Say a customer buys from you today. Last week they saw an Instagram ad. Three days ago they clicked a Google result. Yesterday they opened your email. Which of those deserves the credit for the sale?
That question is attribution, and it's harder than it looks, mostly because the honest answer is "all of them, somehow." Attribution models are different ways of dividing up that credit. None is perfect. Knowing how each works helps you see where it will mislead you.
The common models
Last click. All the credit goes to the final thing the customer clicked before buying. It's simple and it's the default in a lot of tools. It also ignores everything that built the interest in the first place. Brand search and email, which usually come at the end of the journey, look brilliant under last click.
First click. All the credit goes to the first interaction. This favours channels that introduce people to you, such as social ads or content, and ignores what convinced them to buy.
Linear. Every touch gets an equal share. Fair-sounding, but it treats a fleeting banner impression the same as a long visit to your pricing page.
Time decay. Touches nearer the sale get more credit. Reasonable if your sales cycle is short and recent actions matter most.
Position-based. Typically 40 percent to the first touch, 40 percent to the last and the remaining 20 split between those in the middle. A compromise that values both introduction and closing.
Data-driven. A model that learns from your own conversion paths, comparing journeys that ended in a sale with those that didn't. Available in some platforms, and it can be more realistic, but it's a black box and needs plenty of data.
What the same sale looks like under different models
Take a ₹10,000 order with three touches: social ad, search click, email.
| Model | Social ad | Search | |
|---|---|---|---|
| Last click | 0 | 0 | ₹10,000 |
| First click | ₹10,000 | 0 | 0 |
| Linear | ₹3,333 | ₹3,333 | ₹3,333 |
| Position-based | ₹4,000 | ₹2,000 | ₹4,000 |
Same customer, same order, completely different stories about which channel works. Anyone choosing a model is also choosing which channels will look good.
What attribution can't see
Be honest about the gaps. Cookies get blocked or expire. People move between phone and laptop and look like different visitors. Privacy rules and consent choices remove users from the data altogether. And offline influences like recommendations and in-store visits don't show up at all.
Add to that a subtler problem. Attribution measures who was present, not who caused the sale. Someone who was going to buy anyway and typed your brand name into Google gives search all the credit. Whether the ad changed anything is a different question, and tracking alone can't answer it.
A sensible way to start
Pick a model and stay with it. The goal is consistency, so you can compare this month with last month. Position-based or time decay are decent starting points for most businesses.
Look at two models side by side. Compare last click and first click for your main channels. Channels that look much better under one than the other are telling you about their role: closers or introducers.
Make sure the basics are tracked. Use consistent campaign tags (UTM parameters) on every link, with a naming convention people stick to. A large share of attribution trouble is down to messy tagging.
Compare with real results. Do ad platforms add up to more conversions than you recorded in your own orders? They usually do, because each platform claims credit for the sales it touched. Use your own system as the source of truth for totals.
Run a test when the stakes are high. To find out if a channel really drives sales, switch it off for a region or hold back part of an audience, and see what happens. It's the most reliable evidence available.
When to look beyond it
If your spending is large and spans offline channels, attribution alone won't be enough. A technique called marketing mix modelling uses totals over time and doesn't depend on tracking individuals. We compare the two in marketing mix modelling or attribution: which one do you need?.
For now, remember the main thing. An attribution report is a model, not a fact. Use it to ask better questions, and be wary of anyone who treats a single number as the truth about what drove a sale.