What is marketing attribution?

Marketing attribution is the practice of assigning credit for a sale to the marketing touches that preceded it. It matters because budget decisions depend on it: if you cannot say which channel produced revenue, you are funding channels by instinct. The difficulty is that credit is not a physical quantity — a customer who saw an ad, read a review and searched your name was influenced by all three, and no measurement recovers how much each mattered.

Which attribution model is most accurate?

None of them, in the sense of being true. An attribution model is an assumption about how influence works, expressed as arithmetic — last click assumes the final touch did the work, linear assumes all touches are equal, and both assumptions are obviously wrong in most real journeys. The useful question is not which is accurate but which is least misleading for the decision in front of you, and whether your answer changes when you switch models. If it does, you do not have an answer yet.

What is the difference between ROAS and ROI?

ROAS is revenue divided by ad spend. ROI is profit divided by total cost. ROAS ignores your margin and every cost that is not media, so a 4:1 ROAS on a product with a 20% margin is losing money. ROAS is a channel-management number — useful for comparing this week's campaigns against last week's — and ROI is a business number. Reporting ROAS to anyone who thinks in profit is the commonest way marketing performance is overstated without anybody lying.

Is multi-touch attribution worth it for a small business?

Usually not. Multi-touch attribution needs enough conversions for the splits to be stable, and below a few hundred a month you are reading noise with more decimal places. The cheaper method that works at any size: change one thing at a time, keep everything else steady, and watch total results rather than attributed ones. It is slower and it answers the question you actually have, which is whether the spend produced anything.

How do you measure marketing when attribution is broken?

Stop trying to split credit and start looking at whether the whole thing moves. Turn a channel off for a fortnight and watch total sales, not that channel's reported sales. Compare periods where one thing changed. Ask new customers where they heard about you — imprecise, unbiased by any tracking regime, and often the only signal that catches word of mouth. None of this is as satisfying as a dashboard with percentages, and all of it is more honest.

Guide

Marketing attribution: what it can and cannot tell you

How attribution models actually work, why they disagree, and how to act on marketing data when the honest answer is that the credit cannot be split.

Marketing attribution assigns credit for a sale to the marketing touches that came before it — and no attribution model is true. Each one is an assumption about how influence works, written as arithmetic. Last click assumes the final touch did the work; linear assumes every touch mattered equally. Both are obviously wrong about most real customers, which is why two models fed identical data will tell you to fund different channels.

That is not a defect to be tooled away. It is the honest shape of the problem, and knowing it changes how you should read every marketing report you are handed.

The five models, and what each one flatters

Every model is defensible and every model has a channel it favours. Knowing which is the whole skill — because the model you pick decides which team looks good, and people notice that faster than they notice the methodology.

Attribution models: what each assumes, and what it hides
	How it splits credit	What it flatters or hides	When it is defensible
Last click	All credit to the final touch before the sale.	Flatters closing channels — brand search, retargeting — and hides everything that created the demand.	Short sales cycles where one channel genuinely does the work.
First click	All credit to the first touch.	Flatters discovery and ignores everything that got the deal over the line.	Judging whether a top-of-funnel bet is bringing anyone new at all.
Linear	Credit split evenly across every touch.	Treats a newsletter glance and a demo call as equal, which they are not.	A sanity check against last click, not a decision basis.
Time decay	More credit the closer a touch is to the sale.	An assumption about human memory dressed as arithmetic.	Longer cycles where recency plausibly matters.
Data-driven	The platform decides, from its own model.	Unauditable, and the platform grading its own homework has an interest in the answer.	High volume, where you can validate it against a holdout.

The test that matters more than the model

Run your numbers through two models that disagree — last click and first click will do — and see whether your decision changes.

Not an answer

Last click says fund search. First click says fund the podcast. You pick the one that matches what you already wanted to do.

An answer

Both models say the same channel is carrying this, by a wide margin. The disagreement between them is smaller than the gap you are acting on.

If the models disagree about your decision, the data does not support it yet. That is worth knowing before the money moves, and it is the check almost nobody runs.

ROAS is not ROI, and the difference is your margin

ROAS is revenue divided by ad spend. ROI is profit divided by total cost. They are quoted interchangeably and they are not close to the same thing.

