Disclosure: This page contains commercial links. BizBot may earn a commission from qualifying purchases. Paid placements do not buy a better ranking. Read our affiliate disclosure.
Attribution models help marketers understand which channels drive conversions. There are two main types:
Single-Touch Models
- First-Touch: Gives all credit to the first interaction
- Last-Touch: Gives all credit to the final interaction before conversion
Multi-Touch Models
- Linear: Distributes equal credit across all touchpoints
- Time Decay: Gives more credit to recent touchpoints
- Position-Based: Weights the first and last touch most heavily
- Data-Driven: Uses machine learning to estimate each touchpoint’s contribution
Important change since earlier versions of this guide. Four of these models are no longer available in Google’s own tools. In 2023 Google removed first-click, linear, time decay and position-based attribution from both Google Ads and Google Analytics 4. What remains in GA4 is data-driven attribution and last-click variants. The rule-based models are still valid ways of thinking about credit, and other platforms still offer them, but if your plan was to configure a position-based model in GA4, that option is gone. This is covered in detail in the setup section below.
| Model | Best For | Available in Google’s tools? |
|---|---|---|
| First-Touch | Identifying initial awareness channels | No, removed 2023 |
| Last-Touch | Analyzing final conversion drivers | Yes |
| Linear | Evaluating overall marketing performance | No, removed 2023 |
| Time Decay | Longer sales cycles, nurturing leads | No, removed 2023 |
| Position-Based | Complex sales cycles, optimizing journeys | No, removed 2023 |
| Data-Driven | Businesses with enough conversion volume | Yes, the GA4 default |
Setting up attribution involves selecting a model, integrating data sources, configuring it, and analyzing metrics like conversion value, conversion count, and return on ad spend.
The Modern Customer Journey
How Customers Actually Buy
Customers research extensively before purchasing, across many channels, and rarely in a straight line. Journeys are fragmented, span devices, and often include touchpoints you cannot see at all.
That last point is worth sitting with, because it undermines every model in this article to some degree. A conversation with a colleague, a podcast mention, a search on a work laptop before buying on a phone: none of these appear in your data. Attribution models allocate credit among the touchpoints you happened to record, which is not the same as the touchpoints that mattered.
Touchpoints and Channels
The modern customer journey involves:
- Social media
- Online search
- Email and messaging
- In-store experiences
The tracking problem
Attribution has become harder rather than easier, and any guide that ignores this is misleading you:
- Browser restrictions limit third-party cookies and cap the lifetime of many first-party cookies, which shortens the window in which a journey can be reconstructed.
- Consent requirements mean a share of your visitors are never tracked at all. In some European markets that share is substantial.
- Cross-device journeys break unless the user logs in.
- Platform walls mean each ad platform reports on its own contribution using its own rules, which is why the platforms collectively claim more conversions than you actually had.
The practical consequence: treat attribution output as directional evidence, not measurement. Where a decision is large enough to matter, validate it with a holdout test or a geographic experiment, which measures incremental effect rather than allocating credit after the fact.
Factors Influencing the Journey
| Factor | Description |
|---|---|
| Industry | The industry the business operates in (retail, finance, healthcare). |
| Product | The type of product or service (considered purchase, commodity). |
| Target Audience | The demographics and behaviour of the target customer group. |
| Cycle length | A two-day journey and a six-month journey need different models. |
Single-Touch Attribution Models
Single-touch models give full credit to either the first or last interaction. They are straightforward and structurally wrong, which is fine as long as you know which way they are wrong.
First-Touch Attribution Model
Assigns 100% credit to the interaction that first introduced the customer to your brand.
Pros:
- Shows which channels create initial awareness
- Simple to understand and implement
- Useful when the constraint is finding new prospects
Cons:
- Ignores every subsequent touchpoint
- Systematically over-credits top-of-funnel channels
- Limited by your tracking window: a first touch six months ago is usually invisible
Best For: Short sales cycles with a heavy focus on new customer acquisition.
Last-Touch Attribution Model
Gives full credit to the final interaction before conversion.
Pros:
- Identifies what closes
- Simple to set up and analyze, and the default in most tools
- Least affected by tracking gaps, since the last touch is the one you are most likely to have recorded
Cons:
- Ignores everything that created the demand
- Systematically over-credits branded search and retargeting, which tend to appear at the end of journeys the other channels created
- Encourages cutting the awareness spend that fills the funnel it measures
Best For: Short cycles and bottom-funnel optimisation. It remains the most widely used model largely because it is the default, not because it is the best.
| Model | Key Pros | Key Cons | Ideal For |
|---|---|---|---|
| First-Touch | Shows demand generation channels, simple | Over-credits awareness, tracking window limits | Short sales cycles, new lead focus |
| Last-Touch | Identifies closers, easy setup, robust to data gaps | Over-credits brand search and retargeting | Bottom-funnel optimisation |
A useful diagnostic: run both. Where first-touch and last-touch disagree sharply about a channel, that channel is doing a job in the middle of the journey that neither model captures.
