Field notes
Multi-Touch Attribution: What It Buys You, and What It Costs
A multi-touch attribution model tries to credit every step of the buyer's journey, not just the first or last one. Useful when your data is clean. A quiet disaster when it is not.
Multi-touch attribution spreads revenue credit across all the marketing touches a buyer had, using a weighting like linear, U-shaped, or W-shaped. It answers a real question for longer sales cycles: which activities are pulling their weight. But it only tells the truth if the touches are actually captured and the records are clean, which most CRMs are not. On messy data, a fancier model just dresses up a guess. For a short, simple funnel, first-touch or last-touch is often enough.
On this page
Here is a question that decides real budget. A deal closed last month for 40,000. Before that buyer ever spoke to sales, they read a blog post, came back through a paid ad, opened three of your emails, and finally booked a demo from a webinar. Which of those got the credit for the sale?
If your reporting hands the whole 40,000 to the webinar, you will pour money into webinars and starve the blog post that started the whole thing. That is the problem multi-touch attribution sets out to fix. It is a way of splitting the credit for a sale across all the marketing the buyer actually touched, rather than pinning it on one moment.
Whether it is worth doing for your business is a real decision, not a technical detail. We wrote a broader guide to marketing attribution that maps the whole territory; this piece goes deep on the multi-touch approach and, more importantly, on when you should not bother.
What multi-touch attribution is trying to do
Most companies start with a single-touch model, and for good reason: it is simple. First-touch attribution gives all the credit to whatever brought a buyer in the door. Last-touch gives it to whatever was happening right before they bought. We compared those two head to head in first-touch versus last-touch attribution, and the short of it is that each one tells you a true thing and hides another one.
Multi-touch says the truth is in the middle. A buyer with a six-month sales cycle had maybe a dozen meaningful interactions with your company. Some pulled them in, some kept them warm, some pushed them over the line at the end. A multi-touch model tries to give each of those a slice of the revenue, so that when you look at a channel you see the share of money it genuinely helped produce, not just the deals it happened to touch last.
The payoff, when it works, is that you can defend a budget. You can look at the middle-of-the-funnel content that never gets credited in a last-touch world and show that deals which touched it closed at a higher rate. That is the difference between guessing where to spend and knowing.
The common models, in plain terms
The models mostly differ in one thing: how they slice the credit. Three come up again and again.
Linear splits the credit evenly. If a buyer had ten touches before a 40,000 deal, each touch gets 4,000. It is the most honest starting point because it plays no favorites, and it is the easiest to explain to a board. The weakness is that it treats the ad someone half-noticed the same as the demo that closed them, which almost nobody believes is fair.
U-shaped (you will also hear it called position-based) says the first and the last touches matter most, so it hands each of those a big share, commonly 40 percent apiece, and splits the remaining 20 percent across everything in between. The logic is that getting someone interested and getting them to buy are the two hard jobs. This fits a lot of businesses well.
W-shaped adds a third heavy point: the moment a lead turns into a real sales opportunity. So first touch, opportunity creation, and closing each get a large slice, and the rest is spread thin. It suits longer, sales-led deals where that middle conversion is a genuine milestone, not a formality.
There is no model that is correct in some absolute sense. The weighting is a claim about how your buyers actually decide, and the right one is whichever matches how deals really move at your company. If your first conversation and your closing conversation do most of the work, U-shaped will feel right. If there is a clear moment where a curious lead becomes a live deal, W-shaped will.
The prerequisite nobody wants to hear
All of this rests on one assumption, and it is usually wrong. A multi-touch model can only credit the touches it can see. If a buyer's journey had a dozen steps and your systems captured five of them, the model does not tell you that. It quietly divides the credit among the five it has and reports the number with the same confidence it would report a perfect record.
This is where most attribution projects fall apart, and it has nothing to do with the model. Think about what has to be true for the touches to be captured. Every form fill has to land on the right contact record. Paid clicks have to carry tracking that survives all the way to the deal. The same person showing up as three separate records has to be merged into one, or their journey gets chopped into three partial stories. Offline touches, the phone call, the conference conversation, the reply a rep forgot to log, have to make it into the system at all.
Most CRMs are nowhere near this. Contacts are duplicated, campaign tracking is missing or inconsistent, and half the sales activity lives in inboxes rather than in the record. Run a multi-touch model on top of that and you have not measured anything. You have taken a set of guesses and given them decimal places. The report looks precise, so people trust it, and they steer real spending by it. That is worse than having no attribution at all, because a wrong number that looks rigorous is harder to argue with than an honest shrug.
So the first question is never which model to pick. It is whether the underlying data is clean enough to carry any model. Getting a customer record into that state, deduplicated, with touches actually captured and tied to revenue, is most of the work and most of the value. When we help a company get there, that groundwork is the project; the model on top is the easy part. If your team is drowning in the manual side of keeping records straight, a fitted custom AI build can take on the cleanup and matching that no one has time for.
When it is worth the complexity
Multi-touch attribution earns its keep when three things are true. Your sales cycle is long enough that buyers really do have many touches, so a single-touch view genuinely misleads you. You spend enough across enough channels that the answer changes where real money goes. And your data is clean enough that the model is reading reality rather than laundering guesses.
If your funnel is short and a buyer sees one ad and buys a week later, multi-touch is expensive overkill. Last-touch will tell you almost the same thing for a fraction of the effort, and you can put the saved time into the campaigns themselves. Plenty of good businesses never need more than a single-touch model, and there is no prize for building something more complicated than your decisions require.
The trap we see most is a company reaching for a W-shaped model to look sophisticated while its contact data is a mess. The order is backwards. Fix the record first, prove that a touch reliably makes it into the system and ties to a closed deal, then add weighting. In practice, once the data is clean, most companies find a U-shaped model answers the questions they actually had, and they never need to go further.
The whole point of any of this is to connect marketing to money you can bank, which is a harder job than it sounds and the one most reporting skips. We wrote separately about campaigns that connect to revenue, because a model is only useful if it ends in a decision about where next month's budget goes. If you cannot yet trace a single closed-won deal back through the touches that produced it, that is the place to start, not the model.
Common questions
What is multi-touch attribution in plain terms?
It is a way of splitting the credit for a sale across all the marketing a buyer interacted with, instead of giving it all to the first or last thing they did. If someone read a blog post, clicked an ad, and joined a webinar before buying, a multi-touch model gives each of those a share of the revenue so you can see which activities really helped.
What is the difference between linear, U-shaped, and W-shaped models?
They differ in how they slice the credit. Linear splits it evenly across every touch. U-shaped gives the biggest shares to the first and last touch, on the logic that getting attention and closing are the hard parts. W-shaped adds a third heavy point at the moment a lead becomes a real sales opportunity, which suits longer, sales-led deals.
How do I set up multi-touch attribution?
The model is the last step, not the first. Before anything, make sure the touches are actually captured and the data is clean: contact records deduplicated, campaign tracking that survives to the deal, and sales activity logged rather than sitting in inboxes. Once a touch reliably makes it into the system and ties to a closed deal, pick a weighting that matches how your buyers really decide. Most companies land on U-shaped.
When is a simpler attribution model good enough?
When your sales cycle is short and buyers have only a touch or two before they buy, a single-touch model like last-touch tells you almost the same thing for far less effort. Multi-touch is worth the complexity only when cycles are long, spending is spread across many channels, and the underlying data is clean enough that the model reflects reality instead of dressing up a guess.