- Platform numbers will never sum to your revenue. That is a design feature of last-click-per-platform reporting, not a tracking bug.
- Use three layers: platform for optimisation, blended for allocation, incrementality for deciding whether a channel deserves to exist.
- Blended new-customer CAC and marketing efficiency ratio are crude, unglamorous and almost impossible to game. Report them weekly.
- Attribution windows are a reporting choice, not a fact. Write yours down and stop changing them mid-quarter.
- Reconcile to the bank monthly. A measurement stack that has never been checked against banked revenue is a hypothesis.
Why platform numbers never sum
Add up the revenue your ad platforms report and compare it to what the business actually took. In most multi-channel accounts the platforms will claim somewhere between 120 and 200 percent of real revenue. People discover this, assume something is broken, and spend a quarter looking for the bug.
There is no bug. Each platform reports conversions it can plausibly claim under its own attribution settings, in isolation, with no knowledge of the others. A customer who sees a video ad, clicks a social ad a week later, then searches your brand name and converts will appear in three separate reports as three separate successes. Every one of those claims is defensible on its own terms. The sum is nonsense.
Three mechanics drive most of the overlap:
- View-through conversions. Meta's default settings credit conversions from people who saw an ad without clicking it. Useful as a directional signal, catastrophic when added to a click-based number from another platform.
- Differing windows. A 30-day click window on one platform and 7-day click plus 1-day view on another means the same conversion is counted twice, at different times, in different reports.
- Modelled conversions. With consent restrictions and cross-device gaps, a growing share of reported conversions are estimated rather than observed. Reasonable estimates, but estimates, and each platform models only its own share.
The decision this breaks
Allocation. Optimising inside a channel using that channel's own numbers is legitimate, because you are comparing like with like. Moving budget between channels on those same numbers is not, because you are comparing two claims that both include the same customers. This one distinction resolves most cross-channel reporting arguments.
The metrics worth managing to
Most reporting suffers from too many metrics rather than too few. Here is the short list that actually drives decisions, and what each one is for.
| Metric | Calculation | Answers | Fails when |
|---|---|---|---|
| Cost per acquisition | Cost ÷ conversions | Are we inside the ceiling? | Conversions include repeat buyers or micro-events |
| New-customer CAC | Cost ÷ first-time customers | What is growth actually costing? | You have no way to flag first-time buyers |
| ROAS | Revenue ÷ ad spend | Channel-level efficiency | Read without margin; 4x is great or fatal |
| MER | Total revenue ÷ total media spend | Is the whole programme working? | Non-paid growth moves it and gets miscredited |
| Contribution margin | Revenue − COGS − variable − media | Did we make money? | Rarely calculated, which is the problem |
| Payback period | CAC ÷ monthly contribution | Can we fund the growth? | Ignored until cash gets tight |
Two of these deserve more attention than they usually get. New-customer CAC is the honest version of cost per acquisition, and the gap between the two tells you how much of your paid performance is acquisition versus retention wearing acquisition's clothing. Contribution margin after media is the only line here that answers whether the programme made money, and it is astonishing how many mature accounts have never calculated it.
Return on ad spend is meaningless without margin
A 4x return is comfortably profitable at a 70 percent gross margin and loses money on every order at 22 percent. Any report that shows return on ad spend without the break-even line drawn on the same chart is inviting a bad decision. Put break-even on the chart.
Blended reporting: the number that cannot be gamed
Blended reporting takes total media spend from your invoices and total new-customer revenue from your own systems, and divides one by the other. It knows nothing about attribution, ignores every platform's claims, and cannot be improved by changing a setting. That crudeness is precisely what makes it valuable.
# Illustrative. AUD, week ending. Source: your systems,
# not the ad platforms. Spend from invoices, not dashboards.
Total media spend $31,400
Total revenue $196,800
New-customer revenue $121,900
New customers 290
MER 196,800 / 31,400 = 6.27
aMER 121,900 / 31,400 = 3.88 # acquisition only
Blended CAC 31,400 / 290 = $108
# Against an allowable CAC of $131: inside the ceiling.
# Sum of platform-reported revenue this week: $274,100.
# That is 139% of actual. Nothing is broken. That is
# what happens when three platforms each claim a share
# of the same customers.
Track the acquisition-only ratio alongside the headline one. Total marketing efficiency ratio improves whenever retention improves, which is good news that has nothing to do with whether your acquisition spend is working. Separating the two stops a strong repeat-purchase quarter from disguising deteriorating acquisition.