Multi-Touch Attribution Models
Linear Attribution Model
Equal credit to every touchpoint.
Pros:
- Recognizes every touchpoint’s role
- Simple to understand
- No arbitrary weighting decisions to defend
Cons:
- Assumes all touchpoints matter equally, which they do not
- Rewards channels that appear often rather than channels that persuade
Best For: A sanity check against single-touch models rather than a primary model.
Time Decay Attribution Model
More credit to recent touchpoints, usually on an exponential decay with a configurable half-life.
Pros:
- Matches intuition for longer cycles, where recent activity is more predictive
- Handles long journeys without treating a six-month-old touch as equal to yesterday’s
Cons:
- Under-credits the awareness activity that started the journey
- The half-life is a judgement call that materially changes the answer
Best For: Longer sales cycles with nurture activity.
Position-Based Attribution Model
Also called U-shaped. The conventional split gives 40% to the first touch, 40% to the last, and divides the remaining 20% among everything in between, though most tools that still offer it let you change those weights.
| Pros | Cons |
|---|---|
| Credits both demand creation and closing | The weights are a convention, not a finding |
| A reasonable default for considered purchases | Squeezes mid-funnel activity into 20% |
| Easy to explain to stakeholders | No longer available in Google’s tools |
Best For: Complex cycles where both discovery and closing need credit. It is the most defensible rule-based model, which makes its removal from Google’s tools inconvenient.
Data-Driven Attribution Models
Algorithmic Attribution
Data-driven models use machine learning on your historical conversion paths to estimate how much each touchpoint actually contributed, typically by comparing converting and non-converting paths.
Pros:
- Derives weights from your data rather than from a convention
- Adapts as behaviour changes
- Now the default in GA4
Cons:
- Needs meaningful conversion volume. Below that, it is fitting noise.
- Harder to explain, which matters when a finding contradicts what a stakeholder believes
- Correlational, not causal: it identifies touchpoints present on converting paths, which is not the same as touchpoints that caused conversions
- Inherits every gap in your tracking
Best For: Businesses with enough conversion volume for the model to have something to learn from. If you convert a handful of customers a week, a simple rule-based model you understand will serve you better.
Custom Attribution Models
Custom models let you set your own credit rules against your own goals.
| Pros | Cons |
|---|---|
| Aligns with your specific journey | Significant data and engineering effort |
| Weights you can justify to your own business | Needs ongoing maintenance |
| Not dependent on a vendor’s roadmap | Easy to build in your own assumptions and then find them confirmed |
Best For: Larger businesses with analytics resource and a journey that standard models genuinely do not fit.
Comparing Attribution Models
Key Factors
- Cycle length: The single most useful input. Short cycles suit single-touch models; long cycles need time decay or position-based.
- Conversion volume: Determines whether data-driven attribution is viable at all.
- Business goals: Growth through new customers points one way, efficiency the other.
- Channel mix: More channels means single-touch models hide more.
- Tool support: Check the model still exists in your platform before planning around it.
Model Comparison
| Model | Pros | Cons | Best For |
|---|---|---|---|
| First-Touch | Simple, shows awareness channels | Over-credits top of funnel | Identifying what creates demand |
| Last-Touch | Easy, resilient to data gaps | Over-credits brand search and retargeting | Bottom-funnel optimisation |
| Linear | No arbitrary weighting | Rewards frequency over persuasion | A cross-check on other models |
| Time Decay | Suits long cycles | Half-life choice drives the answer | Nurture-heavy B2B journeys |
| Position-Based | Credits both ends of the journey | Weights are conventional; squeezes mid-funnel | Considered purchases |
| Data-Driven | Weights derived from your data | Needs volume; correlational; opaque | Higher-volume businesses |
Setting Up Attribution Models
Tools and Platforms
- Google Analytics 4: Offers data-driven attribution and last-click variants, plus conversion paths reporting. It no longer offers first-click, linear, time decay or position-based, all of which Google removed in 2023.
- Third-party tools: Windsor.ai and Matomo Cloud provide attribution modelling, including rule-based models Google has dropped. Matomo is also worth a look if your reason for moving is data ownership or consent handling.
- Custom solutions: You can model attribution yourself in Python against your own conversion path data, which is the only route to a model you fully control.
