Bid on profit,
not revenue.
The conversion value you send back to a platform is what its bidding model learns from. Zappush computes each conversion's true value, contribution margin or predicted LTV, and delivers it server-side, so Meta and Google stop chasing discount-hunters and start winning your high-margin, repeat customers.
From $59/month · 14-day free trial · Cancel anytime
The value is the training signal
$200 order
50% refund rate
$60 order
resubscribes 6×
Send raw totals and the algorithm hunts the first kind. Send true value and it hunts the second.
- Profit, not top-line
- margin- or LTV-weighted value on every conversion, corrected when orders refund
- 0.5× to 10×
- Google's documented value-rule range for bidding up a cohort by location, device, or audience
- Unlearns bad orders
- negative-value adjustments on refunds, so lookalikes stop cloning your refunders
Raw revenue trains the algorithm to find your worst customers.
Send the browser pixel's raw order total and you teach Meta and Google that a $200 order with a 50% refund rate is worth more than a $60 order that resubscribes six times, so the model goes and finds more of the first kind. The two things that define a good customer, product margin and lifetime value, never reach the platform, because they live in your CRM and your P&L, not in the checkout event.
The browser makes it worse: 15 to 40% of purchase events never arrive at all, lost to iOS restrictions, Safari's tracking prevention, and ad blockers. So the platform trains on a value that is both wrong and incomplete. Budget flows to refunders and one-and-done buyers, POAS erodes while reported ROAS looks fine, and lookalike modelling compounds the mistake by hunting for more of them.
First, know what your best customers look like.
Platforms let you bid more for valuable cohorts, but most marketers never use those levers for a simple reason: they don't know what their valuable cohort looks like. Zappush does, because every order sits on a persistent customer profile.
Rank customers by real LTV, computed from full purchase history, refunds included, not from a single session.
See what the top slice has in common: region, device, acquisition channel, first product bought, full-price versus promo behaviour.
Turn the cohort into bidding inputs: a Customer Match seed, a value rule worth writing, and a value on every conversion that reflects who converted.
Your highest-value cohort
From your order historyWhat they have in common
Then use the levers the platforms already built.
None of this is a workaround. These are documented bidding mechanics; they just need data most accounts don't have.
On Google Ads
Smart Bidding bids per auction on the value you supply. Target ROAS and Maximize Conversion Value read the conversion value on each event and set a tailored bid for every individual auction. Feed margin or LTV instead of raw revenue and the model optimises for profit.
Conversion value rules bid up your cohort. Google adjusts conversion values by geographic location, device, and audience, including Customer Match lists, with multipliers from 0.5× to 10×, consumed at auction time. Your cohort profile tells you which rule is worth writing.
Your old bid adjustments are being ignored. Under value-based bidding Google bypasses manual device and location bid adjustments (only -100% device exclusions survive). Value rules are the sanctioned lever, and most accounts have never written one.
New customers can carry extra value natively. The new customer acquisition goal adds an advertiser-chosen extra value to first purchases, and Google recommends sizing it from customer lifetime value, the number Zappush computes from your order history.
On Meta & TikTok
Value optimisation needs values worth optimising. Meta's highest-value bidding runs on the purchase values your events carry. Zappush delivers a margin- or LTV-weighted value on every conversion via CAPI, server-side, so the model ranks buyers by what they are actually worth to you.
Refunds become negative signal. When an order refunds, Zappush sends the adjustment so the platform unlearns it, instead of seeding lookalikes with your refunders.
Match quality decides how much signal survives. Values only train the model when the event matches a person. Server-side delivery with hashed first-party identifiers keeps Event Match Quality high, so your value signal is used, not discarded.
How it works
Compute the true value of each conversion
Instead of the raw order total, Zappush calculates what the customer is really worth: contribution margin (revenue minus COGS, shipping, discounts, and fees) or a predicted-LTV score built from the persistent profile's purchase history. The number that leaves your server reflects profit, not top-line.
Send it server-side with high match quality
The value is delivered through Meta's Conversions API and Google's server-side connection, keyed to a shared event ID so it deduplicates against your pixel, and enriched with hashed identifiers so it lands with a strong Event Match Quality score instead of being lost to ITP or ad blockers.
Correct the value as reality changes
When an order refunds or cancels, or an LTV estimate resolves into real revenue, Zappush sends an adjustment, including a negative-value event so the platform actively unlearns bad orders. Outliers are capped before they poison the model.
Questions, answered.
- Why shouldn't I just send the order total as the conversion value?
- Because the value is the algorithm's training signal. Raw revenue teaches Meta and Google that a big first order is best, even if that customer refunds or never buys again, so they go find more like them. Sending contribution margin or predicted LTV trains bidding toward your profitable, repeat customers instead. Vendors report 20, 40% ROAS/POAS lifts from making that switch.
- How are refunds and cancellations handled?
- Zappush sends a conversion adjustment, including a negative-value event on Meta, so the platform actively unlearns the bad order instead of building lookalikes of your refunders. Values are also restated as predicted LTV resolves into real revenue.
- How is this different from just turning on Google's Target ROAS or Meta value optimization?
- Those bid on whatever value you feed them. If that value is the browser pixel's raw total, incomplete from ITP and ad blockers, and equal to top-line revenue, you're optimizing on a lossy, misleading signal. Zappush computes the true profit- or LTV-weighted value, delivers it server-side with high EMQ against a persistent identity, and QAs and corrects it, so the bidding you already have finally trains on the right number.
Find out what value your account is bidding on today.
Start free in minutes, or see it live on a call.
From $59/month · 14-day free trial · Cancel anytime