Your Best Customers Are an Audience and Creative Asset

Many accounts feed their prospecting engine the wrong fuel. The default move is to build a lookalike from “all purchasers” or “all website converters,” hand it to the delivery system, and assume scale will sort out quality. It won’t. A seed list is a definition of who you want more of, and if that definition is dominated by one-time, discount-driven buyers, you are paying to find more people exactly like the customers who erode your margin. Your best customers are not just a retention asset. They are the single most undervalued input to both your targeting and your creative — and almost nobody treats them that way.

For the adjacent growth decisions, compare Pre-Peak Meta Ads: Building Demand Ahead of Your Evidenced Ramp and then use Repeat-Purchase Rate: Why Your Only Benchmark Is Your Own Cohorts to pressure-test the operating plan.

The seed is a value judgment, not a list

A lookalike is only as good as the behavior it is asked to imitate. When you seed from “everyone who bought,” you hand the model a blended portrait: the bargain hunter who bought once at 40% off and never returned sits in the same pool as the buyer who came back three times at full price. The algorithm has no way to know one is worth a multiple of the other. It optimizes toward the statistical center of whatever you give it, so a polluted seed quietly teaches the system to chase cheap, shallow demand.

The fix is to stop treating “purchaser” as the unit of value and start treating contribution margin over time as the unit. A high value customer lookalike audience is built from the cohort that actually carries your economics: buyers who repurchase, who buy at or near full price, and who do it without a coupon nudging every order. That is a fundamentally different seed than “people who triggered a purchase event,” and it produces a fundamentally different prospecting pool.

Repeat behavior is the only honest signal

Every other proxy for “good customer” can be faked by a discount. Average order value spikes when you bundle a promo. First-order ROAS looks healthy because the offer did the selling. Even a one-time large order tells you almost nothing — it might be a gift, a one-off, or a deal-driven impulse that never recurs.

Repeat purchase at full price is the one behavior that is expensive to fake and hard to accident into. It means the product delivered, the buyer chose to come back, and they did it without being bribed. That is the closest thing to a clean signal of intrinsic value you will get from your own data. So the qualifying behaviors for your seed should be concrete and behavioral, not aspirational:

• Repeat rate — bought two or more times, ideally within a defined window so you are capturing momentum, not ancient history.
• Full-price share — a high proportion of their orders carry no discount code, or a discount below some threshold you set.
• Margin contribution — they sit in the top tier of cumulative contribution after returns, shipping, and discounts are netted out, not top of gross revenue.
• Low refund and return drag — value is net of the costs they impose, and serial returners can look great on revenue and terrible on contribution.

Note what is missing: gross revenue rank alone. A customer who placed one enormous discounted order can out-rank a quiet, loyal, full-price repeat buyer on a revenue leaderboard while being worth far less to the business. Rank on contribution, not on top line.

Building the seed without fooling yourself

You do not need a perfect customer-data platform to do this. You need a defensible definition and the discipline to apply it.

1. Define the cohort in plain language first. Write the rule before you query it: “two or more orders, majority full-price, top tier of net contribution, returns under a set ceiling.” If you can’t state it in a sentence, you don’t understand it yet.
2. Net out the costs. Strip discounts, refunds, returns, and shipping subsidies before you rank. The whole point is to escape the revenue illusion.
3. Watch the size floor, not just the size ceiling. A seed needs enough recent, qualified members to be statistically usable. If your true high-value cohort is tiny, widen the definition deliberately (loosen the repeat or margin threshold a notch) rather than polluting it with discount buyers to hit a count. Document the trade-off you made.
4. Keep it fresh. Value cohorts drift as the product line and price architecture change. A seed assembled and frozen a year ago is describing a customer who may no longer exist. Refresh on a cadence.

Then let the delivery system do what it is genuinely good at: finding more people who resemble that seed. You are not micromanaging the targeting — you are giving the machine an honest definition of “good” and getting out of the way.

The same cohort is your best creative brief

Here is the part many teams miss. The people who clear your highest-value bar are not only the best audience seed — they are the best creative seed too. They have already told you, through behavior and frequently through words, why the product was worth coming back for at full price.

That is the testimonial that converts, because it is anchored in repeat conviction rather than first-purchase novelty. Discount-driven reviews can praise the deal. Full-price repeat buyers praise the product, the result, the reason they returned. Mine that group specifically:

• Pull testimonials, reviews, and support threads from full-price repeat buyers, not from your general review pile.
• Build creative around the reason for return — the moment the product proved itself — not around the discount that triggered the first order.
• Use their actual language. The phrases a loyal buyer uses to justify a repeat purchase are the angles your prospecting creative should test.

This closes a loop. The same cohort defines who you target and what you say to them. Your highest-value customers describe the value in their own words, that language becomes the creative, and the lookalike points it at people who resemble the people who already believe it. Targeting and message stop fighting each other.

What to watch once it’s live

Two cautions keep this honest.

First, judge the new prospecting pool on forward behavior — repeat rate and contribution of the customers it acquires — not on day-one ROAS. A higher-value audience may show a slightly heavier upfront acquisition cost and a far better contribution curve over the following cycles. If you grade it on first-click ROAS, you will kill the exact thing you were trying to build. Give the delivery system enough recent conversion signal to settle before you draw conclusions; new audiences need a stabilization window, and the early read is noisy by nature.

Second, treat any benchmark as a planning range, not a promise. The thresholds you pick for “full price,” “repeat,” and “top tier” are judgment calls about your economics — there is no universal number, and a clean rule you can defend beats a borrowed one you can’t.

This is the kind of analysis Bach is built to surface — separating contribution-positive cohorts from revenue mirages before a single targeting change is proposed, and never acting until you approve it.

The takeaway

Stop seeding prospecting from “everyone who bought.” Build your lookalike from the buyers who repurchase at full price and carry real contribution margin, because repeat behavior is the only value signal a discount can’t counterfeit. Then pull your testimonial creative from that same cohort, so your targeting and your message describe the same person. Your best customers are not just revenue you already earned — they are the most honest brief you will ever get for the revenue you haven’t.