What blended reporting cannot tell you
It cannot tell you which channel caused the change, and it will not distinguish paid-driven growth from a good PR week, a seasonal lift or a competitor's outage. It is a thermometer rather than a diagnosis. Used correctly, blended sits at the allocation layer: if the blended number is holding while you shift budget between channels, the shift is working regardless of what the platform dashboards claim about it.
Use spend from invoices
Platform dashboards report spend before credits, rebates, currency conversion and agency or consultant fees. Blended reporting built on dashboard spend understates true cost, sometimes by ten percent or more. Take the number from what actually left the bank account.
Attribution windows are a choice, not a fact
Attribution settings do not discover truth. They apply a rule. Changing a window from 7-day click to 30-day click does not make more conversions happen; it changes which ones the platform is willing to claim. Treating those settings as measurements rather than policies causes an enormous amount of confusion.
| Setting | What it actually does | Sensible default |
|---|---|---|
| Click window | How long after a click a conversion may be claimed | Match your real consideration cycle, then leave it alone |
| View window | Whether impressions with no click may be credited | Keep, but report separately — never blend into a click number |
| Attribution model | How credit is divided among touchpoints in one platform | Consistency matters more than which model |
| Conversion counting | Every conversion, or one per click | One for lead generation, every for e-commerce |
| Primary vs secondary | Which actions bidding optimises toward | Only genuine business outcomes as primary |
Three rules make attribution settings useful instead of a source of argument:
- Write them down. One document listing every platform's window, model and counting setting, with the date each was last changed. Undocumented settings drift, and then nobody can explain a year-on-year comparison.
- Change them rarely, and annotate when you do. Any change breaks comparability. Mark the date on every chart so the step-change is not later mistaken for a performance event.
- Never compare across platforms without stating the settings. A 30-day-click channel will always look better than a 7-day-click channel, and the difference tells you nothing about which one is performing.
The underlying signal quality matters more than the model on top of it. Enhanced conversions, consent mode, server-side tagging and offline conversion imports all determine how much the platform can observe rather than estimate. The conversion tracking guide covers that signal path in detail.
Incrementality: would this have happened anyway?
This is the question attribution cannot answer at all. Attribution asks which touchpoint should get credit for a conversion that happened. Incrementality asks whether the conversion would have happened without any paid touchpoint at all. For branded search, retargeting and broad prospecting, the difference between those two questions is often most of the budget.
Geo holdout
The most practical test for most Australian advertisers. Split comparable regions into test and control, switch the channel off in control, and run for long enough to clear the consideration cycle. Two to four weeks minimum, longer for considered purchases.
What makes it work: matched regions on pre-period conversion rate and volume, enough volume in each arm to detect the effect you care about, and no other campaign changes during the window. What breaks it: running it over a holiday period, using regions with different competitive dynamics, or stopping early because the first week looks alarming.
Platform lift studies
Both Google and Meta offer conversion lift testing that holds out a randomised portion of your audience. Cleaner randomisation than a geo test and considerably easier to set up, with the obvious caveat that the platform is measuring its own effectiveness. Useful, worth running, and worth pairing with an independent test before making a large decision on the result.
The switch-off test
Crude and often the most informative. Turn a channel off entirely for two to four weeks and watch total new-customer volume rather than that channel's reported numbers. Most useful on branded search and retargeting, which are the two line items most likely to be harvesting demand that already existed.
What people expect versus what they find
The uncomfortable finding is usually that retargeting and branded search are substantially less incremental than their reported return implies, and that upper-funnel video and prospecting are substantially more incremental than theirs. This is why programmes optimised purely on platform-reported efficiency tend to converge on a highly efficient, slowly shrinking business.
Incrementality testing is expensive in attention and occasionally in revenue, so run it where the stakes justify it: before a large reallocation, on any channel consuming more than about 15 percent of budget, and annually on branded search.
Building the stack
You need four things, and none of them requires an enterprise platform.
- A clean signal layer. Server-side tagging where practical, consent mode configured properly, enhanced conversions on, and offline conversion imports for anything with a sales cycle. This determines the quality of everything above it.
- A source of truth. One system that owns the definition of a customer and an order, usually the e-commerce platform or CRM. When numbers disagree, this one wins. Naming it in advance prevents most reporting arguments.
- A first-party order flag. Some way to mark each order as first-time or repeat. Without it, new-customer CAC cannot be calculated and half the metrics above are unavailable.
- A weekly blended sheet. Spend from invoices, revenue and new customers from the source of truth, sitting alongside each platform's claim so the gap is visible rather than discovered.
