- Value-based bidding optimizes toward the number you report. The platform never checks it. [1]
- Google’s own docs name “profit margins” as a thing the value field may hold. [3]
- Subtract cost of goods, variable costs, and expected returns. What is left is contribution margin.
- The same order is worth $100, $43, $39, or $49 depending on the definition, and each one needs a different target to mean identical profit.
- Refunds are corrected with a RESTATEMENT or a RETRACTION, matched on the order ID you stored at purchase. [4]
- A flat value is the common failure. With no variance, the strategy is conversion-count bidding wearing a costume. [2]
- Case evidence is real but thin: one bank, one retailer, both published by interested parties. [1][6]
A customer buys something for $100. The platform that found that customer is told the conversion was worth $100, because that is the number sitting in the order object when the tag fires. The warehouse pulls a product that cost $45. Payment processing and fulfillment take another $12. Roughly eight orders in a hundred come back, and processing each return costs about $6 on top of the sale it erases.
So the business kept about $39. The algorithm optimizing your account was told $100, and it has been told $100 on every order for as long as the account has existed. Every target you have ever set sits on top of that.
This is the part of Smart Bidding nobody automated. Targeting is gone, bid adjustments are mostly gone, and placements are a suggestion. The conversion value is the one input still entirely yours, and in most accounts it is the one input nobody has ever audited.
The value field is the strategy
Google is unusually direct about this. Its documentation on setting conversion values describes bids being set to maximize conversion value “such as sales revenue or profit margins,” and it treats a single fixed value per conversion action as a legitimate but blunt choice, warning it “may not fully capture the varying values of each transaction.” [3] Conversion value rules exist for the same reason. Google says they let you supply information “that isn’t already reflected in your account (for example, different margins for different types of users, or lifetime value considerations),” applied by audience, location, or device. [2]
Meta arrives at the same place from the other direction. Margin and repeat-purchase behavior are server-side facts that a browser tag cannot see, which is what the Conversions API is for: a connection from “an advertiser’s server, website platform, mobile app, or CRM” into the systems that optimize delivery. [5] If you already run server-side event collection, the hard infrastructure is built. What is usually missing is the decision about what number travels through it.
One order, four honest answers
Here is the same $100 order under four defensible definitions of its value, and the target each one needs to mean the identical amount of money kept.
| What you send as the value | The number | Target ROAS for the same profit | What it quietly assumes |
|---|---|---|---|
| Order revenue | $100.00 | 10.2x | That your margin never varies by product |
| First-order contribution | $43.00 | 4.4x | That nothing ever comes back |
| Expected contribution after returns | $39.08 | 4.0x | That your return rate is stable |
| Twelve-month contribution | $48.54 | 5.0x | That you can fund the gap until the repeat lands |
Every row describes the same business outcome: about $9.77 of ad spend against that order, and about $39 of contribution kept. The targets differ by a factor of two and a half, and none of them is wrong. A 10.2x target is not ambitious and a 4.0x target is not lax. They are the same instruction written in different currencies, which is why a target quoted without its definition tells you nothing at all.
The practical consequence is that the definition has to be decided before the target, and changed only deliberately. Teams get hurt when the value definition changes underneath a target that stays where it was, because the account silently starts buying to a standard nobody chose.
Contribution margin → the revenue ROAS where ads stop paying
If you send order revenue as your conversion value, this is the ROAS at which an order contributes exactly nothing before fixed costs. Everything above it is contribution, everything below it is a subsidy.
Model: breakeven ROAS = 1 ÷ contribution margin. Contribution margin is revenue minus cost of goods, variable costs, and expected returns. Fixed costs are deliberately excluded, because they do not move when the algorithm buys one more order.
Send contribution margin as the value instead and this arithmetic disappears from your head, which is the real argument for doing it. A target of 4.0x on contribution means four dollars kept per dollar spent, in every campaign, in every market, at every margin. Nobody has to remember which category runs at 22% and which runs at 61%.
The revenue that came back
A value pipeline without a refund pipeline reports profit the business never had. Google supports correcting this after the fact through conversion adjustments, and the two modes do genuinely different jobs. A RESTATEMENT replaces the value while the conversion stays counted, which is what a partial return is. A RETRACTION removes the conversion outright, which is what a cancellation is. Both are matched using the order ID you sent at purchase time. [4]
You need: a stable order ID written into the conversion at purchase, stored in the same place your refunds are recorded, and a scheduled job that reconciles the two. Google’s API documentation is explicit that the order ID is the preferred match key.
You get: a Target ROAS that reflects money kept rather than money briefly received. Without it, the strategy is optimizing toward your gross return rate, and it will find the cohorts that return most.
That last point is the one worth sitting with. An algorithm told to maximize pre-refund revenue will happily discover the customers who buy three sizes and keep one. It is not malfunctioning. It is doing exactly what it was told, against a number that stops being true a week later.
What the evidence actually supports
Google publishes a case for Citibanamex reporting a 27% increase in credit card bookings and a 7% lower cost per booked card after moving to value-based bidding. [1] On the Meta side, Søstrene Grene ran a multi-cell conversion lift study comparing predicted lifetime value bid multipliers against its usual automatic bidding, and reported 31% more incremental purchases, 28% higher incremental return on ad spend, and a 24% lower cost per incremental purchase. [6] A European pet retailer reported 12% higher margin and 9% more site visits on the same approach. [7]
Read those honestly. Three advertisers, each published by a party with an interest in the result. The Søstrene Grene number is the strongest of the three because it came out of a conversion lift study rather than a platform-reported comparison, which is the distinction that separates a measured claim from a flattering one.
