The question that started this essay came from a paid search lead at a mid-market UK retailer. She had, in front of us, three dashboards. The first was Google Ads, which reported that her account had produced 41,382 conversions in the previous 30 days at a blended cost per acquisition of £18.40. The second was her GA4 dashboard, which reported 33,109 conversions attributable to Google paid search in the same window at a CPA of £23.01. The third was her Shopify orders dashboard, which reported 27,806 completed orders where the customer's last click before purchase had been a Google Ads click, at what we could reasonably describe as a blended CPA of £27.39. Same account, same period, three different numbers, each of them defensible.
The head of ecommerce, who had also joined the meeting, then asked the obvious question. Which of these numbers should we be running the account on?
The honest answer, in the case of that account and in the case of every account we have audited since, is: none of them, exactly. This essay is about what to run the account on instead.
Where the gap comes from
It helps to be specific about why three broadly credible numbers from three broadly credible systems disagree by such a wide margin. There are, on inspection, three separate mechanisms in play, and most PPC leads I speak with understand one or two of them but rarely all three at once.
The first is measurement window. Google Ads reports conversions using an attribution window that defaults, in most account configurations, to 30-day click plus 1-day view. GA4, by default, uses a 30-day click window but no view-through credit. Shopify's last-click view has no defined window at all — it simply looks at the immediately preceding session on the store. Three different windows produce three different counts before you even get to the underlying model.
The second is the modelled vs observed split. Google Ads' conversion column, since roughly the middle of 2023, is a blend of observed conversions and platform-modelled conversions. The modelled share is not disclosed on a per-account basis but is, by triangulation against holdout tests, often in the 20-35% range for mid-market retail accounts. GA4's number is a smaller blend of observed and modelled — mostly observed for logged-in users, mostly modelled for consented cookieless users. Shopify's number is closest to purely observed but is bounded by the browser session, which for a meaningful share of users has been reset by intelligent tracking prevention.
The third is de-duplication. All three systems, in their own way, are counting individual events. None of them, without substantial custom work, are counting incremental conversions. A customer who was going to purchase anyway and happened to click an ad on the way there is counted by Google Ads, potentially counted by GA4, and definitely counted by Shopify. The ad receives credit; the customer would have converted regardless; the count is real but the incrementality is nil.
Why the gap matters
You could, at this point, reasonably ask: does it matter that the three numbers disagree, if the account is producing good aggregate performance? For most functions, most of the time, the answer used to be no. The dashboard was directionally right. The relationships between campaigns, keywords, and ad groups on the same dashboard were consistent enough to drive good decisions. The absolute number was less important than the relative comparison.
That answer, in 2026, is starting to break down. The reasons are worth listing. Automated bidding — target CPA, target ROAS, and their Performance Max cousins — makes its decisions based on the Google-reported conversion number, which is the most modelled of the three. When the modelled share of that number rises, as it has, the platform's automated bidding decisions are increasingly optimising for a metric that has diverged from the till receipt. The CFO's line-item view, meanwhile, is anchored on Shopify, which is the closest to the till receipt but is also the most under-crediting of the three. The paid search lead is stuck in the middle, defending an account against a CFO who thinks it's under-performing (because the till number is lower than the ad platform number) while the platform's bidder is quietly wasting budget optimising for a number that isn't the till.
"The dashboards aren't lying. They are answering different questions competently. Nobody is asking them the question we actually want them to answer, which is: for every incremental pound we spend, what incremental revenue do we get?"
What to run the account on
My working recommendation, developed over roughly two years of audits and tested against about forty accounts in the last twelve months, is a three-line internal dashboard that most PPC teams can stand up in a fortnight. It does not replace the platform dashboards; those still drive the daily optimisation work. It sits above them and answers the questions the platform dashboards cannot.
The first line is platform-reported blended CPA, aggregated across all paid media, weekly. This is the number the platform bidding algorithms are optimising against, and it is the number the CMO tends to see first. It is worth tracking not because it is the truth but because when it moves, something has happened. Sudden divergence between this line and the other two is a diagnostic signal.
The second line is MMM-derived blended CPA, aggregated across all paid media, monthly. If you do not have MMM in place, this line is initially proxied by a simpler geo-holdout analysis run quarterly. This is the number that captures incrementality — the difference between what the paid line is contributing and what would have happened without it. In the accounts we work with, this number is, on average, 1.5-2.5x the platform-reported number. The ratio is not universal; the direction is.
The third line is contribution margin per pound of paid media, monthly. This is the number the CFO cares about. It rolls in gross margin, returns, promotional discount, and (where the finance function is willing to allocate them) the operational costs of running the paid media function. It is the only one of the three that is expressed as a return on capital rather than a cost of acquisition, and it is the number that, in the end, tells you whether the function is producing value at the level the business needs.
The weekly practice
The three-line dashboard, on its own, is not sufficient. It has to be paired with a weekly practice — a small, boring, repeatable set of questions the team asks itself.
Every week, the team looks at line one and asks: what moved, and why? A 4% shift in blended CPA week-on-week is usually noise. A 12% shift almost always has a cause. The point of the weekly review is not to react to the number but to build a shared team memory of what has caused past movements — creative fatigue, seasonal shift, competitor entry, feed-quality change — so that future movements can be interpreted quickly.
Every month, the team looks at line two and asks: is our incremental cost per acquisition consistent with the price we can defend to the finance function? If the answer is yes, the current spend level is broadly right. If the answer is no, the question is not "how do we reduce CPA" but "what parts of the spend are the least incremental, and can we cut them without materially hurting the aggregate output?" This is the question incrementality analysis is uniquely suited to answering; last-click reporting is not.
Every quarter, the team looks at line three and asks: is the paid media function producing contribution margin at a rate the rest of the business can grow on? This is the question that determines the annual budget conversation. It is also the question most PPC functions do not equip themselves to answer, because they treat the dashboard number as the definitive read on their performance.
The political problem
The reason most PPC functions do not run a version of this dashboard is not that they cannot. It is that the internal politics of doing so are, in most companies, difficult.
The platform-reported dashboard makes the paid function look bigger and more productive than it is. Moving to a three-line internal dashboard almost always reveals that the paid function's contribution to the business is smaller — sometimes materially smaller — than the platform dashboard had suggested. The paid lead who introduces this reporting is, in the short term, damaging their own function's apparent productivity. The head of growth above them may or may not thank them for the honesty. The CFO usually will. Whether the CMO does depends heavily on the CMO.
The right time to introduce the reporting is when the platform-reported and MMM-derived numbers have not yet diverged too widely. If you leave it until they have, the internal conversation about the divergence becomes politically fraught. Better to run both in parallel, at less stakes, for a quarter or two before making the internal dashboard the primary read.
The teams that have done this in 2024 and 2025, in our sample, are in the strongest budget positions in the accounts we look at now. The teams that have not are, currently, defending headline numbers that their finance functions increasingly do not believe. The gap between the two positions is going to widen through 2026 and 2027. Now is a good time to have the conversation.