tl;dr
Static models (Last Touch, First Touch, Linear, U-Shape) apply fixed rules — they don't adapt to how a journey actually played out
Unique is the only static model worth using regularly, for day-to-day creator optimization
Data-Driven adjusts value dynamically based on session quality, channel type, time lags, discount codes, and PPS answers
Marketing Mix Model builds on top of Data-Driven to reallocate branded/direct conversions back to the channel that actually generated the demand
For budget allocation, use Data-Driven + MMM — it's the most complete picture of what's driving conversions
The example journey
All models below are illustrated using this customer journey:
TikTok → Facebook Paid → Influencer → Branded Paid Search → Email Automation → Purchase
Static attribution models
Static models assign value to touchpoints based on a fixed rule — regardless of what actually happened in the journey.
Last Touch gives 100% of the value to the final touchpoint before purchase. In this journey, Email Automation gets everything. Most marketers will recognize this from Google Analytics.
First Touch gives 100% to the first touchpoint. Here, TikTok gets everything — the idea being that the channel that brought the customer in deserves the most credit.
Linear distributes value equally across all touchpoints. With five touchpoints, each gets 20%.
U-Shape gives more weight to the first and last touchpoints, with the remaining value spread evenly across the middle. In this journey: ~30% TikTok, ~13% each for Facebook, Influencer, and Branded Paid Search, ~30% Email Automation.
Unique gives 100% of the value to every channel that had a touchpoint — TikTok, Facebook, Influencer, Branded Paid Search, and Email Automation all get 100%. This mirrors how ad platforms themselves report conversions based on click attribution.
The problem with static models
Static rules don't account for how a journey actually unfolded. Say the first touch was 60 days ago, the customer did nothing for 50 days, and then re-engaged 10 days before converting. Is that 60-day-old touchpoint really that important? Probably not. But a static First Touch model doesn't care — it treats every first touch the same.
That's why static models are generally not recommended for budget allocation decisions. The exception is Unique — it's useful for day-to-day creative optimization, since it shows you which creatives had any touchpoint in a conversion journey at all.
Data-Driven model
Instead of a fixed rule, the Data-Driven model starts from a baseline (U-shape) and then adjusts value dynamically based on what actually happened in each journey.
Adjustments the model makes
Time spent on site — a session where someone browsed for 10 minutes is worth more than a bounce. The model increases the weight of high-engagement sessions accordingly.
Channel type — not all channels create demand equally:
Branded Paid Search is boosted down. It doesn't create intent — it just captures intent that was already there.
Email and owned channels (WhatsApp, etc.) are evaluated in context. If a customer clicks an email two minutes after their last session, the email gets low value — they were already on their way back. If the email brings someone back after two months of inactivity, it gets high value.
Discount codes — if a discount code is linked to a channel in Klar (via the Influencer CRM or Influencer Sheet), two things can happen:
If that channel already has a touchpoint in the journey → it gets boosted
If there's no touchpoint for that channel → a new touchpoint is injected (at last touch for influencers)
Post Purchase Survey answers — the same logic applies. If the customer says "TikTok" and a TikTok touchpoint exists → boost it. If they say "Pinterest" and there's no Pinterest touchpoint → inject one.
The result is a model that reflects the actual quality and context of each touchpoint, not just its position in a sequence.
Marketing Mix Model
Even the best click-based model has a blind spot: journeys with very few tracked touchpoints. A new customer with a single Branded Paid Search touchpoint before purchase isn't a realistic user journey — they knew your brand from somewhere, they just didn't click any trackable ad.
The Marketing Mix Model (MMM) is built on top of the Data-Driven model to handle exactly this. It uses machine learning to understand what actually drove those conversions, by looking at:
Post Purchase Survey answers — especially for short journeys, PPS data gives a direct signal of the customer's perceived source
Correlation between spend, impressions, and engagement and the resulting branded/direct traffic — if you spend €10k more on Meta and branded paid search sessions increase by 5k, those branded conversions should be reallocated to Meta, not credited to Branded Search or Direct
The outcome is that branded and direct traffic gets redistributed to the channels that actually generated the demand. It also surfaces Word of Mouth as a standalone channel — for some brands, this accounts for 50–60% of post-purchase survey responses. Trying to allocate all conversions to paid channels when a significant portion is word-of-mouth leads to completely wrong budget decisions.
What changes with MMM
Compared to a last-click starting point, a typical shift looks something like this:
Meta gets significantly more conversions allocated (often 100–200% uplift) as branded/direct traffic is traced back to Meta campaigns
TikTok, which is heavily top-of-funnel and harder to track via clicks, often sees the biggest relative uplift
Influencer sees moderate uplift if they're already using discount codes (already well-tracked), less so otherwise
Branded Paid Search and Direct are reduced — they were overcounted in click-based models
Word of Mouth appears as a new channel reflecting organic brand strength
When to use which model
Unique — use for day-to-day creative and influencer optimization. Tells you which assets had any impact on a conversion, without worrying about value distribution.
Data-Driven — use as your default attribution view. It's the most accurate picture of what touchpoints contributed and how much, based on actual session data and zero-party inputs.
Data-Driven + MMM — use for budget allocation. This is the model that factors in what's tracked, what's observable through correlation, and what customers themselves report. It gives you the most complete view of true channel contribution — including channels that never generate a trackable click, like TV or word of mouth.
Why dynamic models matter — a quick analogy
Think about how a goal is scored in football. If you only look at the player who put the ball in the net (last touch), you miss everything that made that goal possible. If you credit only the goalkeeper who started the counter-attack (first touch), that's equally incomplete. Every player in the move contributed — and the value of each contribution depends on what they actually did, not just where they were in the sequence. Some players never touched the ball at all but created the space that made the goal possible. That's word of mouth and TV in your attribution — real impact, no trackable touch.