What is an attribution model?
An attribution model is a rule for distributing conversion credit across the touchpoints in a customer journey. A customer sees a TikTok ad, clicks a Meta retargeting ad, opens an email, and searches the brand on Google before buying. Four touchpoints. One sale. An attribution model decides how the revenue gets split.
Different models produce very different answers from the same data. There is no mathematically correct choice; every model encodes assumptions about how marketing influences buying behaviour. Understanding those assumptions is half the skill. For the broader context, start with what is marketing attribution.
Single-touch models
Single-touch models assign 100 percent of the credit to one moment in the journey. They are simple, easy to communicate, and almost always misleading when used alone.
Last Touch (Last Click)
100 percent of credit to the final interaction before conversion. The default in Google Analytics, Shopify, and most ad platforms. Systematically over-credits branded search, retargeting, and anything that sits close to the checkout.
Best for: understanding what closes. Weakness: blind to everything that created intent.
First Touch (First Click)
100 percent of credit to the interaction that started the journey. Useful for revealing which channels bring new customers into the funnel, but ignores everything that happened afterward.
Best for: discovery channel analysis, brand building. Weakness: ignores nurture and conversion touches.
Used in isolation, both models distort budget. Used together as opposite-end sanity checks, they reveal which channels open journeys vs close them, often surprising teams that have only ever looked at last click.
Multi-touch models
Multi-touch models distribute credit across every touchpoint in the journey, using different rules for how credit is weighted. They are the operational backbone of modern attribution.
Linear
Credit distributed equally across every touchpoint. Five touches in a journey? Each gets 20 percent. Treats a minor impression the same as a decisive demo, which is both its strength (no bias) and its weakness (no nuance).
Best for: long B2B cycles, early attribution baselines, fair aggregation.
Time Decay
Recent touchpoints weighted more than earlier ones. Credit halves every fixed period (typically 7 days). Reflects the reality that recent interactions tend to carry more decision weight.
Best for: short to medium sales cycles, ecommerce, shopping campaigns.
Position Based (U-Shaped / 40-20-40)
40 percent to the first touch, 40 percent to the last touch, and the remaining 20 percent split across middle touches. Honours both discovery and close, which matches how considered purchases actually work.
Best for: DTC, subscription products, anything with a clear discovery-to-purchase arc.
Full Path (W-Shaped or Z-Shaped)
Extends position-based with a third "key middle" touchpoint (often a demo, trial signup, or lead capture) and optionally post-conversion touches. The most comprehensive multi-touch view for multi-stage journeys.
Best for: B2B SaaS, long sales cycles with measurable milestones.
For a short-read introduction, see what is multi-touch attribution.
Data-driven attribution
Data-driven attribution (DDA) uses statistical or machine-learning models to distribute credit based on your own conversion patterns, rather than a predefined rule. Google Analytics 4, Meta, and several commercial platforms offer a DDA option.
In principle it is the most sophisticated approach. In practice it comes with trade-offs: it requires large conversion volumes to produce stable outputs, its credit distributions are difficult to audit or explain to a CFO, and each platform's DDA sees only its own data. Different platforms running different DDAs on different data will produce different answers, which is the problem independent attribution was meant to solve.
Marketing mix modelling
Marketing mix modelling (MMM) is not a multi-touch model in the same sense. MMM works at the aggregate level, using statistical analysis of total spend, total revenue, seasonality, and external factors over time. It does not track individual users. It measures the relationship between each channel's spend and revenue, producing incremental contribution estimates.
Where MTA answers "which channels moved this customer?", MMM answers "what would revenue have been without each channel?" The two are complementary. MTA gives daily granularity; MMM gives causal honesty. Most mature teams run both. See the Attriqs MMM module and the deeper piece on reported vs true incremental ROAS.
Side-by-side comparison
| Model | How it splits credit | Best for | Main blind spot |
|---|---|---|---|
| Last Touch | 100% to final interaction | Quick close analysis | Over-credits bottom funnel |
| First Touch | 100% to first interaction | Discovery channel analysis | Ignores nurture and close |
| Linear | Equal across all touches | Long B2B, fair baseline | Over-values minor touches |
| Time Decay | More weight to recent | Ecommerce, short cycles | Under-credits discovery |
| Position Based | 40/20/40 split | DTC, considered purchase | Under-values nurture |
| Full Path | Multi-stage W-shape | B2B SaaS, multi-step sales | Complex to communicate |
| Data-Driven | Statistical, from data | High volume, trusted data | Opaque, data-hungry |
| MMM | Aggregate regression | Incremental, forecasting | No user-level granularity |
The same journey, six models
A $500 sale influenced by: TikTok discovery (day 1), Meta retargeting (day 4), Email open (day 6), Branded Google Search (day 8). Here is how each model distributes the credit.
