Attribution

What it is

Attribution is how you connect an outcome, such as a demo request or a sale, back to the interactions that influenced it. It ranges from simple last-click rules in an analytics tool to multi-touch models, media mix modelling and self-reported surveys. Every model is a simplification, so the question is which distortions you can live with.

Why it matters

Budget, headcount and content decisions all rest on attribution, so a flawed model quietly misallocates spend. Discovery through AI assistants, summaries and answer panels often produces no click or an unlabelled one, so channels that genuinely drive demand can look worthless in the default report. Without deliberate attribution work, teams tend to defund the upper funnel and the brand-building content that language models actually cite.

How it works

Practitioners combine platform data with first-party signals: UTM discipline, server-side logging of referrers, custom channel groups for assistant domains, and a "how did you hear about us" field on forms. They then triangulate with holdout tests, geo experiments or incrementality studies rather than trusting a single model. Many teams run a simple model for day-to-day reporting and reserve heavier analysis for quarterly budget reviews.

When it applies

Attribution applies whenever more than one channel or content asset can influence the same conversion, which is nearly always in B2B and considered purchases. It becomes urgent when organic or direct traffic shifts without an obvious cause, or when AI-assisted journeys start appearing in sales conversations.

Examples

• A SaaS company adds a self-reported source field to its demo form and finds a quarter of new pipeline mentions ChatGPT or Perplexity, none of which appears in last-click reports.
• An ecommerce team compares last-click and data-driven models and sees paid brand search credited far more in the former, prompting a brand bidding holdout test.
• An agency builds a custom channel group that separates AI assistant referrers from general referral traffic so the client can track them as a distinct source.

How it is measured

• Share of conversions with an identifiable source versus unattributed or direct
• Self-reported attribution responses by channel, compared with analytics-reported channels
• Assisted conversions and path length by channel in a multi-touch view
• Incremental lift from holdout or geo tests against the credit the model assigns