Marketing attribution models: which to choose?

Attribution models have a reputation for being the kind of topic that sends B2B SaaS marketers running. And when the question of which model to use actually comes up, it tends to get punted to the data team, deferred until after the next product launch, or quietly dropped.

A model is already being used though. GA4 has a default. Every ad platform has a default. The issue is that no single model fits every scenario. Instead of pretending one model is right for everyone, here's every major one: what it does well, and where it falls short. For the underlying concepts, What is marketing attribution? A beginner's guide is worth reading first.

The basics

A few terms worth defining before getting into the models.

A touchpoint is any interaction a potential customer has with a brand before purchasing. A LinkedIn ad, a blog post, a cold email, a webinar, a sales call. Most B2B buyers go through anywhere from ten to hundreds of touchpoints before converting, spread across weeks or months.

A conversion is the action being measured. In B2B SaaS this is usually a demo booking, a trial signup, or a closed deal, depending on what stage of the funnel is being attributed.

A marketing attribution model is built from sets of rules or algorithms used to decide how much credit each touchpoint receives for a conversion. Different models distribute that credit differently. Some give it all to one touchpoint. Some spread it across many. Some use data to figure out what actually mattered. What is an example of marketing attribution? A B2B journey shows all of them applied to the same customer journey side by side.

An attribution window is the time period during which a touchpoint is eligible to receive credit. A 30-day window and a 90-day window will produce very different results for the same customer journey. For B2B SaaS with longer sales cycles, windows shorter than 90 days often miss meaningful early-stage interactions entirely.

At a glance

Every model covered below sits somewhere on a spectrum from simple to accurate. The simpler ones are easier to set up and easier to explain in a board meeting. The more accurate ones require cleaner data and more infrastructure. Neither end of the spectrum is wrong. It depends on what the business needs and what the data can support.

Single-touch models (first-touch, last-touch) give all credit to one interaction. Fast to implement, easy to understand, and significantly limited for any journey with more than a handful of touchpoints.

Multi-touch models (linear, time decay, position-based, u-shaped, w-shaped) distribute credit across the journey in different ways. More representative of how B2B buying actually works, with tradeoffs depending on which distribution logic is applied.

Data-driven models use machine learning to assign credit based on what the conversion data actually shows.

First-touch, last-touch, linear, time decay, and the position-based models are all forms of rule-based attribution: a rule-based process that applies a fixed formula the same way regardless of how a specific journey played out. Data-driven models trade that fixed process for statistical inference instead.

First-touch attribution

The first touchpoint gets all the credit. The logic is that without the initial interaction (the LinkedIn ad, the Google search, the blog post), the customer never enters the funnel at all. It's genuinely useful for understanding which channels are generating initial brand awareness and what's bringing new prospects into the top of the funnel.

The limitation is that it ignores everything that happened after. The content that built trust, the retargeting ad that brought the prospect back, the demo that closed the deal, all of it gets nothing.

Last-touch attribution

The final touchpoint before conversion gets all the credit. The assumption is that the most recent interaction caused the sale. For short, direct-response campaigns with few touchpoints, this isn't a bad approximation.

For B2B SaaS it's a different story. A prospect who spent four months consuming content and attending webinars before converting looks identical to one who found the site via a single Google search. When a team relies on last-touch as its primary model, top-of-funnel activity, content, and brand-building can be systematically starved of budget because they never receive credit.

Linear attribution

Credit is distributed equally across every touchpoint in the customer journey. At minimum, it acknowledges that the full journey matters, which already puts it ahead of single-touch models.

The weakness is that equal distribution isn't the same as accurate distribution. An email that was never opened gets the same credit as a product demo that directly preceded a purchase. It's a reasonable starting point for teams moving away from last-touch, but not a long-term solution.

Time decay attribution

More credit goes to touchpoints closer to the conversion date, on the assumption that recent interactions did more of the causal work. For short sales cycles this is a fair reflection of reality.

For longer ones it isn't. In B2B SaaS, where deals can take three to six months to close, the blog post a prospect read at the very start of their research can be the most important touchpoint in the entire journey. Time decay treats it as nearly irrelevant.

Position-based attribution (U-shaped and W-shaped)

Position based models split the difference between the two extremes above. U-shaped weights credit toward two specific moments, the first touchpoint and the conversion event, with the remaining credit distributed across everything in between. A common split is 40% to first touch, 40% to the conversion touch, and 20% spread across the middle.

W-shaped extends that logic with a third weighted moment: lead creation. Credit is concentrated at three points, first touch, lead creation (filling a demo form or signing up for a trial), and the final touch before conversion, with the remainder distributed across everything else. The three weighted moments map naturally onto the stages that matter most in a considered B2B purchase: initial awareness, expressed intent, and closed deal. Not the most accurate model available, but the most practical for most B2B SaaS companies without the data infrastructure to support algorithmic attribution.

Data-driven attribution (algorithmic)

Data-driven attribution doesn't use a fixed set of rules. It uses machine learning to analyze actual conversion data and assign credit based on which specific marketing touchpoints statistically correlate with conversion across a company's own customer journeys.

Touchpoints that consistently appear in journeys that convert get more credit. Touchpoints that appear equally in converting and non-converting journeys get less. It's the most accurate touchpoint-level model available because it reflects what actually happened rather than what a predetermined rule assumes happened.

It does require enough deal volume to be statistically reliable. A company closing 15 to 20 deals a month doesn't have enough signal for meaningful output. And it needs clean, centralized data. Fragmented data going in produces fragmented results coming out, regardless of how sophisticated the model is.

Marketing mix modeling: the privacy-safe alternative

Every model above works at the level of the individual touchpoint, which means it needs to track individual users to function. That's a problem once cookies disappear, privacy regulations tighten, or channels like TV, podcasts, and out-of-home advertising don't produce any trackable touchpoints at all.

Marketing mix modeling (MMM) takes a different approach. Instead of tracking individual people, it's a statistical analysis technique applied to aggregated historical data, spend, conversions, and external factors like seasonality, layered together and analyzed through statistical regression, a causal inference method, to measure the effect each channel had on overall results.

Because MMM never touches individual-level data, it's privacy-safe by design, and it works for offline and untrackable channels that touchpoint-based models simply can't see. The tradeoff is granularity. MMM gives a holistic view of what's driving results in aggregate, useful for scenario planning and forecasting methodology, but it can't tell you which single email closed which single deal the way a touchpoint model can.

For B2B SaaS companies, MMM is rarely a replacement for touchpoint-level attribution. It's a complementary, data-driven analytical approach best used alongside it, particularly for measuring the effect of brand and offline spend that never shows up in a CRM.

How to choose

Three things shape the decision: the length of the sales cycle, the number of marketing channels in the mix, and whether the underlying data is clean and centralized enough to work with.

For most B2B SaaS companies with sales cycles longer than 60 days, w-shaped is the most practical starting point. Representative enough to produce useful insight without requiring the data infrastructure that algorithmic attribution demands.

For teams with the deal volume and data quality to support it, data-driven attribution is worth the investment. And for teams with significant offline or brand spend, MMM is worth running alongside whichever touchpoint model gets chosen. Cross-channel attribution: connecting the full journey covers what has to be true of the data before any of these models can produce something trustworthy, and How to measure marketing attribution? A step-by-step guide walks through building that foundation step by step.

The gap in accuracy between a well-implemented model and a rule-based default is significant, and that gap shows up directly in budget decisions.

If you're looking to choose the right attribution model without building the infrastructure from scratch, that's what Cascayd does. Try Cascayd for free.