Your ROAS Went Up. Where Did It Come From?

By Igor Ivitskiy - Thursday October 15, 2026 Share (Twitter) WhatsApp Summarize ChatGPT Perplexity Grok Google AI A student once shared her screen with me during a live session. A small budget and a neat table of numbers. ROAS above 500%. The last line of the table said: no sales. I stopped her right there and asked where that ROAS came from.

If your ROAS is big and beautiful, Google thinks you have sales. If you have no sales, Google is counting something as a sale that isn’t one. I didn’t need to open her account to know where to look. The conversion settings. 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. In my last column here I wrote about blended averages that hide losing segments. This time the problem sits one level lower.

Before you argue about attribution, or pay for a geo holdout, 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. Say an account gets 1,000 orders worth $100,000 and 2,000 leads every quarter on $20,000 of spend, and each lead has 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 change only applies to new conversions, so nothing looks strange.

The next quarter brings exactly the same orders, the same leads and the same money in the bank, and the report shows $140,000 and a ROAS of 7. 0. ROAS went up by a sixth because a person changed one field. Two quarters with the same orders, leads and spend. Only the typed-in lead value changed.

(Source – authors own image) Two quarters with the same orders, leads and spend. Only the typed-in lead value changed. Chart: author’s illustration. 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 I analyze 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 is about half. The Every setting is right for purchases, because one person can really buy twice. For leads, Google’s own help says that usually only one unique lead per ad click adds value. With Every on a lead form, someone who sends the form, sees a typo in the phone number and sends it again becomes two leads.

A thank-you page that loads twice makes two leads as well. The second source of extra conversions is extra counters. More than one in four accounts with purchase tracking have two or more purchase actions inside the conversions column. Sometimes they measure different sales.

Often they count the same order twice, once through the Google tag and once through a GA4 import or a CRM upload. Most audits look for the wrong duplicate. A counter that drifts away from your real numbers is the harmless one, because it looks broken and somebody removes it. The dangerous duplicate matches your back-end almost exactly.

It looks healthy, so nobody touches it. It sits as a secondary action for months, and the day someone adds it to a goal, conversion value doubles in the reports and the bidding of every campaign that uses that goal. No alert fires, because nothing broke.

Two Out of Three Accounts Keep a Primary Action No Campaign Uses About two in three accounts I analyze have at least one conversion action that is marked primary while no campaign goal bids on it. It doesn’t bid today. It carries a status that says it matters, and nobody looks at it because it isn’t in the column.

Nobody remembers why most of them were set up. One edit to a campaign goal, or one custom goal built in a hurry, and Smart Bidding starts to chase an event nobody has checked in years. Custom goals are the usual way in. Google’s help says it directly: a secondary action included in a custom goal is used for bidding once that goal is set in the campaign.

I rarely see anybody check custom goals during an audit. Cheap Placements Fill a Bad Column First Smart Bidding moves budget toward cheap page views when they count the same as orders. In about six in ten accounts, page views, engagement, file downloads, out