Marketing Attribution, Explained Simply

Marketing attribution is the practice of deciding which marketing touches deserve credit for a sale. Every attribution model is a rule you choose, not a fact your data hands you. The reason to pick one is to make spending decisions you can defend in a board meeting, not to produce a dashboard where the numbers happen to flatter whoever built it.

Most attribution arguments are not really arguments about data. They are arguments about credit, and about whose budget survives the next quarter. That is why this topic belongs in an accountability conversation rather than an analytics one.

Attribution Is a Rule, Not a Measurement

Almost every downstream confusion traces back to this one point.

Picture a buyer. She sees a LinkedIn post in March. In April she reads two of your articles. In June she searches your brand name, clicks a paid ad, and buys.

Which of those four touches earned the sale? There is no correct answer. There is only a rule you apply the same way every month. Attribution software does not discover the truth about that buyer's decision. It applies your rule and prints the result.

Once a marketing team understands that, two things change. Arguments get shorter, because everyone stops pretending the tool settled the question. And the choice of model becomes what it always was: a management decision, made by the person accountable for the budget.

The Models, in Plain English

Last click gives all the credit to the final touch before purchase. It is simple, it is auditable, and it systematically flatters brand search and retargeting while starving everything that created the demand in the first place.

First click gives all the credit to the first touch. It has the opposite bias. Top-of-funnel channels look like heroes and the closing work looks like an afterthought.

Linear splits credit evenly across every touch. Time decay weights later touches more heavily. Position-based, sometimes called U-shaped, hands 40% each to the first and last touch and spreads the rest across the middle. All three are compromises. None of them knows anything about your buyer; they just distribute credit by formula.

Data-driven attribution uses the platform's own modeling to assign fractional credit based on observed conversion paths. It is the most sophisticated option available inside the ad platforms, and it is also the least inspectable. You cannot open it up and check the arithmetic.

That last point matters more since Google narrowed the field. In October 2023 Google removed first click, linear, time decay, and position-based attribution from Google Ads and Google Analytics 4, leaving only last click and data-driven, and it moved properties that had not chosen a setting onto data-driven by default (Search Engine Land, "Google has removed attribution models in GA4"). If nobody on your team made a deliberate choice, a choice was made for you.

Then there is incrementality, which is a different animal entirely. Incrementality testing asks whether a channel produced sales that would not have happened otherwise. You turn spending off in some markets, leave it on in others, and compare. It answers the question executives actually care about. It is also slower, costs real revenue while the test runs, and cannot be done for every line item at once.

Which Model Should You Actually Pick?

The honest answer is that the model matters less than the consistency. Still, three defaults cover most mid-market situations.

If your sales cycle is short and mostly self-serve, last click is defensible. It undercounts your brand work, so pair it with a separate view of branded search volume and direct traffic, which is where that demand shows up.

If your sales cycle runs longer than about sixty days and involves a salesperson, put the model in the CRM rather than the ad platform. Stamp first touch and last touch on the record, report both, and let the gap between them tell you which channels open doors and which ones close them. Two imperfect numbers side by side beat one number pretending to precision.

If you spend more than roughly a third of your budget on a single channel, no model will settle the question and you need a holdout test instead. At that concentration, the risk of being wrong is larger than the cost of running the experiment.

What none of these need is a six-figure attribution platform. Every mid-market team we have seen buy one before fixing the plumbing in step three below ended up with a more expensive version of the same disagreement.

Why the Numbers Never Add Up

Ask any operator who runs paid media across three platforms and they will tell you the same story. Add up the revenue that Meta, Google, and your email tool each claim, and the total exceeds what the bank deposited.

Nothing is broken. Each platform reports on its own conversions, inside its own attribution window, using its own rule. All three saw the same buyer. All three counted her. Nobody double-charged you; three separate systems each answered a question about the same purchase, and no one reconciled the answers.

The fix is unglamorous. One system holds the revenue truth, and that system is almost never an ad platform. It is your CRM or your commerce backend, because that is where money actually lands. Platform-reported numbers become directional inputs for pacing and bidding. They stop being the scoreboard.

We wrote about one version of this problem in detail when Meta reports fewer purchases than a business actually made, which is the same reconciliation gap running in the opposite direction.

What Broke Attribution Between 2021 and Now

Three shifts changed what is technically possible to track, and any agency still describing 2019-era tracking as normal has not updated its assumptions.

