Marketing attribution is the practice of assigning credit for a conversion to the marketing touchpoints that led to it — and the reason your Google, Meta and analytics dashboards never agree is that each uses a different attribution model, and no model is the truth. Attribution is an estimate built on assumptions about which touch mattered, not a measurement of cause. For executives, the practical implication is to stop hunting for one perfect number and instead triangulate: use platform attribution to optimise inside channels, and use incrementality tests and marketing-mix modelling to make budget decisions.
TL;DR
- No attribution model is the truth. Each assigns credit by a different rule, so dashboards disagree by design.
- Last-click over-credits the final touch (often brand search) and under-credits demand creation.
- Data-driven attribution may allocate credit better than last-click, but it is still a model shaped by observable data and configuration.
- Adding platform-reported conversions overstates the shared total. Platforms use different windows, eligible interactions, and identity signals, so the same order may appear in several reports.
- For budget decisions, use incrementality and MMM, which answer "what would have happened without this spend?" — the question attribution cannot.
- Executives should triangulate, not adopt one model — attribution to steer channels, incrementality/MMM to allocate budget.
- The goal is better decisions, not a perfect number that does not exist.
Why the dashboards disagree
If Google Ads reports 100 conversions, Meta reports 80, and your analytics platform reports 140 total, none of them is lying — they are answering different questions with different rules. Google counts conversions its attribution model credits to Google; Meta counts what its model credits to Meta; analytics applies its own model to everything. Because a single customer's journey touches several channels, the same conversion gets claimed, in whole or part, by more than one system.
Attribution is fundamentally an assignment problem with no correct answer. A buyer sees a Meta ad, later searches the brand on Google, clicks an email, and buys. Which touch "caused" the sale? Any answer is a modelling choice: last-click credits the email, first-click credits Meta, linear splits it evenly, data-driven estimates weights from patterns. All are defensible; none is fact. The disagreement between dashboards is not a bug to fix — it is the visible result of different assumptions applied to the same ambiguous reality.
Once a leadership team internalises this, the frustration of "why don't the numbers match?" turns into the right question: "which view should we trust for which decision?"
Align the data scope before debating the model
Changing an attribution model will not reconcile reports that describe different sets of events. Before evaluating a channel, align at least six variables:
- Conversion definition — a purchase, qualified opportunity, and raw form submission are different outcomes.
- Business source — the ad platform, GA4, CRM, billing system, or commerce backend.
- Reporting time — interaction date versus conversion date.
- Lookback window — how long a touchpoint remains eligible for credit.
- Eligible interaction — click, engaged video view, or view-through exposure.
- Identity and consent — which journeys can be joined across devices and which events are observed or modeled.
GA4 adds another source of confusion: attribution depends on dimension scope. Session-scoped dimensions continue to use paid and organic last click, while event-scoped dimensions use the property's selected reporting model, which is data-driven by default. Two reports in the same property can therefore differ without either being broken.
The operating rule is simple: reconcile the numerator, denominator, scope, and clock before comparing attribution models. Otherwise, a debate about credit hides a more basic data-contract problem.
The attribution models and what each hides
| Model | How it assigns credit | What it hides |
|---|---|---|
| Last-click | All credit to the final touch | Everything that created the demand |
| First-click | All credit to the first touch | Everything that closed the sale |
| Linear | Even split across touches | Which touches actually mattered |
| Time-decay | More credit to recent touches | Early demand creation |
| Position-based | Most to first and last | The considered middle |
| Data-driven | Modelled weights from patterns | Still one platform's partial view |
First-click, linear, time-decay, and position-based remain useful concepts for diagnosing bias, but Google removed them from GA4 and Google Ads in 2023. GA4 now offers data-driven, paid and organic last click, and Google paid channels last click; Google Ads supports data-driven and last click. A legacy dashboard may therefore name a model that is no longer selectable in the current product.
Last-click deserves special mention because it is still the default many teams live by, and it systematically misleads. It hands credit to the last click before conversion — usually a branded search or direct visit — which means the channels that created the demand (social, video, thought leadership) look worthless while the channels that merely harvested it look brilliant. Cutting budget on that basis defunds the demand creation that fills the top of the funnel, and results decay a quarter later for reasons the last-click report cannot explain.
Data-driven attribution is a genuine improvement — it estimates credit from converting and non-converting paths rather than a fixed rule. It is not, however, a randomized causal experiment. Its output depends on the paths a system can observe or model, the channels eligible to receive credit, and the selected lookback window. It also does not solve duplicate credit across independent platforms.