| Model | TikTok | Meta | Brand Search | |
|---|---|---|---|---|
| Last Touch | $0 | $0 | $0 | $500 |
| First Touch | $500 | $0 | $0 | $0 |
| Linear | $125 | $125 | $125 | $125 |
| Time Decay | $50 | $100 | $150 | $200 |
| Position Based | $200 | $50 | $50 | $200 |
| Incremental (est.) | $275 | $125 | $75 | $25 |
Same data. Six stories. The model you pick decides which channel gets more budget next week. That is why mature teams do not pick just one.
How to choose a model
Four factors usually drive the right starting point.
- 1. Sales cycle length. Short cycles (ecommerce, impulse retail) favour time decay or position based. Long cycles (B2B, SaaS, considered purchase) favour linear or full path.
- 2. Channel mix. Heavy upper-funnel spend (TikTok, podcasts, YouTube) needs first touch or position based to surface discovery value. Narrow paid-search-only mix can get by with last touch.
- 3. Data volume. Low volume favours simple rule-based models that do not require training data. High volume opens up data-driven attribution.
- 4. Offline share of revenue. Businesses with significant phone or in-person conversion must pair their model choice with call tracking; otherwise every model understates channels that drive calls.
Not sure where to start? The Attribution Model Decision Tree recommends one based on your inputs.
Why mature teams run multiple models
No single model is universally correct. The useful skill is not picking "the right one" but running several in parallel and reading the disagreements.
- Agreement across models is a strong signal. If a channel looks excellent under last touch, linear, and position based, confidence is high.
- Disagreement between last touch and first touch reveals whether a channel opens journeys or closes them. Often the same channel looks very different across those two views.
- Gap between attributed and incremental reveals non-causal spend. Branded search is the classic example: high attributed ROAS, much lower incremental ROAS.
Attriqs runs six models simultaneously on a single dataset so the comparison is one click, not six exports. See the multi-touch attribution feature.
Frequently asked questions
What are the main attribution models?
The main marketing attribution models are Last Touch, First Touch, Linear, Time Decay, Position Based (also called U-shaped), and Full Path. Beyond these, Data-Driven Attribution uses machine learning to distribute credit statistically, and Marketing Mix Modelling operates at the aggregate channel level rather than the user journey level.
What is the difference between last click and last touch?
They are the same model in practice. "Last click" is the more common name in Google Analytics and paid search contexts; "last touch" is the more general name used in attribution software. Both assign 100 percent of conversion credit to the final interaction before the sale.
Which attribution model is best?
There is no universally best model. The right choice depends on sales cycle length, channel mix, and what question you are trying to answer. The modern best practice is to run several models in parallel and compare them. If a channel looks strong across every model, confidence is high. If it only looks good under one, the ranking is fragile.
What is the difference between first touch and first click attribution?
First touch and first click are used interchangeably. Both assign 100 percent of credit to the interaction that started a customer journey. First touch is useful for understanding which channels bring new customers into your funnel but is blind to everything that happens afterward.
Is linear attribution too simple?
Linear is mathematically simple but analytically useful. It distributes credit equally across every touchpoint, which treats a minor impression the same as a decisive demo. It is a fair baseline for comparing channels that would otherwise be invisible under last click, and it is especially useful for long B2B sales cycles where every touch matters.
What is position-based attribution?
Position-based attribution (also called U-shaped or 40-20-40) assigns 40 percent of credit to the first touch, 40 percent to the last touch, and splits the remaining 20 percent across the middle touches. It honours both the channel that started a journey and the channel that closed it, which matches how most considered purchases actually behave.
How is data-driven attribution different from multi-touch?
Data-driven attribution (DDA) is a type of multi-touch attribution that uses machine learning rather than predefined rules to distribute credit. Instead of "40-20-40," it learns from your own conversion data which touchpoints matter most. It is sophisticated in principle but often opaque in practice, and usually requires very large data volumes to produce stable outputs.
When should I use MMM instead of multi-touch attribution?
Marketing Mix Modelling (MMM) measures the relationship between total spend and total revenue at the channel level, without tracking individual users. It is the right tool when user-level tracking is degraded (privacy changes, offline channels, or out-of-home media), for measuring incremental contribution, and for forward-looking budget recommendations. Most mature teams use both: MTA for daily granularity, MMM for causal honesty.