None of these publish a single comparable price. Check current rates on the vendor’s own pricing page before committing.
Configuration Steps
- Choose a model that fits your cycle length and conversion volume.
- Set up data: integrate channels, tag campaigns consistently, and set up conversion tracking. Inconsistent UTM tagging ruins more attribution projects than model choice ever has.
- Configure the model according to your tool’s documentation, and record the lookback window you used.
- Test and refine: compare the model’s story against what you know about your business before you act on it.
Analyzing Attribution Data
| Metric | Description |
|---|---|
| Conversion Value | Total value of conversions attributed to each channel |
| Conversion Count | Number of conversions attributed to each channel |
| Return on Ad Spend (ROAS) | Revenue attributed to each channel against its cost |
| Assisted Conversions | Conversions a channel contributed to without closing, which is where single-touch models go wrong |
One check worth running: add up the conversions each ad platform claims and compare the total against your actual order count. The gap tells you how much double-counting is in your reporting, and it is usually larger than expected.
Best Practices and Considerations
Regular Model Review
Review models as your channel mix and cycle length change. Also check that the model you rely on still exists: the removal of four models from Google’s tools stranded a lot of reporting that nobody had revisited.
Combining Attribution Models
Run more than one. Where models agree, you can act with reasonable confidence. Where they disagree, you have found a genuine question rather than an answer. Pick one model as the number of record for reporting, so budget conversations do not turn into arguments about methodology.
Validate with experiments
Attribution divides credit for conversions that already happened. It cannot tell you what would have happened if you had not run a campaign, and that is usually the question you actually care about. For significant budget decisions, run a geographic holdout or a scheduled pause and measure the difference. This is more work than reading a report and it is the only method that answers the causal question.
Challenges and Solutions
| Challenge | Solution |
|---|---|
| Data quality issues | Standardise campaign tagging and audit it regularly. This is the most common root cause of bad attribution. |
| Touchpoint tracking | Encourage logged-in sessions where appropriate; accept that cross-device gaps will remain. |
| Consent and privacy limits | Understand what share of your traffic is untracked, and account for it rather than reporting as if it were zero. |
| Non-marketing factors | Allow for seasonality, pricing changes, and competitor activity when reading results. |
| Resource constraints | Get tagging right first. It costs little and improves every model you might later use. |
Conclusion
Key Points
- Match the model to your sales cycle length and conversion volume
- Four rule-based models were removed from Google Ads and GA4 in 2023; check what your tools actually support
- Every model allocates credit only among the touchpoints you managed to record
- Run more than one model, and treat disagreement between them as information
- Validate significant decisions with holdout experiments rather than attribution reports alone
Why Attribution Modeling Matters
Attribution is how most businesses decide where to spend marketing budget. That makes it worth doing carefully, and worth being honest about its limits. A model that is wrong in a known direction is more useful than one that is wrong in an unknown one.
The Future
| Trend | Description |
|---|---|
| Data-driven as default | Rule-based models are being retired from major platforms |
| Marketing mix modelling | Returning to favour precisely because it does not depend on user-level tracking |
| Incrementality testing | Growing use of holdouts and geo experiments to answer causal questions |
| Server-side and first-party data | Increasing importance as browser tracking narrows |
FAQs
How do I choose an attribution model?
Start with your sales cycle. Short cycles with few touchpoints are served adequately by last-touch. Long, multi-touch cycles need time decay or position-based. Then check conversion volume: data-driven attribution needs enough conversions to learn from, and below that threshold it produces confident-looking noise.
What attribution model works best for ecommerce?
For low-consideration purchases with few channels, last-touch is usually adequate and its bias is well understood. For considered purchases with long research phases, a position-based or data-driven model reflects the journey better. Whichever you choose, watch that branded search does not absorb credit for demand that other channels created.
Is data-driven attribution better?
Better than a fixed rule when you have the volume for it, because the weights come from your data rather than a convention. It is still correlational rather than causal, still limited to the touchpoints you recorded, and harder to explain when it produces an unpopular answer. It is an improvement, not a solution.
Why can’t I find first-click or linear attribution in GA4?
Because Google removed them. First-click, linear, time decay and position-based attribution were retired from Google Ads and Google Analytics 4 during 2023. GA4 now offers data-driven attribution and last-click variants. If you need the rule-based models, you will need a third-party tool or your own analysis.
How does a multi-touch attribution model work?
It splits credit for a conversion across the touchpoints on the path, according to a rule (linear, time decay, position-based) or a learned weighting (data-driven). This shows the contribution of channels that assist without closing, which single-touch models make invisible.
More on this topic
Browse all 53 articles on Data & Analytics.