What attribution cannot answer — and what can
The question executives actually care about is causal: did this spend cause incremental revenue, or would that revenue have happened anyway? Attribution cannot answer this, because it only distributes credit among touches that preceded conversions it can see — it never observes the counterfactual.
Two methods do address causation:
- Incrementality testing — deliberately withholding spend from a group (a geo holdout, an audience holdout) and comparing outcomes. The difference is the incremental effect the spend actually caused. This is the closest thing to a controlled experiment in marketing.
- Marketing-mix modelling (MMM) — a statistical model relating spend across channels (and external factors) to business outcomes over time, estimating each channel's contribution without relying on user-level tracking.
These answer the budget question — how much did this channel really contribute, and what happens if we move money — far more defensibly than arguing over attribution settings. They are covered in depth in incrementality testing and marketing-mix modelling. The trade-off is that they are slower and need scale, which is why attribution still has a job.
A four-layer measurement system
A mature measurement system does not force one report to answer every question. It combines four layers that challenge and calibrate one another:
| Layer | Best question | Typical decision | Main limitation |
|---|---|---|---|
| Finance system / CRM | How much revenue, margin, and new-customer value actually occurred? | Company performance | Does not isolate channel impact by itself |
| GA4 / product analytics | How do users move through and convert on the site or product? | Funnel and journey improvements | Consent, identity, and dimension-scope gaps |
| Ad platforms | Which campaigns and signals help the delivery system achieve its objective? | Daily campaign optimization | Platform-specific view and credit rules |
| Experiments and MMM | Which outcomes were incremental, and how might spend respond? | Budget allocation and scenarios | Scale, time, and data-quality requirements |
A “source of truth” is appropriate for an accounting fact, such as recognized revenue. It is not a complete explanation of causality. The CRM cannot observe every ad exposure, while an ad platform does not know full contribution margin or how much demand would have converted without advertising. Each layer therefore needs a declared role, owner, refresh cadence, and decision threshold.
How to reconcile conflicting marketing reports
When the gap becomes materially larger than normal, use a fixed diagnostic order:
- Confirm the base volume in the CRM, billing system, or commerce backend. This validates actual outcomes, not marketing credit.
- Compare the same event definition. Separate primary outcomes from micro-conversions and qualified opportunities from raw leads.
- Align time zone, currency, tax treatment, refunds, and order status. A gross-versus-net difference can masquerade as attribution drift.
- Check report scope and reporting time. Interaction-time and conversion-time reports answer different questions.
- Record the model, lookback window, and eligible interaction types for every tool. Without this measurement contract, the comparison cannot be reproduced.
- Audit identity and deduplication. UTMs, click IDs, transaction IDs, event deduplication, and CRM joins determine how much of the journey can be connected.
- Interpret the remaining gap only after the controls pass. Unexplained variance should not automatically be labelled a platform error.
This sequence turns a recurring dashboard argument into an auditable process. More importantly, it separates instrumentation defects from legitimate differences in model scope.
Glossary
- Attribution — assigning credit for a conversion across the touchpoints that preceded it.
- Last-click — a model giving all credit to the final touch before conversion.
- Data-driven attribution — a model estimating credit weights from observed conversion patterns.
- Incrementality — the additional outcome caused by spend versus what would have happened without it.
- Marketing-mix modelling (MMM) — statistical modelling of spend-to-outcome relationships over time.
- Counterfactual — what would have happened absent the marketing, the thing attribution never observes.
How executives should actually use attribution
The mature stance is not to pick a winning model but to use each tool for the decision it fits.
- Use platform attribution (including data-driven) to optimise within a channel — which campaigns, audiences and creatives to scale or cut inside Google or Meta. It is good enough for steering.
- Use MER and blended views for the whole-business efficiency check — is total marketing producing proportional revenue, without the double-counting of summed platform numbers. See MER vs ROAS.
- Use incrementality and MMM for budget allocation — the high-stakes decisions about how much to spend and where, which need a causal answer attribution cannot give.
- Treat platform attribution as directional, not gospel — especially for demand-creation channels that last-click structurally undervalues.
The failure mode is using one tool for everything: optimising budget allocation on last-click (defunds demand creation), or trying to run daily campaign decisions off a slow MMM. Match the measurement to the decision's stakes and speed.
