Multi-Touch Attribution

Multi-touch attribution is a measurement framework that assigns fractional credit for a conversion to each touchpoint that contributed to it. Where last-touch attribution gives 100% of the credit to the final touchpoint before conversion, multi-touch divides the credit across all touches in the journey using a defined model.

The most common multi-touch models:

• Linear: equal credit to each touch.
• Time-decay: more recent touches get more credit, older touches get less.
• U-shaped (position-based): 40% to first touch, 40% to last touch, 20% spread across the middle.
• W-shaped: 30% each to first touch, lead conversion, and opportunity creation, with 10% spread across the rest.
• Algorithmic or data-driven: a machine-learning model assigns credit based on which touch sequences correlate with conversion.

No model is correct. Each makes different assumptions about how influence works. The model choice depends on the business: position-based fits long, multi-stakeholder B2B journeys; time-decay fits shorter, transactional journeys.

B2B buying journeys are long. The average enterprise software purchase involves 6 to 10 stakeholders and 27 touchpoints over 6 to 12 months (Gartner). Last-touch attribution credits whichever channel produced the final click, typically branded search or direct traffic, and ignores the brand, content, and demand-creation work that built the awareness in the first place.

The practical effect: under last-touch, brand-building investments look unproductive and capture-channel investments look magical. Teams that report only last-touch attribution typically over-invest in capture channels and under-invest in creation, which is the opposite of the optimal allocation.

Multi-touch attribution corrects the bias. A position-based or W-shaped model surfaces the early-funnel touches (content downloads, podcast listens, employee advocacy posts) that initiated the journey, even when they were weeks or months from the final conversion.

Five practical steps:

1. 1.Choose a model. Position-based and W-shaped are the most common B2B starting points. Algorithmic models require sufficient conversion volume to train on (usually 1,000+ conversions per month).
2. 2.Instrument the touches. UTM parameters on every link, cookie persistence, and deterministic identity matching where possible (form-fill matching, CRM linkage). Without instrumented touches, the model has nothing to work with.
3. 3.Define the conversion event. Be explicit. Closed-won is the truest endpoint but lags by months. Most teams use opportunity creation as the operational conversion event for attribution reporting.
4. 4.Layer in self-reported attribution. Add a "how did you hear about us" question on demo forms and customer onboarding. The deterministic answer is often more useful than the modelled one for hard-to-track channels.
5. 5.Reconcile against revenue. Compare attributed pipeline to actual closed-won revenue by channel. Persistent gaps signal model assumptions that do not fit the business.

The single biggest implementation mistake is treating multi-touch attribution as the answer rather than as a directional input. Use it to make spending decisions, not to litigate channel-by-channel ROI to two decimal places.

Multi-touch attribution has known limitations:

• Cookie loss. Browser privacy changes (Safari ITP, Firefox ETP, Chrome cookie deprecation) have eroded cookie persistence. Attribution coverage has dropped from roughly 90% to 50 to 70% in 5 years.
• Cross-device gaps. A buyer who reads a blog post on mobile, watches a webinar on desktop, and converts via tablet is rarely tracked as one journey unless deterministic identity is in place.
• Offline touches. Conferences, dinners, and partner introductions rarely make it into attribution models, despite being major B2B influences.
• Lag. The full influence of brand-building shows up months or years later. Attribution models that look at the last 30 days under-credit long-cycle work.

The healthy practice is to treat multi-touch attribution as one of several signals, alongside marketing-mix modelling (a top-down revenue-and-spend regression), self-reported attribution, and incrementality tests. Each captures part of the picture; none is complete on its own.