What's the right attribution model?

There isn't one. Use multi-touch for within-channel optimization, self-reported for channel mix decisions, and holdout tests for causal claims. Any vendor selling you a single "right" model is selling you comfort, not accuracy.

Should we buy an attribution tool?

For $10M+ ARR with meaningful paid ad spend ($50K+/month), yes — Dreamdata, HockeyStack, or similar. Below that, self-reported attribution + UTMs in a spreadsheet does 80% of the job at 0% of the cost.

How does self-reported attribution scale?

It doesn't need to scale — it just needs to run continuously. 100-500 responses per quarter is enough to see channel mix trends. Have a marketer categorize responses weekly (10 minutes/week).

Marketing Attribution for B2B Startups: 2026 Guide

Every attribution model is wrong. Multi-touch, first-touch, last-touch, and self-reported all miss reality.

Marketing Attribution: Why Every Model Lies and Which One Lies Least

Marketing attribution is the practice of assigning credit for pipeline and revenue to marketing touchpoints. It is also one of the most quietly dishonest practices in B2B — because every attribution model produces different answers, and each model systematically favors channels the tool vendor can measure. The honest posture: no attribution model is right, several models together are useful, and self-reported attribution is often the least-wrong signal you have.

The four common models

First-touch: 100% credit to the first touchpoint. Overweights top-of-funnel channels (SEO, content, ads). Last-touch: 100% credit to the last touchpoint before conversion. Overweights bottom-of-funnel (branded search, demo request forms, sales outbound). Multi-touch (linear, U-shaped, time-decay): distributes credit across touchpoints. Feels sophisticated but relies on tracking that misses 30-60% of real touchpoints (LinkedIn, podcasts, offline). Self-reported ("how did you hear about us?"): asks the buyer directly. Imprecise but captures dark-social touchpoints no tool sees.

What multi-touch attribution actually misses

Podcast listens (not trackable). LinkedIn scrolling without a click (dark social). Conference conversations. Peer recommendations in Slack communities. Sales reps' personal LinkedIn presence. Direct traffic (often attributed to "direct" when it's actually brand-driven from these dark channels). Companies that rely purely on tool-based attribution consistently under-invest in the channels producing the most demand, because those channels don't show up in the dashboards.

The self-reported attribution question

On every demo request form, ask: "How did you hear about us?" as an open-text field (not dropdown — dropdowns pre-load your assumptions). Review responses monthly. Categorize into: search, referral, social (LinkedIn/Twitter/etc), podcast, community, event, direct/other. In practice, self-reported attribution shows 40-60% of pipeline coming from channels that tool-based attribution assigns to "direct" or "organic search."

The right way to use attribution

Use tool-based multi-touch for directional trends within measurable channels (paid ads, gated content, email). Use self-reported for the big-picture channel mix decision (should we invest more in podcast or SEO?). Use holdout tests for causal claims (turn off a channel in one geo for 90 days, measure impact — this is the only true causal test). Never present a single attribution number to the board without disclosing the model and its limitations.

Common mistakes

Picking one model and treating it as truth. Making channel investment decisions solely on multi-touch attribution (kills dark-social investments). Not asking self-reported attribution. Not running holdout tests. Assigning revenue credit for comp purposes based on attribution models (creates political warfare between sales and marketing).

Frequently asked questions

What's the right attribution model?: There isn't one. Use multi-touch for within-channel optimization, self-reported for channel mix decisions, and holdout tests for causal claims. Any vendor selling you a single "right" model is selling you comfort, not accuracy.
Should we buy an attribution tool?: For $10M+ ARR with meaningful paid ad spend ($50K+/month), yes — Dreamdata, HockeyStack, or similar. Below that, self-reported attribution + UTMs in a spreadsheet does 80% of the job at 0% of the cost.
How does self-reported attribution scale?: It doesn't need to scale — it just needs to run continuously. 100-500 responses per quarter is enough to see channel mix trends. Have a marketer categorize responses weekly (10 minutes/week).