Marketing Attribution: Why the Last Click Rarely Tells the Whole Story
A customer sees a social video, reads a native article a week later, searches for the brand and buys. Last-click attribution gives all the credit to the brand search. Cut the social and native budgets on that basis, and branded searches will quietly start to fall.
What attribution tries to do
Attribution assigns credit for a conversion to the marketing touchpoints that preceded it. It exists because most customer journeys involve more than one interaction, and businesses want to know which investments are working.
The difficulty is that attribution observes correlation, not causation. It records which touchpoints happened before a conversion, not which ones caused it. Every model is a set of assumptions about how to share credit between them.
The common models
| Model | How it assigns credit | Tends to favour | Blind spot |
|---|---|---|---|
| Last click | All credit to the final click | Branded search, retargeting, direct | Ignores everything that created demand |
| First click | All credit to the first known click | Prospecting channels | Ignores what closed the sale |
| Linear | Equal credit to every touchpoint | No channel in particular | Treats a glance and a decisive visit equally |
| Time decay | More credit to touchpoints closer to conversion | Lower-funnel channels | Undervalues early demand creation |
| Position-based | Most credit to first and last, some to middle | Both ends of the journey | Arbitrary weighting |
| Data-driven | Credit based on observed paths that convert vs. don’t | Depends on data | Still correlational; limited to tracked touchpoints |
Why last click persists
Last-click attribution is simple, easy to explain and available everywhere. It also produces stable numbers that make lower-funnel channels look highly efficient, which is comfortable for anyone managing them.
The problem is systematic bias. Channels that capture existing demand, such as branded search and retargeting, receive credit for demand created elsewhere. Channels that create demand, such as paid social, native, video and content, appear inefficient because they rarely deliver the final click.
What attribution cannot see
Even sophisticated models are limited to the touchpoints that can be tracked:
- Ad views without clicks, especially on platforms that do not share impression data
- Offline influences: word of mouth, events, print, TV, podcasts
- AI assistants and search features that inform decisions without a click
- Cross-device journeys where the user cannot be linked
- Visitors who decline tracking consent
- Organic brand awareness built over years
Attribution reports present a precise-looking answer drawn from an incomplete picture.
Platform attribution adds another layer
Each ad platform reports conversions using its own rules and attribution windows, and each tends to claim every conversion it touched. Add up platform-reported conversions and the total frequently exceeds actual sales. This is one reason platform CPA and business CAC diverge.
A practical measurement framework
Rather than searching for a perfect model, we combine methods, using each for what it does best.
1. Attribution for tactical decisions within channels
Use attribution to compare campaigns, ad groups and creative within one platform, where touchpoints and rules are consistent. It is good at telling you which of two similar ads works better.
2. Incrementality testing for channel value
Incrementality tests measure what would have happened without a marketing activity. Common designs include:
- Holdout tests: a random portion of the audience is not shown ads, and conversion rates are compared.
- Geographic tests: spend is changed in some regions and not others, and outcomes are compared.
- On/off tests: spend is paused for a period, with care taken to account for seasonality.
These are the most reliable way to learn whether a channel causes conversions, or merely precedes them. Branded search and retargeting tests often reveal lower incremental value than attribution suggests. Prospecting tests often reveal higher.
3. Marketing mix modelling for budget allocation
Marketing mix modelling uses statistical analysis of spend and outcomes over time, including offline factors and seasonality, to estimate each channel’s contribution. It does not depend on user-level tracking, which makes it increasingly useful as tracking declines. It needs sufficient historical data and careful interpretation.
4. Customer research for what data cannot see
Ask new customers how they heard about you. Self-reported attribution is imperfect, but it captures word of mouth, podcasts, AI recommendations and other influences invisible to tracking. Disagreement between survey answers and attribution data is often where the most useful insights are.
Making decisions anyway
Attribution will never be perfect. Decisions still need to be made. Some principles we apply:
- Never cut demand-creating channels purely on last-click data. Test first.
- Treat branded search with scepticism. Measure how much would arrive organically.
- Look at total business outcomes alongside channel reports. If total new customers fall after cutting a channel that “wasn’t working”, it probably was.
- Agree the measurement approach before the campaign, so results are not reinterpreted afterwards.
- Document assumptions. Every model has them; write them down so they can be challenged.
The bottom line
The last click is a fact about a customer journey. It is not an explanation of it. Businesses that understand this make better budget decisions, and avoid the slow decline that follows when demand creation is cut because a report said it was not working.
Building measurement that balances these methods is central to our analytics and attribution work. For the data foundation underneath it, see first-party data and the future of performance marketing.