Mobile consent

Apple's App Tracking Transparency arrived in April 2021 and required apps to ask permission before tracking users across other companies' apps and sites. Four years on, AppsFlyer put global consent at roughly 50%, up about ten points since launch (AppsFlyer, April 2025). Half of iOS users say yes. Half do not, and their journeys are partially invisible.

The cookie reversal

Third-party cookies did not die, but the ground shifted anyway. In April 2025 Anthony Chavez, VP of Privacy Sandbox at Google, wrote that Chrome would keep its existing approach and would not roll out a standalone third-party cookie prompt (privacysandbox.com, "Next steps for Privacy Sandbox and tracking protections in Chrome"). Google wound down the Privacy Sandbox effort later that year. The result is a stranded middle ground: cookies still work in Chrome, browser-level protections keep tightening elsewhere, and the replacement standards most vendors built toward never arrived.

Committee buying

Buying itself got harder to trace. Gartner's research on complex B2B purchases describes buying groups of six to ten decision makers, each arriving with four or five pieces of independent research they then share internally. Your form captures one name. The other seven read your material, argued about it in a Slack channel you will never see, and left no trace in any system you own.

None of this makes attribution pointless. It makes precision claims suspect. Directional confidence, checked against booked revenue, is the realistic ceiling.

How to Build Attribution You Can Defend

Five steps, in order. Skipping to step four is the most common mistake.

1. Name one source of truth. Pick the system where revenue is recorded and closed, usually the CRM or the commerce platform. Everything else reports to it. Write the choice down somewhere your whole team can see, because the argument you avoid in June is worth more than the hour it takes today.
2. Write down the model and the window. Which model, what lookback period, and which conversions count. Thirty days versus ninety days will change your channel rankings, and a team that never wrote the window down will relitigate every quarterly review. Put both in the same document as step one.
3. Instrument the handoff. Consistent UTM parameters on every paid and email link. A stored source field on every CRM record. Server-side conversion imports so closed-won deals flow back to the ad platforms. This is the boring plumbing work that determines whether the rest of it means anything, and it usually takes an engineer a week.
4. Reconcile monthly. Put platform-reported revenue and booked revenue side by side, every month, and explain the gap out loud. A steady gap is fine; you now know your correction factor. A gap that moves without explanation is a tracking problem, and you have found it early.
5. Test incrementality on the biggest line item. Take whichever channel eats the most budget and run a geo holdout, or a scheduled pause, or a matched-market test. One test per quarter is plenty. It will teach you more than a year of dashboard refinement, because it is the only method here that answers whether the spending caused anything.

Most mid-market teams can finish the first four steps in a month. The fifth is a habit, not a project.

What This Means for Your Agency Relationship

Attribution is where agency accountability either exists or quietly does not, so a few questions separate the two.

Ask which model reports on your account, and why that one. A partner who owns the answer will name the model, name the tradeoff, and tell you which channel it flatters. A vendor will say "we use data-driven" and change the subject.

Ask them to reconcile their reported numbers against your closed revenue. Not once, in front of the CFO, as a challenge. Monthly, as routine. The agencies who resist this are usually the ones whose reports need the gap to stay unexamined.

Ask what they would stop spending on if attribution said it was not working. An honest answer names a channel. That is uncomfortable to say out loud, and it is the whole point of measuring anything. Our piece on the tough questions to ask your marketing agency covers the rest of that conversation.

Watch what they report on. Reports built on impressions, reach, and engagement are reports designed to avoid this subject entirely. The metrics that predict revenue look different and behave differently.

The Bottom Line

Attribution will not tell you the truth about why someone bought. It will give you a consistent, defensible rule for allocating credit, which is enough to run a budget on as long as everyone knows which rule is in play.

Gartner's 2026 CMO Spend Survey, based on 401 marketing leaders surveyed between January and March 2026, put average marketing budgets at 7.8% of company revenue, and found 56% of CMOs saying they lack the budget to deliver their 2026 strategy (Gartner, May 2026). When money is that tight, the cost of a measurement system nobody trusts is not the software fee. It is the quarter spent defending spending you cannot explain.

Pick your model on purpose. Reconcile against money. Test the big line items. Everything after that is refinement.

If you want an outside read on whether your current setup would survive that scrutiny, our marketing audit covers exactly this ground.