A customer sees your Instagram ad on Monday, googles you on Wednesday, clicks a search ad, reads reviews, and finally buys on Friday from a link in your email. One sale. Now open your dashboards: Meta claims it, Google claims it, your email tool claims it, and your analytics has its own opinion. Add up what every platform reports and you've apparently sold the product three times.
- Attribution is a modelling choice, not a measurement — different models crown different channels from identical data.
- Last-click over-credits closers like brand search and retargeting and erases demand creation.
- Anchor on blended MER and escalate to incrementality tests when decisions matter.
Nobody is lying. Each platform is answering honestly under its own rules about who deserves credit. That's the trap: attribution is not a measurement — it's a modelling choice, and different models produce different winners from identical facts.
The models, in one glance
All attribution models are just credit-splitting rules applied to the same journey:
- Last-click gives everything to the final touch. Simple, standard — and it systematically crowns the closers (brand search, retargeting, email) while erasing whatever created the demand.
- First-click gives everything to the introduction. Great for understanding discovery, useless for understanding closing.
- Linear splits credit evenly — diplomatic, and about as decisive as it sounds.
- Time-decay weights later touches more heavily — a compromise with the same closer bias, softened.
- Data-driven lets an algorithm assign weights — smarter, but opaque, and still built on incomplete tracking.
Change the rule, change the hero. The journey never changed at all.
ROAS profit calculator
Enter your real numbers. This works out whether your ROAS is actually making money â or quietly losing it.
Adjust the numbers above.
Why this ruins real decisions
The trap springs when someone treats one model's output as objective truth. A team on last-click concludes that upper-funnel video “doesn't convert,” kills it, and then watches brand-search volume quietly sag a quarter later — because the thing that created the searches is gone. The dashboard applauded the cut right up until the damage arrived. Meanwhile platforms grade their own homework with generous view-through windows, which is how three channels can each claim the same rupee of revenue with a straight face.
Two honest analysts, identical data, different winners. If your conclusion changes with the model, the model is the conclusion.
How working marketers stay sane
- Pick one model as your operating lens and be loudly aware of its bias. Last-click undervalues demand creation — adjust with judgement, not faith.
- Compare models occasionally. Where first-click and last-click wildly disagree about a channel, that channel is doing an unglamorous job one lens can't see.
- Anchor on blended reality. MER — total revenue over total spend — ignores attribution entirely. When platform ROAS climbs but MER doesn't, the platforms are trading credit, not creating sales.
- Escalate to incrementality tests for the big-money questions. Causation beats credit assignment every time.
The one-line takeaway: Attribution models are lenses, not truths. Know which lens you're using, check the others when decisions matter — and let blended revenue arbitrate when the dashboards start arguing.
Interviewers ask about attribution not because they expect you to solve it — nobody has — but to see whether you know it's unsolved. The candidate who says “here's the model I'd use, here's its bias, and here's how I'd sanity-check it” sounds like an operator. Now you can be that candidate.
What to say when the platforms disagree in a meeting
Sooner or later you will sit in a room where Meta claims one thing and Google claims another and someone senior asks which is right. The wrong answer is to defend a platform. The right answer sounds like this: “Both are reporting honestly under their own credit rules, and they overlap — if we add them up we will double-count. Here is total revenue over total spend, which cannot double-count, and here is what it says.”
Then offer the next step rather than the argument: a hold-out or geo test on whichever channel the decision hinges on. This reframes the discussion from “whose dashboard do we believe” to “what would we have to observe to know” — and that shift is one of the fastest credibility gains available to a marketer at any level.
The double-count, in actual numbers
Make the trap concrete for a second. Say you spend ₹1,00,000 across Meta, Google and email in a month, and your bank statement shows ₹4,00,000 of real revenue. Now open the dashboards: Meta reports ₹2,40,000, Google reports ₹2,00,000, and email reports ₹1,20,000 — a cheerful ₹5,60,000 in total. That is ₹1,60,000 of revenue that exists nowhere in your account. Nobody invented it; the same customers were simply counted again by every platform they happened to touch on the way to a single purchase.
Now watch what MER does to the panic. Total revenue (₹4,00,000) over total spend (₹1,00,000) is a blended 4.0 — a figure that physically cannot double-count, because it begins from money that actually landed. Trust the summed dashboards instead and you’d act on an apparent 5.6, scaling budget toward a profit that was never there. When those two numbers drift apart, believe the one tied to your bank balance, not the three tied to a pixel each grading its own homework.
The figure that should genuinely worry you is the gap between them. A widening spread between what the dashboards add up to and your blended MER usually means the platforms are competing harder to claim credit — more retargeting, more brand search — while the total barely grows. If you pour in more budget every time one dashboard’s ROAS ticks up, you’re paying to shuffle credit around, not to create customers. Track MER month over month; it is the one scoreboard that cannot flatter you.