The Challenges of Multi-Touch Attribution
Why multi-touch attribution is hard to get right, from biased exposure and messy data to privacy gaps and internal buy-in, and what it is still good for.
Multi-touch attribution promises something every marketer wants: a clear account of which touchpoints led to each sale. In practice, getting an accurate answer from it is harder than most implementations admit.
The problems fall into four groups. Some are statistical, some are about data, some come from privacy changes, and some are about people. Knowing them helps you decide what MTA can tell you and where you need a different method. If you need a primer first, start with what multi-touch attribution is and how its models work.
Targeted ads don't reach a random audience
The deepest problem is that modern advertising is targeted. Ads are shown to people the platforms predict are likely to convert. That makes it hard to tell whether the ad caused the purchase or simply reached someone who was already on their way to buying.
Retargeting is the clearest example. It shows ads to people who have already visited your site or added something to a cart. Many of them would have bought anyway. Attribution sees the ad in the path and gives it credit, so retargeting and branded search often look far more effective than they are.
There are two common ways to separate media effects from existing intent:
- Randomized controlled tests. Holding out a random group from seeing the ads isolates the true effect. It is the most direct method, but tests cost money, take time and are hard to run across every campaign and tactic.
- Propensity adjustments. A model estimates how likely each person was to be exposed to the media, based on their behavior and characteristics, and includes that likelihood alongside ad exposure. This helps separate the effect of intent from the effect of the ad.
Other hidden factors distort credit too: the delay between seeing an ad and acting on it, whether an impression was actually viewable, and each person's baseline chance of converting with no ads at all. An attribution model that ignores them will over-credit the media closest to the sale.
The underlying point is that MTA measures which touchpoints were present, not which ones made the difference. That question belongs to incrementality measurement.
The data is fragmented and inconsistent
MTA is only as good as the data feeding it. The old rule applies: garbage in, garbage out.
Building a single view of each customer means merging ad platform data, web analytics, CRM records and sales activity. Each source has its own identifiers, definitions and gaps. Doing this well is time-consuming, and errors carry straight into the results.
Some practical requirements come up again and again:
- Consistent tracking. UTM parameters have to be applied the same way across every team and campaign, or traffic gets misattributed before any model runs.
- A data foundation beyond the CRM. Many teams find CRM and marketing automation reporting too limited and move attribution data into a warehouse with a business intelligence layer.
- An iterative build. Trying to design a complete data system up front is usually overkill. It works better to build what the current questions require and add tools and people as the questions get harder.
There is no single right model
Attribution isn't one problem. A company's products, customers and sales channels each shape the right approach. A high-consideration B2B purchase with a sales team looks nothing like an impulse buy on a phone.
Choosing between linear, time-decay, position-based and data-driven models is a real decision, because each one tells a different story about the same journeys. The guide to attribution models covers how they differ.
Models also tend to favor the lower funnel. Clicks near the purchase are easy to see, while the ads that put a brand into someone's consideration set are not. Brand effects are especially hard to quantify at the user level, so teams need people inside the business who understand why upper-funnel media matters, or budgets drift toward whatever closes.
Privacy changes and dark social hide much of the journey
MTA depends on recognizing the same person across sites, apps and devices. That has become much harder. Apple's App Tracking Transparency, browser tracking protections in Safari and Firefox, and privacy regulations have all cut the user-level signals attribution relies on. Chrome still supports third-party cookies, but the cross-site picture is far less complete than it was.
Even with perfect tracking, many influences were never visible. The term "dark social", first used by journalist Alexis Madrigal in 2012 and popularized in B2B marketing by Chris Walker of Refined Labs, covers the touchpoints that shape a purchase but that attribution software can't measure: a podcast mention, a post in an industry forum, an ad shared in a company Slack channel, a conversation over lunch. The buyer learned about you there, but attribution records the visit as direct or organic search.
Channels people don't click, such as connected TV, radio and out-of-home, have the same problem. For more on how these shifts changed measurement, see life after third-party cookies.
Attribution fails without buy-in
Many attribution projects stall for organizational reasons rather than technical ones.
When an attribution model is built in isolation, without marketing, sales, finance and IT involved from the start, its results are easy to dismiss. Cross-functional buy-in has to come early, and it depends on explaining how the model works in plain, transparent terms.
Expect sticker shock. A working model often changes reported numbers noticeably, and executives need to be prepared for that before the first report lands. Ongoing education helps: raw data views for analysts, short executive summaries, and regular reviews built around specific business questions.
Attribution also needs senior support and a long-term budget. It is an ongoing capability, not a one-time project.
What multi-touch attribution does well
None of this means MTA has no value. When the data is sound and the team trusts it, it helps in three ways:
- Ending credit disputes. Arguments between marketing and sales about who drove a deal give way to shared data about the whole journey, and teams can decide where to put resources based on it.
- Fewer irrelevant messages. Knowing what a customer has already seen and done lets teams stop sending generic messages and tailor communication to the situation.
- A clearer view of the journey. Seeing the sequence of interactions helps improve campaigns and the customer experience within the channels you can track.
Using MTA where it fits
The most reliable way to work with multi-touch attribution today is to use it for what it is good at, tactical comparisons within trackable channels, and to answer budget questions with methods that don't depend on following individuals.
Marketing mix modeling and incrementality measurement work from aggregate data, include channels people don't click, and estimate what actually caused sales. Many teams now combine the two approaches; MTA vs. MMM explains how they compare.
Attribution can tell you who touched the customer. Deciding where the next dollar goes depends on knowing what made a difference.
Frequently asked questions
What are the biggest challenges of multi-touch attribution?
The main ones are biased exposure (targeted ads reach people already likely to convert), fragmented and inconsistent data, choosing a model that fits the business, gaps in tracking caused by privacy changes and untrackable channels, and getting marketing, sales and finance to trust the results.
Why does multi-touch attribution over-credit retargeting?
Retargeting shows ads to people who have already visited your site or shown interest. Many of them would have bought anyway, but because the ad was in their path, attribution gives it credit. Separating the ad's real effect from existing intent requires a test or a statistical adjustment.
What is dark social in attribution?
Dark social describes the influences on a purchase that attribution software can't track, such as a podcast mention, a recommendation in a private chat or Slack channel, or a conversation with a colleague. When the buyer later arrives, the visit is usually logged as direct or organic search.
Is multi-touch attribution still worth using?
Yes, for the right questions. It is useful for comparing creative, keywords or audiences within channels you can track well. For deciding how to split budget across channels, especially ones people don't click, aggregate methods like marketing mix modeling and incrementality measurement are more reliable.