Marketing Attribution Models: Which One to Use and When to Stop

The models, and how each one lies

There are four you will meet, and each systematically over-credits one part of the journey.

Linear and time decay

Linear splits credit evenly across every touch. It is the fairest sounding and the least decisive, because it tells you that everything helped a little and gives you nothing to act on. It is genuinely useful for one thing: seeing which channels appear in journeys at all, including ones that never win first or last touch and would otherwise be invisible.

Time decay gives more credit to touches nearer the sale. It is a reasonable compromise for businesses with a long consideration period, where the early touches are real but the later ones clearly did more work. It still under-credits the thing that started the relationship, just less brutally than last touch.

There are more sophisticated approaches, including data-driven models that infer weights from your own conversion patterns. They need a lot of conversions to be meaningful, which puts them out of reach for most small businesses and makes them a distraction until you are much larger.

What none of them can see

This is the part that matters most and gets discussed least. Attribution only counts touches that were tracked, and a large share of real influence is never tracked at all.

Someone hears about you on a podcast and searches your name a week later. Someone reads your post at work, sends it to a colleague on a messaging app, and the colleague arrives with no referrer. Someone sees you mentioned in a newsletter and types your domain directly. Every one of those shows up as direct traffic or branded search, and last touch attribution will hand the credit to your brand, which is not a channel you can invest in.

The practical consequence is that your dark channels are systematically undervalued in every model, and the channels that get overvalued are the ones that happen to be trackable. Budget decisions made purely on attribution data drift toward whatever is easiest to measure, which is not the same as whatever works.

The cheapest fix nobody uses

Add one optional field to your signup or enquiry form: how did you hear about us? Free text, not a dropdown.

Self-reported attribution is imprecise and people misremember. It is also the only instrument that can see the podcast, the forwarded message and the conversation. Run it beside your analytics and the gaps between the two are the most interesting data you will have.

The pattern to expect is that a channel your analytics rates poorly gets mentioned constantly in the free text. That is not a contradiction to resolve; it is the tracked model showing you its blind spot.

• One optional free-text field on the form, not a dropdown of your guesses
• Read the answers monthly, in full, rather than counting them
• Compare against your analytics and treat the gaps as findings
• Expect word of mouth and content to be understated by tracking
• Do not replace analytics with it, run both

When to stop thinking about this

Attribution work has sharply diminishing returns and a strong pull toward becoming the work itself.

If you have three marketing channels, you do not need a model. Look at first touch and last touch side by side, note where they disagree, and spend the saved time making the content better.

The point at which a proper model earns its cost is when you are deciding between meaningfully different budgets across many channels, and when the volume of conversions is high enough that the patterns are real rather than noise. Below that, the honest position is that you are directionally informed and precisely uncertain, and acting on the direction is the correct response.