Est. reading time: 7 minutes
Most accounts that come to us are optimizing for purchases, and on paper it works. Cost per purchase falls, volume climbs, the dashboard is green. Then we segment the revenue and find the campaigns are systematically acquiring the cheapest buyers in the catalog. One-time discount purchasers, low order values, no repeat behavior. The account is winning the metric and losing the business.
That happens because Meta delivers exactly what you ask for. Optimize for purchase count and every buyer is worth one point, whether they spend $18 once or $240 a quarter. Value-based lookalikes exist to fix that mismatch. They let you hand Meta the difference between your customers instead of just the list of them, so the model goes looking for the expensive kind.
What changes when you add value to the seed
A standard lookalike treats every person in the source audience as equal. Meta profiles the shared traits across the list and ranks the wider population by similarity to that composite. The composite is an average, and if your customer base spans a wide value range, it’s an average of people you’d pay very different amounts to acquire.
A value-based lookalike takes the same source with one addition, a value assigned to each customer. Meta weights the profile toward the high-value end of the list, so the pattern it models is “what do my best customers have in common” rather than “what do all my customers have in common.” The output is an audience ranked by resemblance to the people who actually drive your revenue.
The mechanics are otherwise unchanged. The source still needs to meet Meta’s minimum of 100 people from a single country, the percentage still controls how deep into the similarity ranking the audience reaches, and the audience still feeds delivery the same way. The value column is the only new input, which is convenient, because it’s also the input most brands get wrong.
Where the value data goes wrong
The tool is only as honest as the numbers in it, and we see three recurring failures in the value layer.
The value column is a proxy for the wrong thing
Brands frequently upload last order value as the customer’s value. That trains Meta to find people who resemble your biggest single transactions, which is a different population from your most valuable customers. A buyer with one $200 order and a buyer with six $45 orders are not the same acquisition target, and only one of them compounds.
Use lifetime value, or the closest approximation your data supports, such as total historical spend per customer. Any ESP or platform that tracks purchase history can produce this. Klaviyo and Shopify both hold it natively, and if your list lives in ActiveCampaign workflows, the same purchase data can drive the export. The point is that the value column should encode the customer relationship, not the receipt.
The pixel isn’t sending purchase values at all
Value-based approaches also depend on your conversion events carrying value and currency parameters. A purchase event that fires without a value tells Meta a conversion happened but not what it was worth, which starves every value-weighted feature in the account at once. This is a tracking configuration problem before it’s a targeting problem, and it’s one of the first things we check in an audit. If events are firing without values, fix that before building anything on top of it.
The values are stale
A value column exported eighteen months ago describes who your best customers were, filtered through whatever you were selling and promoting then. Customer value distributions shift with catalog changes, pricing changes, and acquisition mix. Rebuild the source on a schedule, and date the exports so nobody has to guess.
Pair the audience with value optimization, or leave money on the table
A value-based lookalike changes who Meta looks for. It doesn’t change what Meta optimizes for once the campaign runs. If the campaign objective is still set to maximize purchase volume, delivery will drift back toward the cheapest conversions inside your value-weighted audience, and you’ll have blunted your own tool.
The complement is value optimization at the campaign level, where Meta bids to maximize total purchase value rather than purchase count. Eligibility depends on your account sending enough valued purchase events, which is the same plumbing as above, and the learning phase still applies. An ad set needs roughly 50 conversion events per week to exit learning, and value optimization concentrates spend on fewer, higher-value conversions, so budget and volume have to support that threshold. This is one of the places where reading the right numbers matters, and we broke down which ones deserve your attention in the Meta metrics that actually drive optimizations. For value-based campaigns, the headline metric shifts from cost per purchase to ROAS and average order value, because a rising cost per purchase can be the system working correctly.
Judged together, the audience and the optimization goal form one instruction. Find people like my best customers, and spend toward the highest-value outcomes among them. Judged separately, each one gets undermined by the other’s default.
The feedback loop that makes it compound
Everything above describes the setup. The returns come from the loop that follows, because value-based targeting improves with every cycle of accurate data you feed it. We covered the general principle in how smarter inputs train Meta’s system faster, and the value-based version of the loop runs in four steps.
- Send valued purchase events continuously, so the system learns from what customers actually spend rather than from a static upload.
- Refresh the seed from live customer data, so newly acquired high-value customers make it into the model.
- Read results on value metrics over a window long enough to include repeat purchases, since the whole premise is that the second order matters.
- Cut the segments that produce volume without value, even when their cost per purchase looks better.
That last step is where discipline gets tested. Value-based campaigns will often lose the cost-per-purchase comparison against a volume-optimized campaign in the same account, and someone will propose shifting budget to the “cheaper” one. The comparison is wrong. The campaigns are buying different things, and the only fair scoreboard is revenue per dollar over a full customer window.
Who this is for, and who should wait
Value-based lookalikes reward accounts with real value variance and enough data to express it. If your order values sit in a narrow band and repeat purchase is rare, the value column adds little, and standard seeds built from your full buyer list will perform about the same. If your customers range from one cheap order to years of repeat spending, the variance is exactly the signal this tool exists to use, and running volume-optimized campaigns in that account means paying the same price for both kinds of customer.
Brands earlier in the curve should get the plumbing right first. Valued purchase events, a clean customer list with purchase history, and a working definition of what a good customer costs and returns. Those assets pay off in every part of the account, and value-based targeting is simply the first place they compound. When the account is ready for that shift and you want a second set of eyes on the setup, we do that work.