# Run this once a month. Illustrative figures, AUD.
PLATFORM CLAIMS
Google Ads $612,000
Meta $398,000
Other $84,000
Sum of claims $1,094,000
SOURCE OF TRUTH
Total revenue $812,000
New-customer revenue $503,000
THE GAP
Claims as % of actual revenue 135%
Expected range for 3 channels: 120-160%
Above 200% -> check view-through and windows
Below 100% -> check tracking, not attribution
THE DECISION NUMBERS
Media spend (invoiced) $131,500
Blended new-customer CAC $118
Allowable CAC $131
Status: inside ceiling, room to scale
A gap that sits stable month to month is normal and can be ignored. A gap that moves sharply is a signal worth chasing: it usually means an attribution setting changed, a tag broke, or the channel mix shifted toward something with a longer window.
What to report, and how often
Reporting fails in two directions. Too frequent, and you react to noise. Too detailed, and nobody reads it. The fix is to match each cadence to the decisions it is meant to support.
| Cadence | Audience | Contains | Decides |
|---|---|---|---|
| Weekly | Whoever runs the accounts | Spend pacing, blended CAC, anomalies | Nothing structural — catch breakage |
| Monthly | Marketing lead | Channel CAC vs ceiling, reconciliation, impression share | Allocation between channels |
| Quarterly | Leadership or board | Contribution after media, payback, incrementality results | Total budget, channel set, allowable CAC |
The quarterly report is the one that matters and the one most often missing. It is also the only one where contribution margin after media belongs, because that is the number that answers whether the programme made money rather than whether it looked busy.
Put the caveats in the report
State the attribution windows, note where numbers are modelled, and flag any settings changed during the period. A report that presents modelled conversions as observed fact is not more confident, it is less useful, and the credibility cost lands the first time someone checks.
Frequently asked questions
Why do my ad platforms report more revenue than my store?+
Because each platform claims conversions independently under its own attribution rules, with no knowledge of the others. A customer touched by three channels is counted by all three. Add view-through conversions and differing windows and the sum routinely reaches 120 to 200 percent of actual revenue. This is expected behaviour rather than a tracking fault, and the fix is to stop summing platform numbers rather than to try to make them agree.
What is a good MER?+
It depends entirely on gross margin and on how much of your revenue is not driven by paid media, so cross-business comparisons are close to useless. What matters is your own trend against your own break-even, which you can calculate from contribution margin. An e-commerce business at 60 percent gross margin might break even around 1.7 and target 3 or above, while a low-margin retailer needs considerably more.
Should I use a multi-touch attribution tool?+
Rarely, and almost never before the basics are in place. Multi-touch tools depend on cross-site identity resolution that privacy changes have steadily degraded, and they produce a precise-looking number built on assumptions most users never examine. Clean first-party tracking, honest blended reporting and periodic incrementality tests will outperform a multi-touch deployment in most businesses, at a fraction of the cost and complexity.
How do I measure paid media with a long sales cycle?+
Import closed-won revenue back into the ad platforms through offline conversion imports, so bidding optimises toward revenue rather than form fills. In the meantime, identify an upstream event with a demonstrated relationship to closed revenue and optimise to that, while tracking the conversion rate between the two. Reporting lags the sales cycle, so compare cohorts by lead-creation month rather than by close month.
How often should I run incrementality tests?+
Annually on branded search, before any large reallocation, and on any channel consuming more than roughly 15 percent of budget. Testing is expensive in attention and sometimes in revenue, so reserve it for decisions where the answer would actually change what you do. A test whose result would not change anything is not worth running.
Which number should I show the board?+
Contribution margin after media, new-customer CAC against the allowable ceiling, and payback period. Three numbers that describe whether the programme made money, whether it is inside its limits, and whether it can be funded. Return on ad spend by channel belongs in the appendix, because it invites comparisons between platforms whose numbers were never comparable.
Where this fits
Run the numbers
The paid media metrics calculator derives every figure in the table above from spend, impressions, clicks, conversions and revenue, including profit on ad spend. The CAC payback calculator covers acquisition cost, lifetime value and payback period.
Measurement is not a reporting task bolted on after launch. It decides which decisions you are able to make, and the layer you use determines whether a decision is sound or merely confident. Platform numbers for optimisation. Blended for allocation. Incrementality for existence.
The framework sits underneath the strategy and the budget, because neither can be judged without it. If you would like the current state of your measurement assessed alongside the accounts themselves, the free audit starts exactly there, and paid media consulting engagements treat it as the first piece of work.