<b>Multi-touch attribution is mandatory in B2B — and still systematically biased</b>

<b>Multi-touch attribution is mandatory in B2B — and still systematically biased</b>

Single-touch models (first- or last-touch) are indefensible for B2B, where a closed deal commonly involves 6–10 stakeholders and dozens of touches across months. Yet the move to multi-touch attribution (MTA) trades one bias for several subtler ones worth naming.

<b>Known MTA biases:</b>

— <b>Touch-counting bias.</b> Linear models reward channels that generate many cheap touches (email, retargeting) over channels that generate few decisive ones (a partner-led executive briefing).
— <b>Tracked-touch bias.</b> MTA can only credit digitally captured interactions. Partner-driven offline influence — a referral conversation, an event introduction — is structurally undercounted, again penalizing partners.
— <b>Position-weighting arbitrariness.</b> U-shaped, W-shaped, time-decay: each encodes an untested assumption about which positions matter. The choice of curve, not the data, often drives the conclusion.

The rigorous alternative is incrementality testing — holdout or geo experiments that estimate causal lift rather than dividing fractional credit. MTA answers 'what got touched'; incrementality answers 'what would not have happened otherwise.' These are different questions, and partner value lives in the second.

<b>A practical synthesis:</b> use MTA for operational pipeline reporting and incrementality experiments for budget and partner-commission decisions. Do not let a correlational credit-split masquerade as causal proof of contribution.

<b>Implications:</b> the model you pick is a hypothesis about influence, not a measurement of it.

<b>Open questions:</b> how do you run a clean partner-channel holdout when partners self-select the accounts they work?