Why platform attribution falls short
- Short windows: defaults of 30 days for clicks and 1 day for views.
- Many people: the person who clicked isn't the person who signs.
- Hidden visits: AI traffic often arrives without a referrer.
Four layers that work together
| Layer | What it tells you |
|---|---|
| Platform conversions | Short-term response to ads |
| First AI touch on the account | Which accounts AI brought in, from the company record |
| Self-reported source | "ChatGPT or another AI assistant" on forms and calls |
| Opportunity influence | Opportunities with any AI touch before creation |
How to set it up
- Store first-touch source and buyer question on the company record.
- Add an AI option to "How did you hear about us?"
- Report opportunities with an AI touch every month.
- Compare win rates and deal sizes with and without an AI touch.
A simple influence report
| Quarter | Opportunities created | With an AI touch before creation | Share | Win rate with AI touch | Win rate without |
|---|---|---|---|---|---|
| Q1 | |||||
| Q2 |
Questions
Is self-reported attribution reliable?
It's imperfect but catches what tracking misses, especially AI recommendations. Use it alongside tracked data.
Should we use multi-touch models?
If you already have one, add AI sources to it. If not, start with first AI touch and influence.
How long should we wait before judging?
At least one full sales cycle.
Why AI makes attribution harder
AI assistants compress research. A buyer can learn about your category, compare vendors and decide whom to contact in one conversation, without visiting ten websites along the way. Fewer visits mean fewer tracked touches, and many AI visits arrive without a referrer. Traditional multi-touch models see less of the journey than before.
What to accept, and what to fix
Accept that some influence will never be tracked. Fix what you can: tag every ad, identify companies, ask buyers where they heard of you, and record the first AI touch on the account. Together, those give a picture that is incomplete but honest, which is better than a precise model built on missing data.
Present attribution with confidence levels
Label each figure: tracked (high confidence), self-reported (medium), estimated influence (low). Leaders can then weigh the numbers appropriately.
Start here: AI ads for B2B. More articles in Measurement and reporting.