[ Tracking and attribution ]

Multi-touch attribution: 7 models for course creators

Quentin BarrochéApril 25, 202611 min read

← All articles

Picture this. A prospect sees your Facebook ad on Monday, comes back on Tuesday by typing your name into Google (organic result), downloads your lead magnet on Thursday through an organic search, opens 3 emails from your funnel over a week, then buys your 1,497 euro course the following Monday.

Who made the sale? Facebook, Google, your SEO, your email marketing, or your sales page?

That is the 1,497 euro question multi-touch attribution answers. And the answer changes radically depending on the attribution model used. This article explains the 7 main attribution models on the market, with concrete examples, then the single rule DataHide follows.

The problem: last-click attribution lies to you

By default, many analytics tools attribute the entire conversion to the last click before the sale, whatever the channel. This model, called "last click" or "last touch", is simple, but it massively distorts reality.

Let's go back to the previous example. With last click:

  • Meta: 0 euros of revenue attributed
  • Google: 0 euros
  • SEO: 0 euros
  • Email and sales page: 1,497 euros (the last click, from an email, takes it all)

You conclude that your email marketing does all the work and that Meta is useless. A major strategic mistake: without the Meta ad, this prospect would never have heard of you, and would never have joined your list. You cut the source that brings the traffic and credit the one that only closes the sale.

The platforms make the opposite mistake: each one claims the sale when a click on its ad comes before it, but only within its own window. Across 7,164 sales tracked at 18 businesses, 58% came more than 7 days after the click. Meta stops counting at 7 days. This is exactly what leads many course creators to say "Facebook doesn't work anymore" when Facebook is precisely what feeds their funnel. They measure badly, so they decide badly.

What is multi-touch attribution?

Multi-touch attribution recognizes that a B2C or B2B sale (and especially a course at 500 or 5,000 euros) involves several touchpoints with your brand before the purchase decision.

The role of multi-touch attribution is to split the credit for each sale among all the touchpoints that contributed, according to a precise mathematical formula. Each model offers a different formula, with its own logic.

For multi-touch attribution to work, you need to identify each touchpoint and follow the same user's journey across them. That is exactly what modern server-side tracking makes possible, by getting around the limits imposed by iOS 14 and ad blockers.

The 7 multi-touch attribution models

Here are the 7 main models on the market. For each one, I give the mathematical formula, the ideal use case and the biases to know about. DataHide does not make this a setting: the single rule it applies is explained further down.

Model 1: First Touch

Formula. 100% of the credit to the first identified touchpoint.

In our example: Meta receives 1,497 euros, every other channel receives 0.

Use case. This model overvalues cold-traffic acquisition channels (display ads, video, top-of-funnel content marketing). Useful when your goal is to evaluate a channel's ability to build awareness and reach a new audience.

Bias. Massively underestimates conversion channels (brand search, follow-up emails, retargeting). If you run your budget on First Touch, you will cut your closing channels and wonder why your leads no longer convert.

Model 2: Last Touch (last click)

Formula. 100% of the credit to the last identified touchpoint before the conversion.

In our example: the sales page, reached from an email, receives 1,497 euros, every other channel receives 0.

Use case. It is the default model in many web analytics tools. It is simple to understand and to measure. Relevant only for impulse purchases with a very short decision cycle (impulse e-commerce, microtransactions).

Bias. Disastrous for course creators and B2B SaaS. Overvalues conversion channels and completely underestimates the top of the funnel. It is the model that can make you believe a retargeting campaign is very profitable when it is just capturing prospects already warmed up by another source.

Model 3: Linear

Formula. Credit split equally among all touchpoints in the journey.

In our example with 5 touchpoints (Meta, Google, SEO, email, sales page): each one receives 1,497 / 5 = 299.40 euros.

Use case. The recommended baseline model to start a multi-touch analysis when you do not yet have a view on the relative weight of each channel. Well suited to long course funnels where each touchpoint plays a clear role (discovery, consideration, decision).

Bias. Tends to dilute the performance of channels that rarely appear. If the average journey has 10 touchpoints, each channel gets 10% and precision is lost.

