Attribution is the art of assigning credit for a conversion to the marketing channel that produced it. A customer sees an ad, clicks an email, visits the website, and buys. Which channel deserves credit? The answer matters because it determines where you spend money next.
The problem is that attribution is impossible to measure accurately, and almost nobody acknowledges this.
Why attribution is hard
A customer’s journey typically looks like this:
Day 1: Google search, reads blog post, does not convert. Leaves site. Day 4: Sees a social media ad, clicks through, browses, does not convert. Day 7: Email from your newsletter, clicks through, converts.
Who should get credit? The traditional answer is “last-click attribution” — the email gets credit because it was the last interaction. But the email wouldn’t have worked without the original education from the blog post. And the social media ad increased familiarity. All three were necessary.
The problem gets worse with multiple devices and platforms. A customer uses an iPhone to click an ad, a laptop to read content, and a tablet to buy. Cookies and pixels don’t follow people across devices perfectly. A great deal of the journey is invisible.
Worse, iOS and privacy regulations are killing third-party cookies, which means many journeys are now completely untrackable.
The attribution models nobody should trust
Last-click attribution. Credits the last channel before conversion. Wildly overvalues direct traffic and brand keywords, undervalues awareness-building channels. Bad for measuring marketing mix because it ignores all the legwork that led to the decision.
First-click attribution. Credits the first interaction. Useful for understanding where people first hear about you, but useless for determining which channels drive conversion.
Linear attribution. Credits all channels equally. Superficially fair but disguises actual impact.
Time-decay attribution. Credits interactions more heavily if they are recent. Better than linear but still arbitrary.
Multi-touch attribution. A statistical model that estimates credit based on the probability each interaction led to conversion. Sounds sophisticated, is actually often wrong because the model makes assumptions about behavior it cannot verify.
None of these models is correct. They are all approximations, and they are all wildly different from each other. It is entirely possible that the same conversion gets attributed to different channels depending on which model you use.
What to measure instead
Stop trying to attribute. Measure differently.
Incremental impact. Turn off a channel entirely for a week and measure the change in conversions. If conversions drop 10%, that channel was driving 10% of conversions. This is crude but honest. Do it quarterly, not constantly — constantly toggling channels is disruptive.
Cohort analysis. Group customers by their first interaction source and compare their lifetime value. Customers who first came from organic search might convert at 5%, but then spend £500 over a lifetime. Customers from paid ads might convert at 15%, but only spend £200 lifetime. This tells you which sources are actually valuable.
Surveys. Ask customers “where did you first hear about us?” and “what was the final factor in your decision?” This is qualitative and unreliable but often reveals what quantitative models miss.
Regression analysis. Model conversions as a function of spend across channels. If you spend 10% on channel A and 90% on channel B, and conversions grow 5% when you shift to 20% on channel A and 80% on channel B, that suggests channel A is more efficient. This requires sufficient data and change to detect, but it works.
Controlled experiments. If you can, run true experiments: do not run a channel for one cohort of people, run it for another cohort, and compare outcomes. This is the most honest measurement available.
What attribution models are actually good for
Attribution models have one legitimate use: internal consistency. If you pick a model and stick with it, you can see trends over time. “Last-click attribution says email is our best channel, and it’s been improving” is meaningful relative data even if it’s not true absolute data.
They are useful for spotting obvious patterns. If every model shows organic driving significant conversions, organic probably is. If every model shows a channel driving nearly nothing, it probably is.
They are useful for benchmarking. Marketing teams knowing that other teams use last-click attribution can compare results.
They are useless for precise optimization. “We should shift 5% budget from paid search to email because email gets higher attribution credit” is a bad decision. You do not know that enough.
The honest approach
Build a mental model of your customer journey rather than relying on a model to build it for you.
Customers typically: 1. Discover you through search, a recommendation, or an ad. 2. Learn about you through content, a free trial, or talking to someone. 3. Compare you to alternatives through reviews, conversations, demos. 4. Make a decision through pricing, a specific need, or a timely promotion. 5. Convert.
Each stage has channels that work better. Awareness is search and ads. Consideration is content and reviews. Decision is email and sales. Assign your spending proportional to what stage of the funnel is your actual bottleneck.
Is most of your drop-off in awareness? Spend on ads. In consideration? Content and partnerships. In decision? Email and sales outreach.
This is rougher than attribution modeling but much more honest.
The metrics that actually matter
Stop obsessing over attribution and start measuring what you actually care about:
CAC (customer acquisition cost). Total marketing spend divided by customers acquired. If it is £200 per customer and your average customer lifetime value is £1,000, you are doing well. If it is £500 and LTV is £600, you are losing money. Attribution does not affect this calculation.
LTV by cohort. Separate customers by when they were acquired and how much they spend. This tells you if your marketing is improving quality or just volume.
Payback period. How long until a customer’s spending covers the cost of acquiring them? Short payback periods are usually better.
Efficiency ratio. Revenue from marketing divided by marketing spend. A 3:1 ratio means for every pound of marketing you get three pounds of revenue. Attribution does not affect this.
These metrics completely bypass the need for attribution. They tell you if your marketing works.
The future of attribution
Apple’s privacy changes, regulatory pressure on tracking, and the death of third-party cookies are making attribution harder and less accurate. The future is not better attribution models. It is marketing teams accepting that perfect attribution is not available and measuring differently.
The teams that adapt fastest to measuring incrementally, through cohort analysis and experimentation, will adapt fastest to a world without tracking. The teams that cling to attribution models will find themselves unable to measure at all once tracking disappears entirely.
Embrace the chaos. Measure what you actually care about. Stop expecting attribution to be more accurate than it can be.