A student shared her screen during a live session. Small budget, neat table, ROAS above 500%. The last line of the table said: no sales. That anecdote, from Igor Ivitskiy's column at PPC Hero, captures a problem I see in nearly every pipeline review where paid media gets discussed. The number looked great. The number was also fiction.
Her case was easy because the lie was total. The hard cases are accounts where the ads really do sell and the reported number is only partly wrong. There, a bad setting doesn't look like a problem. It looks like a good year.
Before you argue about attribution models, before you pay for a geo holdout, before you commission an incrementality study, check whether the number was measured at all. In most accounts I open, part of it wasn't.
Some of Your Conversion Value Was Typed In by a Person
Every conversion action with a static value carries a number someone typed into a settings field. Google Ads adds it to conversion value next to real revenue from orders, then divides the total by cost. The totals don't show which part was measured and which part was typed.
Ivitskiy walks through the math: an account gets 1,000 orders worth $100,000 and 2,000 leads every quarter on $20,000 of spend, with each lead assigned a static value of $10. Reported conversion value is $120,000, a ROAS of 6.0. Before the next quarter starts, someone decides leads are worth more and changes the value to $20. The next quarter brings exactly the same orders, the same leads, and the same money in the bank. The report shows $140,000 and a ROAS of 7.0. ROAS went up by a sixth because a person changed one field.
A static value is fine when it equals your lead-to-sale rate times your average sale. The trouble starts when that value changes in the middle of the period you compare. Last year's ROAS and this year's ROAS are then measured in different units, and the chart still draws them as one line.
One Lead Is Often Counted More Than Once
In four out of five accounts Ivitskiy analyzes that feed leads into bidding, at least one lead action in the conversions column counts every conversion instead of one per click. Across all lead actions in that column, it's about half. The "Every" setting is right for purchases, because one person can really buy twice. For leads, Google's own documentation says that usually only one unique lead per ad click adds value.
Cometly's troubleshooting guide puts it bluntly: many advertisers accidentally set lead generation actions to "Every," causing massive over-counting when users refresh thank-you pages or when tags fire multiple times. The mistake is common because the default setting isn't always obvious, and the inflated numbers feel like validation rather than error.
Foxwell Digital's audit process recommends checking the "Repeat rate" column in your conversion actions. Ideally, this value should be close to 1. A repeat rate close to 2 or higher is a sign of implementation issues causing duplicate reporting.
The Platform Grades Its Own Homework
The structural problem runs deeper than settings. Databox research on attribution found that summing all platform-reported conversions routinely produces 150 to 250 percent of actual closed customers. Google Ads claims 47 conversions. Meta claims 52. LinkedIn claims 31. Your CRM shows 38 closed customers. Someone is lying, and it's not your CRM.
One in four GTM leaders in that research said at least a quarter of last quarter's pipeline was misattributed due to missing or incorrect click data. Nearly 7 percent reported error rates of 50 percent or more. And 32 percent reported spending 16 to 30 hours per month just cleaning and reconciling attribution data before any analysis happens.
One agency audit found a client's campaign showing 94 conversions per month while the CRM had 11 leads. Google was counting existing customers clicking remarketing ads and landing on thank-you pages, the same person converting twice after clearing cookies, bots hitting the page, and people who bounced in under three seconds. None of that is a lead. None of it is revenue. But Google counted every pixel fire.
ROAS Doesn't Tell You If You Made Money
Even when the measurement is clean, the number itself is incomplete. My Digital Hero's breakdown shows what happens when you apply margin to a ROAS that looks healthy. With 30 percent margin and £1,000 of ad spend, a ROAS of 2 produces £2,000 in revenue and £600 in gross profit. You lost £400. A ROAS of 3.3 produces £3,300 in revenue and £990 in gross profit. You broke even. A ROAS of 4 finally produces £200 of actual profit.
The HQ Digital's benchmarks make the variance stark: a fashion brand with 25 percent gross margin needs 4x ROAS just to break even, a target most Meta campaigns can't sustain consistently. A SaaS business with 70 percent margins can profitably run at 1.43x. The formula is simple: break-even ROAS equals 1 divided by your gross margin percentage. But I rarely see this number in the same deck as the ROAS chart.
Shopify's guidance frames it correctly: without knowing your break-even point, you can't tell whether your advertising efforts are generating profit or quietly eroding your margins.
The Audit Before the Attribution Debate
The fix isn't complicated. It's just rarely done.
First, pull every conversion action in your account and check the counting method. "One" for leads, "Every" for purchases with unique order IDs. Enrich Labs' best practices guide calls this the first rule: only primary conversions that map to revenue or sales-qualified leads should train Smart Bidding. Soft events stay secondary or out of Ads entirely.
Second, check for duplicate conversion actions. GrowthSpree's audit data shows that roughly 70 to 80 percent of B2B SaaS Google Ads accounts have at least one significant tracking issue. About 60 percent have multiple events under one goal. The typical dashboard CPL versus true cost per SQL gap runs up to 12x.
Third, reconcile platform conversions against CRM actuals weekly. The Databox research recommends a four-tier trust hierarchy: CRM closed-won data first, server-side event data second, GA4 third, platform dashboards last. Platform data is for optimization, never for budget justification.
Fourth, calculate your break-even ROAS and put it on the same slide as your reported ROAS. If your margin is 40 percent, your break-even is 2.5x. If your reported ROAS is 3x and your break-even is 2.5x, you're making 50 cents of gross profit per dollar of ad spend. That's the number your CFO actually needs.
The Number Your Board Needs
The next time someone presents a ROAS chart that goes up and to the right, ask three questions. Did anyone change a static conversion value during the period? Are leads counted once per click or every time the tag fires? What's the break-even ROAS for this margin structure?
If the room goes quiet, you've found the problem. The number wasn't measured. It was assembled from settings, defaults, and platform self-reporting. Model or it didn't happen. And if the model is built on bad inputs, the output is just a story someone told themselves about a good year.