Model 4: Time Decay

Formula. The closer in time a touchpoint is to the conversion, the more credit it receives. Exponential decay (typically a 7-day half-life).

In our example, the Meta click on Monday (7 days before the sale) receives much less than the sales page the following Monday (0 days).

Use case. Often presented as a good compromise for B2C funnels. It reflects well the fact that a prospect who consumed your content yesterday is more engaged than one who saw your ad 30 days ago. It weights each channel's role fairly without penalizing the top of the funnel.

Bias. Slight overweighting of end-of-journey channels. Less suited to very long sales cycles (courses at 5,000 euros and above, B2B with a buying committee).

Model 5: Position-Based (U-Shape)

Formula. 40% to the first touchpoint, 40% to the last, 20% split among the touchpoints in between.

In our example: Meta receives 598.80 euros, the sales page receives 598.80 euros, and the 3 middle touchpoints (Google, SEO, email) share 299.40 euros (99.80 euros each).

Use case. Relevant for course funnels at 1,000 to 5,000 euros where you want to value both the acquisition source (which brings the prospect) and the conversion source (which convinces them to pay).

Bias. Underweights mid-journey touchpoints, which can be structurally important (nurturing email sequences, discovery webinars).

Model 6: W-Shape

Formula. 30% to the first touchpoint, 30% to the lead magnet sign-up (often identified as a major touchpoint), 30% to the last touchpoint, 10% split among the others.

Use case. Suited to funnels with a lead magnet or a structured webinar where the sign-up is a critical intermediate conversion event. Widely used in B2B SaaS.

Bias. Requires manually defining the "major" touchpoints. More complex to configure.

Model 7: Data-Driven (algorithmic)

Formula. A machine learning algorithm analyzes all converted and non-converted journeys, then gives each touchpoint the credit that matches its real impact on the probability of conversion (Shapley value model).

Use case. The most accurate model in theory. It needs a large volume of conversions to train the model.

Bias. Black box: you do not know exactly why the model gives a given weight to a given channel. Hard to explain to a client or a partner. Sensitive to data volume: unreliable on small volumes.

What each model highlights

No model is neutral: each one answers a different question.

The question you are askingModel that answers itLimit
Which channel makes my offer known?First TouchIgnores everything that closes
Which channel triggers the purchase?Last TouchIgnores everything that brought the prospect
Which channels take part in the journey?LinearDilutes credit when the journey is long
Who followed up with the prospect just before the purchase?Time DecayUnderweights discovery
Who brings the prospect and who closes?Position-Based (U-Shape)Underweights the middle of the journey
What role does the webinar sign-up play?W-ShapeRequires defining the major steps
What is each channel's real impact?Data-DrivenBlack box, needs a large volume

The rule DataHide follows

DataHide does not make you pick a model. It applies a single rule, the same for everyone:

  • each sale goes to the last paid ad clicked in the 30 days before it;
  • the sale is dated on the closing day, not on the day the lead signed up.

Why this choice. Your steering question is simple: which ad brought this sale, and how much did it cost you. By only counting clicks on paid ads, email, organic search or the sales page cannot steal a campaign's credit. With a 30-day window, you recover the sales Meta stops counting after 7 days: the median time between click and purchase is 10 days. And by dating the sale at closing, your monthly number matches the sales actually closed that month.

In our example, DataHide credits the 1,497 euro sale to Monday's Meta ad, and files it on the purchase day. A sale with no ad click in the previous 30 days is not tied to any campaign.

How to implement multi-touch attribution in practice

Three required steps.

Step 1: track each touchpoint individually. Install a pixel or tracking script that captures every interaction (ad view, click, form submission, email open, purchase). Without exhaustive tracking, multi-touch attribution does not know how many touchpoints exist.

Step 2: identify the same user across channels. The core challenge. On the web, matching relies on cookies + server identifiers (fbp, fbc) + hashed email/phone (server-side). Modern server-side tracking is essential for this matching to work despite iOS 14, ITP and ad blockers.

Step 3: apply an attribution rule. Either manually with an Excel formula (doable but fragile), or with a dedicated tool that does the calculation for you. DataHide applies a single rule: the last paid ad clicked in the 30 days before the sale, with the sale dated at closing.

