Every marketing organization has an attribution model. Most of them are fiction, and the more confident the marketing team is about their attribution math, the more fictional it probably is. This is not a cynical take — it’s just the honest accounting.
The reason matters. If you don’t know what’s actually driving pipeline, you don’t know what to keep doing, what to stop doing, and what to invest in. Attribution that pretends precision it doesn’t have leads to worse decisions than no attribution at all.
Why Attribution Models Lie
1. First-touch and last-touch are both wrong.
Every attribution model is some variation on “which touchpoint caused the pipeline.” First-touch attributes to the top of the funnel — the ad, the content, the event that the customer first saw. Last-touch attributes to the bottom — the demo signup, the pricing page visit, the meeting booked. Both are accurate for about 5% of buyers and misleading for the other 95%.
The reality is that enterprise buyers typically interact with 15 to 30 touchpoints over a 6 to 18 month journey before they buy. Attributing the pipeline to any single touch is arithmetically convenient and substantively wrong.
2. Multi-touch attribution models are tunable fiction.
“U-shaped,” “W-shaped,” “time-decay,” “position-based.” These models assign weights across touchpoints. The weights are chosen by the marketing team. Different weights produce different conclusions. The model isn’t discovering truth — it’s reflecting whatever assumptions got baked into the weights.
3. The touchpoints that matter most are invisible.
Word of mouth. Private conversations. A customer mentioning your product in a Slack community the marketing team doesn’t see. These are often the dominant drivers of pipeline, and they leave no data. The attribution model compensates by over-crediting the visible touchpoints — producing a model that systematically over-weights what’s measurable.
What Marketing Should Actually Do
1. Measure what’s unambiguously attributable.
Direct response: someone clicked on an ad and signed up for a demo. That’s attributable. Count that. Optimize that. Don’t pretend the attribution extends further than it actually does.
2. For everything else, run experiments.
Turn off a channel for a quarter. See what happens. If pipeline drops, the channel was contributing. If it doesn’t, the channel was vanity. This is crude but more honest than model attribution.
3. Ask the customer.
In customer onboarding, ask: “How did you first hear about us? What made you evaluate us? Who else did you consider?” This is qualitative data, but it’s qualitative data from the only source that actually knows — the customer. It’s worth more than any attribution model.
The Danger of Attribution Theater
The deeper risk of bad attribution is that it’s self-reinforcing. Marketing gets credit for what the model says they did. The team doubles down on those channels. The channels that were actually driving pipeline but weren’t captured by the model get underfunded. Over time, marketing optimizes itself into a corner.
I’ve watched this happen in multiple organizations: the attribution model shows content marketing is high-ROI, so the team doubles content investment. Pipeline drops. The model still says content is the dominant driver, because the model can’t see the shift in word-of-mouth that was actually the wind behind the growth. The team can’t figure out what’s broken because their own measurement is lying to them.
The fix is to be less confident about what your marketing is doing. Treat attribution as directionally useful, not precisely true. Validate with experiments and customer conversations. And resist the executive pressure to present attribution as if it were accounting.
What to Tell the Board
When a board or executive team asks “which channel is driving the pipeline,” the honest answer is usually “we can attribute roughly X% with high confidence. For the remaining Y%, we have hypotheses we’re testing, but we don’t have certainty.”
This sounds weak. It’s actually the only honest version. CMOs who present confident attribution numbers are either using flawed models or hiding what they don’t know. The ones who present with appropriate uncertainty build more trust over time, not less.
Marketing attribution is useful as a working hypothesis, not as a source of truth. Treat it that way, and you’ll make better decisions. Treat it as precise, and you’ll eventually invest in the wrong things for long enough that the growth trajectory suffers.
The measurement is valuable. The certainty is the lie.