How Space Ads approaches attribution
We start by agreeing the business outcome with the leadership team: recognized revenue, contribution margin, a new customer, or a sales-qualified opportunity. We then verify that GA4, the CRM, and ad platforms receive the intended event with a stable identifier and no duplication. Only after that do we compare models and lookback windows. Platform data steers daily campaign decisions; CRM outcomes and blended metrics control the whole business. For a material budget reallocation, we recommend an incrementality test or MMM when scale and data history make the method credible. The reporting layer keeps disagreements visible instead of blending them into false precision, so decision-makers can distinguish a transaction fact, modeled credit, and an estimate of causal impact. This is the measurement foundation behind our web analytics, performance marketing, and fractional CMO work.
Stop doing / Do instead
| Stop doing | Do instead |
|---|---|
| Treating one attribution model as the truth | Triangulate — different tools for different decisions |
| Making budget decisions on last-click | Use incrementality and MMM for allocation |
| Cutting channels last-click undervalues | Check demand-creation value with incrementality |
| Summing platform-attributed conversions | Use MER for the whole-business efficiency view |
| Arguing over attribution settings | Run a holdout test to get a causal answer |
| Expecting dashboards to agree | Understand each model's bias and use it accordingly |
FAQ
What is marketing attribution?
Marketing attribution is the practice of assigning credit for a conversion to the marketing touchpoints that preceded it. Because a customer journey usually spans several channels, attribution uses a model — last-click, data-driven, and others — to decide how to split the credit. It is an estimate based on assumptions, not a measurement of cause.
Why do Google, Meta and analytics report different numbers?
Because each uses a different attribution model and sees a different slice of the journey. A single conversion can be claimed, in whole or part, by multiple platforms, so their totals disagree by design. None is lying; they are answering different questions with different rules.
What is wrong with last-click attribution?
Last-click gives all credit to the final touch before conversion, usually a branded search or direct visit. That over-credits channels that harvest existing demand and under-credits channels that create demand (social, video, thought leadership). Cutting budget on last-click defunds the demand creation that fills the funnel, and results decay later.
Is data-driven attribution better?
It is an improvement — it estimates credit from observed patterns rather than a fixed rule — but it is still a model, and it still only sees one platform's touchpoints. It does not solve cross-channel over-claiming or answer whether spend was incremental; it makes one platform's internal story more sophisticated.
How should executives measure marketing?
Triangulate. Use platform and data-driven attribution to optimise within channels, MER for the whole-business efficiency check, and incrementality tests or marketing-mix modelling for budget allocation, which needs a causal answer attribution cannot provide. Match the measurement tool to the decision's stakes and speed.
What is the difference between attribution and incrementality?
Attribution distributes credit among touchpoints that preceded conversions it can observe. Incrementality measures the additional outcome caused by spend versus what would have happened without it, by withholding spend from a control group. Attribution answers "which touch gets credit?"; incrementality answers "did this spend cause anything?"
Key takeaways
- Attribution assigns credit by a model; no model is the truth, so dashboards disagree by design.
- Last-click over-credits demand harvesting and under-credits demand creation.
- Data-driven attribution is better but still model-based and platform-limited.
- Attribution cannot answer causation — use incrementality and MMM for budget decisions.
- Executives should triangulate: attribution to steer, MER to check, incrementality/MMM to allocate.
Sources and further reading
- Google Analytics Help — Attribution and currently available models
- Google Analytics Help — Traffic-source dimension scopes and their attribution models
- Google Analytics Help — Reporting model, eligible channels, and lookback-window settings
- Google Ads Help — Attribution models in Google Ads
- Google for Developers — Meridian open-source marketing mix modeling
- Meta Marketing Science — Robyn open-source marketing mix modeling
Continue learning
- Incrementality testing: geo experiments across Meta and Google
- Marketing-mix modelling with Meridian and Robyn
- MER vs ROAS: the blended metric your board should track
- Server-side tagging and the Conversions API
- Web analytics that ties spend to revenue
- Fractional CMO: measurement a board can decide on
Continue reading
Your First Senior Marketing Hire: CMO, VP, or Fractional?
When founder-led marketing hits its ceiling, the instinct is to hire a CMO. Often that is too senior, too early, and too expensive — a VP who still executes, or a fractional lead for direction plus specialist hands, fits better until the company is large enough for a full CMO to have leverage.
Who Owns Your Google Ads, Meta, and GA4 When You Leave an Agency?
You should own your ad accounts and data; the agency is a revocable manager, not the owner. But access is not ownership — the billing profile and account structure decide. This is the platform-by-platform mechanics of what you keep, what you can lose, and how to check before it's too late.
SMS & RCS Marketing: Revenue From a List You Already Own
SMS and RCS are owned channels: revenue from a list you already have, not paid media you rent. They deliver immediacy and high engagement, but only work with clean consent, disciplined frequency and RCS's branded, richer messaging replacing plain texts.