Example: a coach who thinks "Facebook is dead"

Let's take an illustrative example: a business coach thinks Meta no longer works for him. In his Business Manager, his Meta campaigns show very few sales for a lot of spend, while his Google campaigns on his brand name look very profitable.

The natural conclusion he was about to act on: cut Meta entirely and put everything into Google.

Once each Stripe sale is tied to its real click, the picture changes. A good share of buyers clicked a Meta ad, then took more than 7 days to buy: Meta no longer counted them. Without Meta, some of these sales would simply not exist. Meta's ROAS is lower than Google's, but not zero, and its role is structural.

The right decision: keep Meta for its acquisition role and work on the creatives, rather than cutting the channel that feeds the whole funnel.

Limits of multi-touch attribution

Multi-touch attribution is not magic. Three important limits to keep in mind.

Tracking is imperfect. Even with server-side tracking, you do not capture 100% of touchpoints. Reels ad views that do not generate a click, Instagram DM conversations, word-of-mouth recommendations, branding on untracked podcasts: all these touchpoints are invisible. Multi-touch attribution assigns credit based on what it sees, not on what really happened.

Models are approximations. No model perfectly reflects the psychology of a purchase decision. A channel's role can be structurally crucial (reassuring, building trust) without generating a measurable click. Models make up for this blur by spreading the credit, but the imprecision remains.

Attribution does not replace marketing common sense. If your attribution tells you Facebook performs at 0.9x ROAS and Google at 4x, that does not mean Facebook is profitable in economic terms. At 0.9x ROAS, you lose money, even with accurate attribution. Attribution clarifies each channel's role; it does not replace analyzing gross margin per sale.

Frequently asked questions

What is the difference between multi-touch attribution and multi-channel attribution?

Multi-channel and multi-touch often mean the same thing in everyday language. Strictly speaking, "multi-channel" means "more than one different channel" (Facebook + Google), and "multi-touch" means "more than one touchpoint" (which can be several on the same channel, for example 3 successive Facebook ad views). Modern attribution combines both dimensions.

How many days of history for multi-touch attribution?

The most common window is 30 days, and it is the one DataHide uses. In our data, 38% of buyers purchase within 48 hours and 85% within 60 days. For longer cycles (B2B, very expensive courses), some tools push it to 60 or 90 days, knowing the dilution will be stronger.

Can you change attribution model retroactively?

Some tools, such as Hyros or Google Analytics 4, let you reread the same sales with another model. DataHide made a different choice: a single, stable rule, so a month's numbers do not change depending on who reads them and with which setting. Simpler tools (Stape, server GTM without proprietary logic) do not attribute sales at all.

Does multi-touch attribution work for B2B with a long cycle?

Position-Based or W-Shape models with a long window are the most common in B2B. For sales where the decision involves several people (buying committee), you should ideally pair tracking with a CRM that follows each contact person (HubSpot, Pipedrive). In DataHide, the sale stays tied to the last paid ad clicked in the 30 days before it, and the closer portal shows who closed it.

What attribution should I keep on Meta Ads and Google Ads alongside my tool?

Keep the standard last-click attribution on Meta and Google (it is their optimization signal) and use your third-party tool (DataHide, Hyros) for consolidated attribution on your real sales. The two measurements coexist: Meta sees its world through last click to optimize its bids, and you steer your overall budget with this consolidated attribution.

Recap: why last click is no longer enough

Last click has been the standard since Google Analytics has existed. It was designed for short purchase cycles and few channels. It no longer fits the reality of 2026, where a course creator combines several acquisition channels and a buyer goes through several touchpoints before paying.

If you run your ad budget on last click, you cut structurally useful channels and over-invest in end-of-journey channels. You optimize locally and destroy value globally.

Understanding the models keeps you from misreading your numbers. To steer your budget, what you need above all is a stable rule tied to real sales: at DataHide, the last paid ad clicked in the 30 days before the sale, with the sale dated at closing. Your budget decisions then rest on what actually sells.

You can try DataHide free for 14 days, with no credit card, after a 10-minute setup on a video call with the founder.

You don’t need more leads. You need more sales.

14-day free trial, no credit card required. Set up in 10 minutes, on a video call with the founder.