Marketing attribution

The models, and what each one hides

An attribution model is a rule for splitting one conversion across several touches. Every model is a simplification, and each one systematically flatters a different part of the funnel.

• First-touch — all credit to the first interaction. Flatters discovery channels like SEO, social, and PR, and makes closing channels look worthless.
• Last-touch — all credit to the final interaction before conversion. The default in most analytics tools, and the reason branded search and retargeting always look like heroes: they are simply what people touch last.
• Linear — credit split evenly across every touch. Honest about the fact that several things contributed, useless for deciding what to cut.
• Time-decay — later touches get more weight. Reasonable for long sales cycles, arbitrary in how fast the decay is set.
• Position-based — usually 40/20/40 across first, middle, and last. A compromise that nobody can defend from first principles but most teams find survivable.
• Data-driven — the platform models the incremental contribution of each touch. Only meaningful with a large, steady conversion volume, and it is a black box you are asked to trust.

Which one is *correct* is the wrong question. The useful question is what decision you are making. Deciding where to spend the next dollar of acquisition budget is a last-touch and incrementality question; deciding what content to write more of is a first-touch question. Feeding both decisions from one model is how teams end up defunding the channel that fills their pipeline.

Why attribution breaks for small startups

Attribution was designed for businesses with thousands of conversions a month. Under that volume it degrades into noise, and three forces make it worse. Small numbers mean a single week's difference between two channels is usually random. Privacy changes — cookie restrictions, tracking prevention, and mail clients that pre-fetch images — quietly delete the trail. And the highest-intent discovery now happens in places that pass no referrer at all: a recommendation inside a private Slack, a podcast, or an AI assistant that names you in an answer the user never leaves.

Our own site shows the gap plainly. Over 28 days, Google Search Console recorded 10,617 impressions and 98 clicks for agentceres.com, and every one of those clicks ties back to a specific query and page. There is no equivalent record for the visitor who arrived because an assistant mentioned us inside a conversation we cannot see — which is exactly why generative engine optimization is measured with brand-mention tracking and self-reported answers rather than a referrer column.

What to use instead when your numbers are small

Below roughly a hundred conversions a month, a simpler stack beats a modelled one — and it is what most funded startups quietly fall back to anyway.

• Ask. A single optional *How did you hear about us?* field on signup outperforms any model at this size, because it captures the untrackable channels that produce your best customers.
• Tag what you control. Consistent UTM parameters on every link you place, and a link tool such as Dub if you place many, so at least the paid and campaign side is unambiguous.
• Pick one source of truth. Ad platforms count conversions they helped cause; your own analytics counts conversions that happened. They will never match, so decide in advance which one you argue from.
• Test by turning things off. With small numbers, pausing a channel for two weeks and watching total signups tells you more about incrementality than any attribution report can.
• Judge channels on cohort economics. Customer acquisition cost and retention by source beat click-path credit, because a channel that sends cheap signups who churn is not a channel that works.

Attribution is a decision aid, not a scoreboard. When it stops changing a decision, stop refining it — and if your problem is that too few visits arrive to attribute in the first place, why a site gets no traffic is the upstream question to answer first.