A campaign returning 4:1 ROAS on a product with a 20% gross margin earns $0.80 of gross profit for every $1.00 of media — before the agency, the tools, or the hours. It is losing money and reporting a number that sounds like winning. Nobody has lied; the wrong ratio was quoted to an audience thinking in profit.

Use ROAS for channel management, where comparing this week against last week is the job. Use ROI, or cost per customer against what a customer is worth, for anything a business decision rests on.

Why your platforms will never agree

Google Ads and Google Analytics report different conversion counts for the same campaign, permanently, and it is not a misconfiguration.

Different objects. Ads counts conversions against a click. Analytics counts sessions and applies its own model.
Different dates. Ads credits a conversion to the day of the click; Analytics to the day of the conversion. Over a long cycle these are different months.
Different windows. Each has its own lookback period, and each is configurable independently.
Different losses. Consent choices, ad blockers, iOS restrictions and cross-device journeys remove different subsets from each one.

Pick one as the system of record for each kind of decision, write down which, and stay with it. The gap is worth measuring once so you know its size; chasing it to zero is a project with no payoff.

What to do instead when the numbers cannot settle it

These are less satisfying than a dashboard and considerably more reliable.

Turn it off.Pause a channel for two weeks and watch TOTAL sales, not that channel's reported sales. The cleanest signal available to a small business, and almost nobody runs it.
Change one thing. Hold everything else steady for a period long enough to cover your sales cycle, and compare totals.
Ask."How did you hear about us" on the order form is imprecise, unbiased by any tracking regime, and often the only thing that catches word of mouth at all.

How Rallik handles this

It reports what it can measure and says so when it cannot — which, for anything involving attribution, is more often than most dashboards will admit. Channel performance is compared against that channel's own history rather than against a benchmark or a cross-channel split it cannot honestly compute, and a proposal that rests on an unmeasurable claim says which part is unmeasurable.

That is a smaller promise than most marketing tools make. It is the one that survives contact with your accounts.

Common questions

What is marketing attribution?: Marketing attribution is the practice of assigning credit for a sale to the marketing touches that preceded it. It matters because budget decisions depend on it: if you cannot say which channel produced revenue, you are funding channels by instinct. The difficulty is that credit is not a physical quantity — a customer who saw an ad, read a review and searched your name was influenced by all three, and no measurement recovers how much each mattered.
Which attribution model is most accurate?: None of them, in the sense of being true. An attribution model is an assumption about how influence works, expressed as arithmetic — last click assumes the final touch did the work, linear assumes all touches are equal, and both assumptions are obviously wrong in most real journeys. The useful question is not which is accurate but which is least misleading for the decision in front of you, and whether your answer changes when you switch models. If it does, you do not have an answer yet.
What is the difference between ROAS and ROI?: ROAS is revenue divided by ad spend. ROI is profit divided by total cost. ROAS ignores your margin and every cost that is not media, so a 4:1 ROAS on a product with a 20% margin is losing money. ROAS is a channel-management number — useful for comparing this week's campaigns against last week's — and ROI is a business number. Reporting ROAS to anyone who thinks in profit is the commonest way marketing performance is overstated without anybody lying.
Why do Google Ads and Google Analytics report different numbers?: Because they are counting different things with different rules. Ads counts a conversion against the click that led to it and credits it to the click's date; Analytics counts a session and applies its own attribution model and lookback window. Consent choices, ad blockers and cross-device journeys then remove different subsets from each. The gap is normal and its size is worth knowing, but chasing the two into agreement is work with no payoff — pick one as the system of record for each decision and stay with it.
Is multi-touch attribution worth it for a small business?: Usually not. Multi-touch attribution needs enough conversions for the splits to be stable, and below a few hundred a month you are reading noise with more decimal places. The cheaper method that works at any size: change one thing at a time, keep everything else steady, and watch total results rather than attributed ones. It is slower and it answers the question you actually have, which is whether the spend produced anything.
How do you measure marketing when attribution is broken?: Stop trying to split credit and start looking at whether the whole thing moves. Turn a channel off for a fortnight and watch total sales, not that channel's reported sales. Compare periods where one thing changed. Ask new customers where they heard about you — imprecise, unbiased by any tracking regime, and often the only signal that catches word of mouth. None of this is as satisfying as a dashboard with percentages, and all of it is more honest.