custom Audiences from five source types: a customer list you upload, website traffic captured by the Meta Pixel, app activity captured by the Meta SDK, engagement with your Facebook or Instagram content, and offline activity such as in-store purchases. Google Customer Match uses contact information your customers gave you: email addresses, phone numbers, mailing addresses, and mobile device IDs. Both platforms require that you have the right to use that data for advertising.
The six audiences that earn their place in almost every eCommerce account are: non-buying subscribers, first-time buyers, one-time buyers who have gone quiet, repeat customers, lapsed customers, and your highest-value customers. Each one has a different job. Subscribers need a first purchase, first-time buyers need a second, repeat customers need to be recognized rather than discounted, and high-value customers are the seed you use to find more people like them.
Export the customer records that match one behavioral rule, such as exactly one order in the last 180 days. Include email, phone, and any other identifier you have permission to use, because match rates rise with every extra field. Upload the list to Meta or Google, or connect a segmentation tool that syncs the segment automatically. Then set a refresh cadence, because a list that is not updated goes stale as customers move between segments.
Meta Custom Audiences can be built from behavior Meta itself observes, such as video views, Instagram engagement, or pixel events, as well as from a list you upload. Google Customer Match is built only from contact data you already hold. The practical difference is intent: Google reaches people at the moment they search, while Meta reaches people who are not looking for anything, which makes Meta stronger for demand creation and Google stronger for demand capture.
RFM scores every customer on Recency, Frequency, and Monetary value, then groups them into segments with a shared next best action. Those segments map directly onto ad audiences: declining-recency customers become a win-back audience, one-time buyers become a second-purchase audience, and top-value customers become both a VIP audience and the seed for a value-based lookalike. Nexus by Omniconvert calculates the segments and syncs them to Meta Ads and Google Ads so the lists refresh themselves.
Yes, and for a different reason than before. Meta Advantage+ and Google Performance Max now decide much of the targeting themselves, so custom audiences are less often a hard targeting rule and more often a signal you hand the algorithm: a seed for a lookalike, an audience suggestion, or an exclusion. The quality of your first-party data therefore matters more than it did when you were choosing interests by hand.
They are, and audiences built from your own customer data are the most durable kind. Google confirmed in April 2025 that it will not remove third-party cookies from Chrome and retired the Privacy Sandbox APIs in October 2025, but browser tracking is still limited and Apple's App Tracking Transparency still restricts app-level signal. Consented first-party data, matched server-side, is the part of the stack you control. You still need a lawful basis to use it for advertising under GDPR and similar laws.
Do not build eleven audiences. Build two. Export the customers with exactly one order in the last 180 days and give them a second-purchase offer. Export the top 20 percent by revenue and use them twice: once as a VIP audience that never sees a new-customer discount, and once as the seed for a value-based lookalike. Those two lists usually change more in an ad account than a month of creative testing, because they replace guessed intent with recorded behavior. Once they work, add the remaining four and automate the refresh, because the value of a custom audience decays the moment the underlying segment stops being true.
Turn your customer data into ad audiences
Nexus by Omniconvert scores every customer on recency, frequency, and monetary value, builds the segments automatically, and syncs them to Meta Ads, Google Ads, and your ESP as audiences that refresh themselves. Built on 13 years of eCommerce data across 7,000+ websites and 15+ industries.