What nobody publishes is adoption. Neither platform reports how many advertisers send margin or lifetime value rather than revenue, and no independent survey with real data access does either. Any specific percentage you have seen for this is somebody’s guess wearing a decimal point.
Getting there without breaking the account
- Start by measuring the gap. Compute contribution margin per order for last quarter and compare it to what you have been sending. The ratio between them is how wrong every target in the account currently is.
- Store the order ID in the conversion from day one, before you change any value. Everything downstream depends on it.
- Move in one step, not two. Change the value definition and restate the target in the same change, so the account never runs a new definition against an old target.
- Use category-level margin rather than one blended multiplier. A single average across a catalog with real spread just relabels revenue.
- Ship the refund reconciliation job in the same pass, on a daily schedule. A weekly one is usually fine. A missing one is not.
- Hold a baseline. Keep a market or a campaign on the old definition long enough to tell a real improvement from a seasonal one, then judge it on where the next dollar lands rather than on the headline ratio.
- Leave lifetime value until last. First-order contribution with returns subtracted is most of the benefit at a fraction of the modeling risk.
The uncomfortable part of all this is that it is not a media problem. Nothing in it requires a new campaign type, a new platform feature, or a better bid strategy. It requires knowing what a product costs you, what fraction of it comes back, and whether the buyer returns, and then putting those three facts into one field that has been sitting there, filled with the wrong number, the entire time.
Sources
- Google Ads Help · About value-based biddingdefinition, and the Citibanamex results of 27% more bookings and 7% lower cost per booked card
- Google Ads Help · About conversion value rulesthe margin and lifetime-value wording, and the audience, location, and device conditions
- Google Ads Help · Set up conversion values for value-based biddingvalues as sales revenue or profit margins, and the limits of a single fixed value
- Google Ads API · Upload conversion adjustmentsRESTATEMENT and RETRACTION semantics, and order ID as the preferred match key
- Meta for Developers · Conversions APIthe server, platform, app, or CRM connection into Meta’s optimization systems
- Churney · Søstrene Grene: predicted lifetime value acquisitionvendor-published case study reporting a multi-cell Meta conversion lift study: 31% more incremental purchases, 28% higher incremental ROAS, 24% lower cost per incremental purchase
- e-dialog · Fressnapf: 12% higher margin with value-based biddinga single agency-published case, reporting 12% higher margin and 9% more website visits
Frequently asked questions
What is value-based bidding?
It is any bid strategy that optimizes toward the conversion value you report rather than the count of conversions. Google defines it as bidding that “enables you to maximize the total value of conversions generated by your campaigns.” The strategy names are Maximize conversion value and Target ROAS on Google, and value optimization or Target ROAS on Meta.
Can I really send profit instead of revenue as the conversion value?
Yes, and the platforms invite it. Google’s own setup documentation describes bids being set to maximize conversion value “such as sales revenue or profit margins,” and its conversion value rules exist so you can reflect “different margins for different types of users, or lifetime value considerations.” The value field is a number you control, not a reading the platform takes.
What should I subtract from revenue to get a usable conversion value?
Cost of goods, the variable costs that scale with an order such as payment fees and fulfillment, and the expected cost of returns. What is left is contribution margin. Fixed costs stay out of it, because they do not change when the algorithm buys one more order.
Does sending a lower value mean I will get fewer conversions?
It means your target has to be restated on the same basis, not that volume drops. One order can honestly be worth $100 in revenue or $39 in expected contribution, and a 10.2x revenue target and a 4.0x contribution target can describe exactly the same spend and the same profit. What changes is that the second one is optimizing toward the thing you actually keep.
How do refunds get corrected after the fact?
Through conversion adjustments. Google’s API supports a RESTATEMENT, which replaces the value while leaving the conversion counted, and a RETRACTION, which removes the conversion entirely. Both are matched on the order ID you sent at purchase time, which is why storing that ID is a prerequisite rather than a nice-to-have.
Is there evidence that value-based bidding works?
There is case evidence, and it is worth reading for what it is. Google publishes Citibanamex reporting a 27% increase in credit card bookings and a 7% lower cost per booked card. Søstrene Grene ran a multi-cell Meta conversion lift study on predicted lifetime value bidding and reported 31% more incremental purchases and 28% higher incremental ROAS against its usual automatic bidding. Both are single advertisers, published by interested parties.
What share of advertisers actually do this?
Nobody credible publishes that number. Neither Google nor Meta reports how many advertisers send margin or lifetime value instead of revenue, and no independent survey with platform data access does either. Treat any specific adoption percentage you see as commentary.
What is the most common way this fails?
A flat value. If every conversion carries the same number, a value strategy has no distribution to learn from and behaves like conversion-count bidding with worse reporting. Google’s own guidance is that a single fixed value per conversion action “may not fully capture the varying values of each transaction,” and that transaction-specific values are the point.
Do I need the Conversions API for this on Meta?
For anything computed after the browser closes, yes. Margin, expected returns, and lifetime value are server-side facts, and Meta’s Conversions API exists to connect data from “an advertiser’s server, website platform, mobile app, or CRM” to its optimization systems. A browser tag cannot know what an order cost you.
Mikołaj Salecki
Writes about media, tech, and AI business for people who actually run digital. Former agency lead. Skeptic of frameworks that read better than they perform.
