Profit on ad spend: bidding to margin, not revenue.
Two orders of the same value are not worth the same. Until your bidding knows that, it will keep buying turnover and calling it growth.
Here is a question most ecommerce dashboards cannot answer: which of your campaigns made money last month. Not which reported the best ROAS, but which actually generated profit after the cost of goods. For most accounts these are different lists, and the difference is where growth quietly goes wrong.
A 4x ROAS sounds excellent. On a product with 15% gross margin, it is a loss once you account for the ad cost. A 2.5x ROAS sounds mediocre. On a 70% margin product, it is a profit engine. Revenue-based ROAS treats these two as the wrong way around, and Smart Bidding, optimizing on revenue values, obediently follows. The algorithm scales the impressive-looking loser and constrains the boring winner. This is not a bidding flaw. It is a values flaw. The machine optimizes exactly what you feed it.
The mental model: POAS
Profit on ad spend is the same arithmetic as ROAS with one substitution: contribution margin instead of revenue in the numerator. Gross profit generated per unit of ad spend. It is a small change to a formula and a large change to behavior, because it re-ranks everything. Campaigns, products, audiences, and channels all reorder when judged on what they contribute rather than what they turn over. The first time a team sees its account ranked by POAS, at least one flagship campaign usually drops from hero to passenger.
You do not need perfect unit economics to start. Even margin bands (high, mid, low) applied to your catalog will re-rank your campaigns more truthfully than exact revenue ever will. To find where your own profitability line sits, run your numbers through the free breakeven ROAS calculator.
A worked example: where ROAS and profit disagree
The argument is more persuasive as arithmetic than as principle, so here is a full month decomposed. The figures below are illustrative rather than a client account: they are constructed to be realistic for a home and decor catalog, which is the category where margin spread within a single account tends to be widest. Nobody publishes their cost of goods, so the margins here are modeled. The structure of the result is what generalizes, and we see it in some form in most catalogs we take over.
Take a store spending $30,000 a month in one shared shopping campaign, returning $126,000 in revenue. Blended ROAS of 4.2x, comfortably above a 4x target, and a campaign nobody would flag in a monthly review. Now split it by the three product lines inside it.
- Premium wool rugs. $9,000 spend, $25,200 revenue, 62% gross margin. ROAS 2.8x, the worst in the account. Contribution $15,624, so profit after ad spend is $6,624.
- Mid-range synthetics. $12,000 spend, $54,000 revenue, 38% gross margin. ROAS 4.5x. Contribution $20,520, so profit after ad spend is $8,520.
- Clearance and entry price points. $9,000 spend, $46,800 revenue, 14% gross margin. ROAS 5.2x, the best in the account. Contribution $6,552, so the line loses $2,448.
The ranking inverts completely. By ROAS the order is clearance, then mid-range, then premium. By POAS it is premium at 1.74, mid-range at 1.71, and clearance at 0.73. The line with the strongest ROAS in the account is the only one losing money, returning 73 cents of gross profit for every dollar of ad spend, and it has been the line the team points at in every review because 5.2x is the number that reads as success.
A POAS of 1.0 is breakeven on ad spend alone, before any overhead. That is the line to internalize, because it is unintuitive after years of ROAS thinking: 2x POAS is a strong business, and anything under 1x is subsidizing turnover with your own capital.
What the reallocation is actually worth
The instinct at this point is to switch the clearance line off, which is usually wrong for reasons covered further down. The disciplined move is to move budget toward the lines that convert spend into contribution, then measure whether the contribution followed.
Continuing the same illustration: total contribution across the three lines is $42,696 against $30,000 of spend, a blended POAS of 1.42. Move the $9,000 sitting in the clearance line into premium and mid-range and the honest assumption is that it performs worse than the existing spend there, because you are buying further down the demand curve. Assume it returns 1.4 rather than the 1.7 those lines currently deliver, and contribution goes to roughly $48,700. Same $30,000 of spend, about 14% more gross profit, and a reported ROAS that falls from 4.2x to roughly 3.6x.
That last sentence is the whole political problem with this work. The correct decision makes the headline number worse, which means it has to be agreed before it is made rather than explained afterwards. When we published the Rugs Outlet engagement, the visible result was ROAS moving from 2.5x to 4.5x over four months, because that is the number a case study can state cleanly. The internal version of that work is always messier: some campaigns are deliberately allowed to look worse.
Where POAS misleads, and what to do about it
POAS is a better objective than revenue, not a complete one, and the honest failure modes are worth naming because bidding to margin naively will cost you growth.
- First-order margin is not customer value. A thin-margin entry product that reliably produces a repeat buyer can be the most valuable acquisition in the catalog, and a first-order POAS model will strangle it. If you can measure repeat behavior by product line, use contribution over the first year rather than the first order.
- Returns distort margin more than most teams model. A line with a 30% return rate and a restocking cost has a materially lower true margin than its spreadsheet, and the categories with the highest return rates are often the ones with the best headline margin.
- Discount depth moves margin per order in ways a static margin band cannot see. If promotional intensity varies through the month, a fixed margin assumption will systematically misprice the promotional weeks.
- Shared costs sit outside the calculation. POAS is gross profit over ad spend; it says nothing about fulfillment, support load, or the fixed cost base. A 1.2x POAS product line can still be unprofitable at the business level.
