Attribution models assign credit for conversions across the many touchpoints in a customer journey.
Marketing attribution is the practice of assigning credit for revenue back to the marketing touchpoints (ads, content, events, emails) that influenced the customer's journey. Every model — first-touch, last-touch, linear, U-shaped, time-decay, data-driven — is directionally wrong because attribution is fundamentally unknowable at the individual level. The useful question isn't 'which model is right?' but 'which model is least misleading for the specific decision I'm making?'
(1) First-touch — 100% credit to the first touchpoint. Good for evaluating top-of-funnel channels. Underweights everything downstream. (2) Last-touch — 100% to the last touchpoint. Overweights bottom-funnel channels (branded search always looks great). (3) Linear — equal credit across all touchpoints. Fair but uninformative. (4) U-shaped — 40% first-touch, 40% last-touch, 20% distributed. Popular pragmatic choice. (5) Time-decay — more credit to recent touchpoints. Useful for short sales cycles.
Use first-touch to answer: 'which channels bring in new audiences?' Use last-touch for: 'which channels close deals?' Use U-shaped for: 'balanced view of full funnel.' Use time-decay for: 'which channels drove recent revenue?' The mistake is picking one model and reading it as truth. Sophisticated marketing teams look at multiple models side-by-side and use the disagreement to spot channel-mix issues.
GA4's data-driven attribution and platform-specific models (Meta, Google Ads) use ML to weight touchpoints based on observed conversion patterns. In principle: better than a static rule. In practice: opaque, sensitive to data volume (needs 400+ conversions/month per channel to be meaningful), and vulnerable to platform-specific bias (Meta will attribute more to Meta). Useful as one input, dangerous as sole source of truth.
The only real way to know what marketing actually caused: hold-out tests where you turn a channel off in some markets/segments and measure the delta. Ghost bidding tests (Meta, Google) simulate incrementality without full holdouts. Geo-experiments (turn off paid search in 5 states, measure delta) are the gold standard. Every marketing team should run at least one incrementality test per year per major channel — attribution reports without incrementality checks systematically over-credit paid channels.
B2B complicates attribution: multi-person buying committees, long sales cycles, self-reported source data ('how did you hear about us?') that reveals 20%+ of true influence not captured in tracking. Standard B2B pattern: (a) tracked attribution for online channels, (b) self-reported source for offline/word-of-mouth influence, (c) revenue-attributable pipeline reporting from Salesforce/HubSpot campaign influence, (d) periodic incrementality tests. The combination beats any single model.
Investor directory · Fundraising library · Articles A–Z · Company funding database