A customer sees a video on Instagram on Monday. On Wednesday they search for the brand on Google and read a few pages. On Saturday a reminder email arrives, and on Sunday they type the website address directly and buy.
Which channel earned the sale?
Every honest answer begins with "it depends on what you mean by earned".
Models are opinions with maths attached
The common attribution models each encode a different belief:
- Last click believes the final touch did the work. It gives the sale to the direct visit, which is like crediting the cashier for the purchase.
- First click believes discovery is everything. It credits the Instagram video and ignores everything that kept the customer interested.
- Linear believes every touch mattered equally — tidy, but rarely true.
- Time decay believes recent touches mattered more.
- Position-based believes the first and last touches mattered most and the middle mattered a little.
None of these is wrong in the sense of being badly calculated. They are simply different assumptions. Change the model and the same data tells a different story — which channel "works", which deserves budget, which should be cut.
The problems underneath the model
Even before you choose a model, the data itself has gaps:
- Tracking loss. Browser privacy features, consent choices and app tracking restrictions mean many journeys are only partly visible.
- Cross-device behaviour. The same person on a phone and a laptop can look like two strangers.
- Offline conversions. Phone calls, showroom visits and WhatsApp chats often never reach the analytics platform.
- Platform self-reporting. Each ad platform measures its own contribution, and their totals regularly add up to more than the sales that actually happened.
So what should a business do?
I treat attribution as a decision tool, not a scoreboard. A few principles help:
Pick one source of truth for revenue. Usually the CRM or the order database — not the ad platforms.
Use more than one lens. Compare platform-reported results, analytics attribution and blended numbers such as total marketing spend divided by total new customers. When they disagree, that disagreement is information.
Close the loop. Send offline outcomes — qualified leads, bookings, sales — back into the ad platforms. This improves both measurement and optimisation.
Test for incrementality. The cleanest way to know whether a channel works is to change it and watch what happens: geographic holdouts, budget pauses, or structured on/off tests. It is less precise than a dashboard, but far more honest.
The question behind the question
Most attribution debates are really budget debates. The useful question is not "which channel gets the credit?" but "if I moved the next ₹1 lakh from here to there, would the business be better off?"
Attribution models can help you form a hypothesis. Only a well-designed test can answer it.