Attribution is an assumption, not a fact

Every attribution report is an argument.

It is an argument about which interactions deserve credit, over what window, under what rules, with what confidence. The spreadsheet formatting can make that argument look like a fact. It is still a model.

This matters because teams make budget decisions as if the model were the market. Then they are surprised when cash, pipeline quality, or incrementality tests disagree.

Attribution is useful. It is not certainty.

Platforms are not neutral witnesses

Each ad platform has an incentive to demonstrate value. That does not make the people building the reports dishonest. It does mean self-attribution should be read as interested evidence.

A click-based model, a view-through window, a modeled conversion in a privacy-constrained environment: each choice changes who looks effective.

Last click tends to reward the end of the journey. First click tends to reward the introduction. Position-based and data-driven models distribute credit differently again. None of them invents ground truth. They organize incomplete observations.

If your Google numbers and Meta numbers and CRM numbers each claim the same revenue, you do not have triple confirmation. You may have triple counting risk.

Multi-touch journeys expose the fantasy of a single cause

Most considered purchases are not one-touch events.

A person may see a social ad, search the category a week later, click an email, return through branded search, and convert on a retargeting offer. Asking which ad “caused” the sale is often the wrong question. The better questions are which investments made the journey more likely, which ones are redundant, and which ones look productive only because they sit near the purchase.

First-click thinking fails here in an obvious way. So does last-click thinking. Both can still be useful as lenses. Trouble starts when one lens becomes the budget constitution.

Lead generation adds another layer. A form fill is not revenue. Sales acceptance, opportunity creation, and closed-won outcomes can reorder which channels looked efficient at the click or MQL stage. If marketing optimizes for the early proxy while sales lives with the later truth, the organization will fight over “lead quality” indefinitely.

Better tracking improves evidence. It does not eliminate judgment.

Server-side collection, cleaner first-party events, and disciplined CRM joining are worth doing. They reduce loss, duplication, and blind spots. They make the argument higher quality.

They do not produce causal certainty on demand.

Incrementality testing, geo splits, holdouts, and careful before-after reads with controls get closer to causality. Even those require judgment about design, contamination, and what the business can afford to learn.

I am careful with this claim because the industry likes magic words. “Attribution solved.” “AI-assigned credit.” “Single source of truth.” Useful systems can improve decision quality. No system removes the need to interpret conflicting evidence.

How I use attribution without worshipping it

A practical operating stance:

1. Treat each platform’s conversion report as directional evidence inside that channel.
2. Reconcile to business systems for revenue, margin, and pipeline outcomes.
3. Expect disagreement, and investigate the disagreement instead of forcing one number to win every meeting.
4. Use experiments when the decision is large enough to justify the cost of learning.
5. Document the attribution assumptions behind any CAC or ROAS target.

That fifth point is underrated. Teams argue about performance when they are actually arguing about windows, models, and definitions.

If your target assumes a seven-day click window and the platform view includes a broader view-through definition, you are not in the same conversation.

What founders should ask vendors and teams

Before the next measurement debate, ask:

• Which events are counted as conversions, and where do they originate?
• What is the lookback window, and why?
• How are view-through conversions treated?
• How do we prevent double counting across platforms?
• Which decisions are we willing to make on modeled data, and which require stronger causal evidence?

These questions are not anti-platform. They are pro-judgment.

You can see how ADSRUNNER thinks about first-party tracking and attribution inside the platform. The useful ambition is clearer evidence for human decisions, not a black box that claims the argument is over.

The sentence worth remembering is this.

Attribution can organize your evidence. It cannot replace responsibility for the assumptions underneath it.