A 133,055-conversion month that produced 668 sales
In April 2026, one of our clients — a Nordic sports retailer running Google Shopping campaigns across Europe — saw 133,055 conversions in their Google Ads dashboard. The reported conversion value was 6,478,582 SEK on 129,520 SEK of spend. A 50x ROAS month.
Their actual sales count for April, measured server-side through Digger: 668 orders, 1,164,176 SEK in revenue. A real ROAS of 9x.
9x is a great month for an e-commerce business. 50x is not what happened. The dashboard number was 199 times higher than reality — and that mismatch doesn’t just inflate a vanity metric. It directs the AI that decides how the budget gets spent.
This is a story about what Smart Bidding actually optimises for when nobody has audited the conversion configuration in two years. We’ve seen it in two of the eleven e-commerce accounts we analysed for this study. The other nine have the opposite problem, which we’ll get to.
What Smart Bidding actually optimises for
When you switch a Google Ads campaign to Maximize Conversions or Target ROAS, you hand over budget control to a machine learning model. That model bids on clicks most likely to produce a “Conversion.”
The word “Conversion” sounds self-explanatory. It isn’t. In Google Ads, a Conversion is whatever you’ve configured a Conversion Action to be. It could be a purchase. It could be an add-to-cart. It could be a page view. The system will obediently optimise toward any of these — it just needs to know which.
For the account in our case study, here’s what Google Ads counted as a “Conversion” in April:
| Conversion action | Category | Times fired | Recorded value |
|---|---|---|---|
| Page View | PAGE_VIEW | 129,349 | 126,644 SEK |
| Begin Checkout | BEGIN_CHECKOUT | 1,227 | 2,857,501 SEK |
| Add to Cart | ADD_TO_CART | 1,224 | 2,577,392 SEK |
| Add Payment Info | (other) | 715 | 707 SEK |
| Purchase (Google tag) | PURCHASE | 326 | 525,785 SEK |
| Purchase (server-side) | PURCHASE | 215 | 390,552 SEK |
| Sum reported as “Conversions” | 133,055 | 6,478,582 SEK |
Ninety-seven percent of those “conversions” were page views. Not sales. Not even cart actions. Just page loads. The remaining three percent were funnel signals — add-to-cart, begin checkout, add payment info — events that happen on the way to a sale but are not a sale.
Actual purchases as counted by Google’s own Purchase action: 326. Counted by the server-side pixel: 215. Both numbers are well below the 668 the back office actually recorded — likely because client-side tags are blocked by iOS Safari, cookie consent declines, and ad blockers on a meaningful share of buyers. That’s a separate failure mode, and a separate story.
The point for now is simpler: Smart Bidding doesn’t know any of this. It treats the 133,055 figure as the target and allocates budget accordingly. The algorithm bids most aggressively for clicks likely to produce page views — because that’s the volume signal — and only marginally for clicks likely to complete a checkout.
The audit that surprised us
We’ve been analysing the conversion configuration of every account under our MCC for an internal series on attribution accuracy. Eleven of those accounts are e-commerce (Shopify and WooCommerce). For each, we pulled the full set of Conversion Actions from the Google Ads API along with their April fire counts.
Two of the eleven had funnel events configured as Conversions. The example above was the more extreme of the two — 99% of its counted conversions came from non-purchase categories. The other had Begin Checkout firing as a primary conversion, accounting for 92% of its counted conversions.
The other nine had clean configurations. One Purchase action. Reasonable click windows. No funnel events. No page view tracking masquerading as conversion tracking.
Those nine accounts have the inverse problem — and the size of the gap is comparable, just in the opposite direction. More on that shortly.
Why this happens
If you’ve ever set up Google Ads conversion tracking with one of the standard wizards — particularly the Google Shopping App or Google Tag Manager templates — you’ve likely been offered a default set of event tags that include Page View, Add to Cart, Begin Checkout, Add Payment Info, and Purchase. The wizards make it easy to enable all of them. They don’t make it equally easy to think about which of those should count toward the metric Smart Bidding bids on.
Two years ago, when the Performance Max rollout pushed agencies to enrich their conversion signals to help the algorithm learn faster, this default became common advice. More conversion signals were supposed to give Smart Bidding more data points and faster optimisation.
What actually happened, in cases like the one we audited, is that every funnel step started being counted toward the bidding signal. Each customer journey contributed five or six “conversions” — page view, add to cart, begin checkout, payment info, purchase (sometimes twice). A buyer who visited four product pages before purchasing counted as nine conversions, not one.
Smart Bidding then optimised toward maximising that count. Which means it optimised toward maximising page views. Which is not what the business owner wants.
What it costs
The direct financial cost of this misconfiguration is hard to isolate in a single number, because it shows up as opportunity cost — budget that could have gone to high-intent search but went to high-volume display impressions instead. Smart Bidding is a redistribution machine, and a bad signal redistributes budget toward the wrong places.
A rough sketch from the case account: the campaigns with the highest reported “conversion rates” in Google Ads were the ones generating the most page views per click. Those were not the campaigns with the highest actual sales conversion rates. The algorithm boosted bids on the page-view-heavy campaigns and reduced them on the sale-heavy ones. The mismatch between reported ROAS (50x) and actual ROAS (9x) is what that redistribution looks like at the top line.
