Attribution is the question every CMO gets asked and no model honestly answers: which marketing dollars created this revenue? Every attribution model is a simplification, and every simplification has a direction of bias. The danger is not using a simple model - it is forgetting which way your model lies.
First Touch
First touch gives 100 percent of the credit to the channel that first brought the buyer into your world. Its story is about demand creation: what made this person aware of us at all?
How it lies: it systematically over-credits top-of-funnel channels - content, paid social, events - and under-credits everything that turned interest into a deal. Optimize purely on first touch and you will fund awareness while starving the nurture and sales-enablement programs that actually convert. It also depends entirely on tracking the true first touch, which cookie loss, dark social, and word-of-mouth make increasingly fictional. The recorded "first touch" is often just the first touch you could see.
Last Touch
Last touch gives all the credit to the final interaction before conversion. Its story is about demand capture: what closed the loop?
How it lies: it over-credits the bottom of the funnel, especially branded search and direct traffic. A buyer who read your content for six months and finally googled your company name gets attributed to "branded search," and the six months of content that did the persuading gets zero. Optimize on last touch and you will conclude that branded search is your best channel - which is like concluding that the cash register is what sells the groceries.
Multi-Touch
Multi-touch models spread credit across the journey: linear (equal credit to every touch), time-decay (more credit closer to conversion), U-shaped (40/40/20 weighted to first touch and conversion), W-shaped (adding opportunity creation as a third anchor).
How it lies: multi-touch feels rigorous but inherits every tracking gap - it just distributes credit across an incomplete record with more decimal places. The weights themselves are arbitrary: nobody has evidence that the first touch deserves exactly 40 percent. And multi-touch counts touches, not influence - a webinar someone attended for 45 minutes and a display ad that technically loaded on a page both count as one touch. The precision is cosmetic; the model is still opinion expressed as arithmetic.
The B2B-Specific Failure
All contact-level attribution breaks in B2B because deals are made by buying committees, not contacts. The person who filled the form is rarely the person who signed. If your attribution only follows the lead that converted, you miss the five other stakeholders marketing touched along the way. Account-based attribution - aggregating every touch across every contact at the account before the opportunity - is messier but far closer to how enterprise deals actually happen.
How to Actually Use Attribution
Run two models and read the disagreement. First touch and last touch bracket the truth. A channel that ranks high on both is genuinely strong. A channel that ranks high on exactly one is telling you what stage of the funnel it works at - which is more useful than any single blended number.
Treat attribution as budget allocation, not performance review. The question is not "which channel gets the trophy" but "where does the next dollar go." For that decision, directional accuracy is enough - and chasing the last 10 percent of accuracy costs more than the decisions it improves.
Cross-check with incrementality when stakes are high. Attribution measures correlation in observational data. For your biggest channels, run the real experiment: pause the channel in a geography, run a holdout, compare. Teams that do this routinely discover that a channel attribution loved was harvesting demand that would have converted anyway.
Ask the buyer. Self-reported attribution - a "how did you hear about us?" field on the demo form - is unfashionable and consistently surfaces the channels tracking cannot see: podcasts, communities, colleague recommendations. When self-reported and tracked attribution disagree sharply, the self-reported answer is usually pointing at your blind spot.
Attribution done well is not a scoreboard. It is an instrument panel - and like any instrument, the reading only helps if you know how the gauge is calibrated and which direction it drifts.