ROAS (return on ad spend)

The ROAS formula and what to put in each side

The core formula is: **ROAS = revenue attributed to ads ÷ ad spend** for the same period and rules. Revenue may be gross sales, net of discounts, or contribution. Pick one and label it. Spend should include media invoices at minimum; some teams add platform fees. Do not mix currencies or leave refunds unaddressed if you claim "net" ROAS. Example: $12,000 attributed revenue on $3,000 spend → 4× ROAS.

That 4× only means something relative to margin structure: a 4× on thin-margin goods can lose money after COGS, shipping, and returns while a lower ROAS on high contribution can be fine. ROAS is not profit. It is not CAC. It does not know whether the buyer was new. Write the definition beside every dashboard tile.

When finance and growth use different revenue bases, every weekly meeting becomes a conversion argument about arithmetic. Shopify, ad platforms, and your warehouse of truth will disagree; choose a source of record for board numbers and treat platform ROAS as directional for in-channel optimization.

Platform ROAS vs blended ROAS

Platform ROAS is what Ads Manager or Google Ads reports using that platform's attribution window, click and view rules, and modeled conversions. Google's ROAS definition is the standard in-channel framing. Blended ROAS (or marketing efficiency ratio variants) uses total ad spend across channels divided by a store-level revenue number you choose. Often total revenue or total new-customer revenue.

Platform figures optimize toward what the auction can see; blended figures answer whether the whole paid program pays the bills under your books. Use platform ROAS to compare campaigns inside one channel with a stable attribution setting. Use blended ROAS or, better, blended new-customer CAC for company-level efficiency. When platform ROAS looks strong and bank cash looks weak, suspect returning-customer credit, over-attribution, discounting, or rising COGS.

Not only "creative fatigue." Do not average platform ROAS across Meta and Google into one fake number; their models are not on the same scale. Report them side by side with spend weights.

Attribution windows and the vanity trap

Attribution decides which sales a campaign may claim. Last-click, data-driven, 1-day click, 7-day click, view-through, and engaged-view options change ROAS without any real-world change in customer behavior. Longer windows and view-through credit inflate ROAS and encourage over-spend on upper-funnel or retargeting that harvests demand created elsewhere. Shorter click windows look harsher and can under-credit assisted paths. Pick settings deliberately and freeze them when comparing periods.

Note iOS and privacy changes that increase modeled conversions. Precision theater is still theater. Incrementality tests (geo holdouts, PSA tests, platform lift studies) are how you challenge ROAS that only exists inside a walled garden. I treat unexplained ROAS spikes after an attribution toggle as definition changes, not performance miracles. Document every setting change with a date so historical charts stay honest.

New-customer ROAS vs retargeting mirages

Retargeting and existing-customer campaigns often show high ROAS because they reach people already likely to buy. Including recent site visitors and past purchasers. That can be efficient for harvesting, but it is a poor signal for prospecting efficiency. New-customer ROAS (or new-customer revenue ÷ prospecting spend) answers whether ads create buyers you did not already have.

Without a new-versus-returning split, scaling "what works" can mean buying your email list back at auction prices. Implement new customer flags from Shopify or your CDP in offline conversions or platform audiences carefully, with privacy and matching limits understood. Compare prospecting ROAS to retention revenue from email and SMS so channels are not double-credited in your head. When new-customer ROAS collapses while blended ROAS holds, you are living on warm audiences.

That is the moment to fix offer, landing conversion rate, creative, or product. Not to lower target ROAS until the account empties the warm pool.

ROAS targets, contribution, and break-even math

A "good ROAS" is the ratio that clears your contribution target after COGS, shipping, variable payment fees, expected returns, and the discount rate in the offer. Not a universal 4× from a thread. Break-even ROAS is roughly 1 ÷ contribution margin percentage when ROAS uses gross revenue and costs are defined consistently.

If contribution margin is 50%, break-even ROAS is about 2× before fixed overhead; you still need room for overhead, returns variance, and growth investment. Build targets from your P&L, not from competitors' screenshots. Separate targets for prospecting and retargeting. Recalculate when discount depth, freight, or returns rate change. A ROAS hit funded by 30% off sitewide is a different product than the same ROAS at full price.

Finance and growth should share one break-even sheet. If that sheet does not exist, any ROAS target is a guess with a media budget attached.

When to prefer CAC, LTV, and payback over ROAS

ROAS optimizes revenue per ad dollar; ecommerce survival optimizes contribution per customer over time. Customer acquisition cost focuses on cost per new buyer. Customer lifetime value estimates what that buyer is worth over a defined window. Payback asks how many months until contribution covers CAC. A channel with lower ROAS but faster payback and higher repeat purchase can beat a high-ROAS channel that only closes one low-margin order.

Use ROAS for in-flight campaign triage. Use CAC, LTV, and payback for budget ceilings and hiring plans. Shopify's customer acquisition cost overview is a better company-level frame than a single ad account tile. First-party order and repeat data from your store beats platform-only stories when the two conflict. If you cannot compute a simple LTV or second-order rate from Shopify, fix that instrumentation before you scale spend on ROAS alone.

Efficient advertising is a contribution system; ROAS is one gauge on the panel, not the whole cockpit.