Three months in: a stable pattern
We pulled the same data for February and March 2026 to check whether April was an outlier. It wasn’t.
| Month | Google “Conversions” | Actual sales (server-side) | Ratio |
|---|---|---|---|
| February | 100,712 | (tracking not yet live) | — |
| March | 118,710 | 258 | 460x |
| April | 133,055 | 668 | 199x |
The ratio improved from 460x to 199x not because the Google Ads configuration was fixed, but because server-side tracking caught more of the actual sales as it was rolled out across the customer’s site. The Google side has been wrong, consistently, for as long as we have records.
The other account in our audit with misconfigured tracking shows a similarly stable pattern — different magnitude, same structural issue every month. This isn’t drift. It’s the configuration doing exactly what it was set up to do.
How to check your own setup in five minutes
Open Google Ads. Navigate to Tools & Settings → Conversions → Summary. Look at the column labelled “Include in ‘Conversions’.” Filter to Yes.
You’ll see one row per Conversion Action that counts toward Smart Bidding’s optimisation target. Now look at the Category column:
- ✅ Purchase — correct for e-commerce
- ✅ Submit lead form — correct for lead generation
- 🚨 Page view — incorrect; this is not a purchase
- 🚨 Add to cart — incorrect; this is intent, not revenue
- 🚨 Begin checkout — incorrect; same reason
- 🚨 Add to wishlist — incorrect; same reason
If any of the four flagged categories are set to Include = Yes, your Smart Bidding has been optimising toward something other than completed sales. They can stay enabled as Secondary actions — they’re still useful as funnel signals for diagnostics — but they should not be in the bidding metric.
Then compare the number Google Ads reports as Conversions for the last 30 days against your back office order count for the same period. The ratio should sit between 0.9 and 1.1. Anything outside that window means there’s a measurement error worth investigating — in either direction.
The opposite failure mode
For most of the eleven accounts in our study, the problem isn’t over-counting. It’s the opposite: Google Ads sees a fraction of the actual sales.
One Nordic supplement retailer in our study has a clean conversion configuration — single Purchase action, properly categorised, no funnel pollution. In April, Google Ads recorded 2 conversions. The back office, measured server-side, recorded 32 sales. Google saw 6% of the truth.
Across the seven undercount cases, the visibility loss ranged from 6% to 56% — and the stability across three months was even tighter than for the over-count cases. Same percentage lost, every month.
When Smart Bidding believes a campaign produced 6% of its actual sales, it concludes the campaign is unprofitable and reallocates budget away from it. The campaign was working. The measurement wasn’t.
That’s the gap Digger was built to capture.
Why server-side data matters
The reason this case study could even be written is that we had a second source of truth running alongside Google Ads. Digger writes a server-side record of every form submission and every order, with the Google Ads ad ID and click ID captured at the moment of conversion. When Google Ads loses a conversion to cookie consent or ITP, Digger still has it. When Google Ads inflates a conversion by counting funnel steps, Digger’s record of actual orders stays grounded.
Without a second source, neither failure mode is visible. The dashboard says what it says. Smart Bidding believes what the dashboard says. The budget moves where Smart Bidding wants it to move. And the only signal that anything is wrong is a quarterly conversation with the finance team about why ROAS reports look great while bank balances don’t.
What to do next
If you’ve never audited your Conversion Actions, do that this week. Five minutes in the Conversions panel will tell you whether your Smart Bidding has been optimising toward sales or toward something else.
If the audit reveals a clean configuration but a wide gap between Google Ads conversions and back office orders, the failure mode is the inverse — server-side measurement is the only way to close it.
Either way: the AI is doing exactly what you’ve told it to do. The question is whether you’ve been telling it the right thing.
Frequently asked questions
How do I check if Smart Bidding is optimising toward page views?
In Google Ads, go to Tools and Settings, Conversions, Summary. Filter Include in Conversions to Yes. If any row has the category Page View, Add to Cart, Begin Checkout, or Add to Wishlist, Smart Bidding is bidding on something other than completed sales. Only Purchase (for e-commerce) or Submit lead form (for lead generation) should be included in the primary conversion metric.
What is the right ratio between Google Ads conversions and actual sales?
Between 0.9 and 1.1 over a 30 day window. If reported conversions are higher than back office orders by more than 10 percent, funnel events like Page View or Add to Cart are likely counted as conversions. If lower, client-side tracking is missing sales due to iOS Safari, cookie consent declines, or ad blockers.
Should I keep Add to Cart as a conversion in Google Ads?
Keep it enabled as a Secondary conversion action for diagnostic visibility, but remove it from the Include in Conversions metric that Smart Bidding bids on. Smart Bidding will optimise toward whatever is in the primary metric. Adding intent signals there pulls budget toward high-volume clicks instead of high-value clicks.
Why does my Google Ads dashboard show high ROAS while actual sales are flat?
Two common causes. Either funnel events are counted as conversions, inflating both the count and the assigned value, which is what produced the 50x reported ROAS vs 9x real ROAS in the case study. Or the dashboard is right about ROAS on tracked conversions, but a large share of actual sales never gets tracked, so the campaign appears to perform on a small subset of buyers. Server-side attribution catches both failure modes.
Get Started — Book a Free Demo
This article is part of a series on the gap between platform-reported and actual revenue. The precursor is how AI optimising perfectly toward the wrong customers shows up in a related failure mode. Next up: how Nordic e-commerce stores lose 30–94% of their sales to client-side tracking failures — and what that does to Smart Bidding’s budget allocation.
Methodology note: data drawn from eleven anonymised e-commerce accounts under one MCC, February–April 2026. Sources: Google Ads API (Conversion Action configuration and campaign metrics), Digger server-side attribution.
The Google Ads to back office gap was 199x on the case account, 460x the month before. Book a free demo to see what your number is.
