Insights / Perspective / Digital campaigns and sales growth
PerspectiveAttribution is imperfect, and that is fine
No attribution model tells the whole story of how customers choose. Use several views, test with real experiments and stop overclaiming credit.
No model tells the whole story of how customers choose. A buyer might see a social post, hear about you from a colleague, search your name a week later and finally click an email before they order. Every attribution model picks one way of dividing the credit for that sale, and every one of them is partly wrong.
That is not a reason to give up on measurement. It is a reason to stop treating any single report as the truth. The organisations that spend well on marketing are not the ones with the cleverest model. They are the ones that look at the evidence from several angles, test their assumptions and are honest about what they do not know.
Why every model gets it partly wrong
Attribution software can only count what it can see. It sees clicks, form fills and tracked visits. It does not see the conversation at a trade show, the recommendation in a group chat or the billboard on the drive to work. Privacy changes in browsers and phones have made the visible part smaller still.
Each model then applies a rule to the visible part. Last-touch gives all the credit to the final click, which flatters search and email because people use them when they have already decided. First-touch gives it all to the opening contact, which flatters awareness channels. Weighted and data-driven models spread the credit, but they still only spread it across the touches they recorded.
The core problem
Attribution describes the path a customer took. It does not tell you what would have happened if one step had been missing. That second question is the one budget decisions actually depend on.
Use several views, not one
Compare first-touch, last-touch and incremental tests side by side. Where they agree, you can act with confidence. Where they disagree, you have found the channels that deserve a closer look.
A typical mid-sized business-to-business firm might find that paid search looks excellent on last-touch and modest on first-touch, while industry events look poor on last-touch but start a large share of the deals that eventually close. Neither view is wrong. Together they suggest search captures demand and events create it, which is a very different conclusion from “cut events”.
- First-touch shows which channels introduce new people to you.
- Last-touch shows which channels are present when people decide.
- Multi-touch or data-driven views show which channels tend to appear together.
- Incremental tests show what changes when a channel is switched off or turned up.
Run experiments to find out what really works
The most reliable way to learn what a channel contributes is to change it and watch the result. Pause a channel in one region and keep it running in another. Double spend for a fixed period. Hold back a random group of customers from an email programme. Then compare sales, not clicks.
These tests do not need to be elaborate. They do need a clear question, a fixed period long enough to cover your normal buying cycle and an agreement in advance about what result would change the plan. Without that agreement, every result gets explained away.
- Pick one channel where the models disagree most, or where spend is largest.
- Choose a comparable control: a region, a customer segment or a time period.
- Agree the measure that matters, usually orders, qualified leads or revenue.
- Run the change for at least one full buying cycle.
- Record the result and the decision it led to, even if the result was “no difference”.
There are trade-offs. Pausing a channel costs you whatever it was genuinely contributing during the test. Regional splits are noisy when regions differ in size or season. But a modest, well-designed test usually teaches more than a year of arguing over dashboards.
Avoid overclaiming
Credit is shared across the journey. When every channel owner reports the sales their channel touched, the total often adds up to far more than the business actually sold. That is a sign the reports are being used to defend budgets rather than to understand customers.
Once we stopped asking which channel won the sale and started asking which mix of channels grew sales, the budget meetings got a lot shorter.
Marketing director, professional services client
Leaders can help by changing the question. Instead of asking each team to prove its value in isolation, ask for the combined view and for the experiments that would settle open questions. Reward teams for honest uncertainty as much as for good numbers.
What good looks like
Mature attribution is a habit rather than a tool. Teams look at two or three models each month, know which channels the models disagree about, and keep a short list of experiments to resolve those disagreements. Budget moves are small and frequent, and each one is checked against what actually happened to sales.
It also means being clear about the limits. A good monthly report says which conclusions are well supported, which rest on a single model and which are still open. That honesty makes the report more useful to a board, not less, because it shows where the next pound of spend is a safe bet and where it is a deliberate experiment. Over a year, the list of open questions should get shorter, and the share of budget sitting on well-tested channels should grow.
Start this quarter with one comparison and one test. Put first-touch and last-touch reports side by side, pick the channel where they disagree most and design a simple pause or increase. Whatever you learn will be more useful than a more sophisticated model you cannot check.
Source: Intelliversal analysis.