- Thin-margin catalogs need volume that margin-only bidding will not chase. A book retailer operating on single-digit percentage margins cannot bid to margin the way a furniture brand can, and the answer there is usually a much lower breakeven target with tighter operational discipline rather than a POAS objective.
The practical resolution is to bid to the best value you can defend rather than the most sophisticated one you can construct. First-order contribution beats revenue. First-year contribution beats first-order contribution where the data supports it. Perfect unit economics you cannot compute monthly is worse than margin bands you can, because a value nobody maintains drifts back to being wrong within two quarters.
Teach the machine: margin-aware conversion values
The mechanical fix is to change what a conversion is worth. Instead of sending the platform the order revenue, send a value that reflects margin: either true per-product profit if your data supports it, or a margin-weighted revenue figure if it does not. Target ROAS then becomes, in effect, target POAS without the platform needing to know the difference. The algorithm starts chasing the baskets that matter.
- Best: pass per-order gross profit as the conversion value, computed from product-level cost data
- Good: weight revenue by category-level margin bands where per-SKU cost data is not available
- Minimum: exclude or down-weight the pathological cases — heavy discounting, high-return categories, loss-leader SKUs
- Always: document what the value means, so nobody reads a profit-based number as revenue
Structure follows margin
Values are half the answer; structure is the other half. When high-margin and low-margin products share a campaign, they share a target, and the blended average mis-serves both. Segmenting structure by margin tier (the custom-label work we covered in the Shopping feed essay) lets each tier carry a target its economics can support. High-margin lines get room to be aggressive. Thin-margin lines get held to the efficiency they require to exist. Returns deserve the same honesty: a product line with a 30% return rate has a very different true margin than its spreadsheet suggests, and its bidding should know that too.
What changes when you make the switch
Expect reported ROAS to look worse on the campaigns that were flattered by the old math, and expect that to be uncomfortable in the first month. Expect a shift of budget toward products nobody was excited about, because they quietly earn. And expect the conversation with finance to get easier, because for the first time the ad platform and the P&L are speaking the same language. Revenue is what the dashboard celebrates. Margin is what the business keeps. Bid to what the business keeps.
It is also worth being clear about what this fixes and what it does not. Bidding to margin corrects a misallocation, which is why the first month usually produces more contribution on identical spend. It does not lift the ceiling on the business. Once budget is pointed at the right economics, the binding constraint moves elsewhere, into demand, catalog, or operations, and that is a different problem with a different diagnosis.
What is POAS and how is it different from ROAS?
POAS is profit on ad spend: gross profit generated for every unit of ad spend, where ROAS measures revenue for every unit of ad spend. The arithmetic is identical apart from the numerator, and the substitution changes conclusions rather than refining them. A four times ROAS on a product carrying fifteen percent gross margin returns sixty cents of gross profit per dollar spent, which is a loss. A two and a half times ROAS on a seventy percent margin product returns a dollar seventy-five, which is a strong business.
What is a good POAS?
A POAS of 1.0 is breakeven on ad spend alone, before any overhead, so the useful benchmarks sit above it. Below 1.0 you are subsidizing turnover with your own capital. Around 1.5 is a functioning acquisition engine for most ecommerce businesses. At 2.0 and above, advertising is comfortably contributing to fixed costs and profit. The right target for your business depends on your fixed cost base and how much of first-order margin you are willing to reinvest in acquiring a customer you expect to buy again.
How do I send margin-based conversion values to Google Ads or Meta?
Send gross profit rather than order revenue as the conversion value, computed from product-level cost data at the point the order is confirmed. Where per-item cost data is unavailable, weight revenue by category-level margin bands, which is materially better than revenue and takes a day rather than a quarter. Target ROAS bidding then behaves as target POAS without the platform needing to know the difference. The important discipline is documenting what the value represents, so nobody later reads a profit figure as revenue.
Will my reported ROAS get worse if I switch to margin-based bidding?
Yes, on the campaigns that were flattered by revenue-based values, and you should agree this before making the change rather than explain it afterwards. Budget moves away from high-revenue low-margin lines toward lower-revenue higher-margin ones, so reported revenue per dollar falls while gross profit per dollar rises. If the reporting change is not agreed in advance it will be read as a performance regression, and the correct decision gets reversed in month two.
Does bidding to margin work for thin-margin catalogs?
The principle holds but the target changes. A retailer operating on single-digit percentage margins has very little room between breakeven and loss, so the useful work is a much lower breakeven target combined with tight operational discipline, rather than a margin-weighted bidding objective that will simply starve the account. Margin-aware values are most powerful where the spread within the catalog is wide, because it is the spread, not the level, that revenue-based bidding gets wrong.
Should I use first-order margin or lifetime value in the conversion value?
Use the best figure you can compute reliably every month. First-order gross profit is a large improvement on revenue and is available to almost everyone. First-year contribution is better where you can measure repeat purchase behavior by product line, because it stops you strangling a thin-margin entry product that reliably produces a repeat buyer. What does not work is a sophisticated lifetime value model nobody maintains: a value that stops being updated drifts back to being wrong within two quarters, and wrong values are worse than crude ones.
Written by Sam Nouri, founder, adsrunner. If this resonated and you want to apply it to your own account, you can book a strategy call or run a free audit.
How we research, source figures, and handle corrections: editorial policy.