PostHog Attribution: How It Works and Where It Stops

Everything below comes from PostHog's own documentation, read in September 2026. PostHog ships quickly and the marketing analytics product is still a beta, so check the docs for anything that has moved since.

PostHog captures campaign parameters on every event: the UTM set, the click IDs and the referrer. What turns them into attribution is where they end up on the person, and there are two kinds of property with different rules.

First-touch attribution in PostHog is therefore a breakdown: pick your signup or purchase event, break it down by Initial UTM Source or Initial Referring Domain, and read the bars. Last touch is the same breakdown on the latest properties. PostHog's own tutorial on first and last touch attribution works exactly this way.

Marketing analytics is the newest piece and the closest PostHog comes to an attribution product. It is an opt-in beta you switch on under feature previews.

• Spend from eight platforms: Google Ads, LinkedIn Ads, Meta Ads, Pinterest Ads, TikTok Ads, Reddit Ads, Bing Ads and Snapchat Ads, with clicks, impressions and cost, converted to your base currency.
• Matching on UTMs: campaigns are tied to traffic through utm_campaign and utm_source, by campaign name or ID, with manual mappings for the ones that do not line up.
• Conversion goals are PostHog events, actions or data warehouse tables. A warehouse table needs a UTM campaign column, a UTM source column and a timestamp.

That puts cost per conversion per campaign on one screen, which is genuinely useful. Two things follow from the design. The match is only as good as your UTM discipline, because a campaign that arrives with a different utm_campaign than the ad platform's name is a campaign with spend and no conversions. And the conversion is whatever you define as a goal, so a signup is easy while a closed deal is only as good as the warehouse table you build for it.

PostHog documents these itself on its campaign attribution troubleshooting page. None of them is a bug; they follow from attribution living on a browser-side identity that has to be joined to a person.

• Anonymous visitors never reach a person. With the default person_profiles: 'identified_only', events from anonymous users do not update person profiles, so the campaign on those events never lands anywhere until the user is identified.
• identify() placed late. If the user is identified after the signup form rather than on it, the conversion happens before the person exists and the latest campaign is lost.
• Safari and in-app browsers. Safari clears cookies and localStorage aggressively, and the LinkedIn, Instagram and Facebook in-app browsers keep isolated storage, so the first-touch cookie is often gone by the second visit.
• A second device. Cross-device journeys are only joined when you call identify() on both, which a marketing visitor rarely gives you before they convert.
• Ad blockers. PostHog says a reverse proxy typically increases event capture by 10 to 30 percent, which is also a measure of what an unproxied setup is missing. The managed proxy is free on PostHog Cloud and worth switching on either way.

The practical effect is that PostHog's first-touch numbers are most reliable for users who converted in one sitting on one device, and least reliable for exactly the long, multi-visit journeys a B2B or high-ticket buyer takes.

PostHog's data warehouse syncs managed sources, including HubSpot, Salesforce, Pipedrive, Attio, Close, Stripe and Shopify, every few minutes, and lets you query them with SQL next to your events. That is how a closed deal gets into PostHog: as rows in a table, joined to people by something both sides share.

SELECT
  p.properties.$initial_utm_source   AS first_touch_source,
  p.properties.$initial_utm_campaign AS first_touch_campaign,
  count()                            AS deals_won,
  sum(d.amount)                      AS revenue
FROM hubspot_deals AS d
JOIN hubspot_contacts AS c ON c.id = d.contact_id
JOIN persons AS p ON p.properties.email = c.email
WHERE d.dealstage = 'closedwon'
GROUP BY first_touch_source, first_touch_campaign
ORDER BY revenue DESC

It runs, and it gives you first-touch revenue by campaign. It also quietly assumes five things worth knowing before a number from it reaches a board slide:

1. One contact per deal. In HubSpot a deal and its contacts are linked through associations, not a column on the deal, so the first join above is already a simplification you have to replace.
2. The email matches. The person in PostHog was identified with the same email the CRM holds, which breaks on a work address against a personal one.
3. The person was identified at all. Every deal from a lead who never logged in or filled a tracked form drops out of the join, silently.
4. First touch is the model. Any other split needs the ordered touch history, not the person property.
5. Spend is somewhere else. The query has revenue but no cost, so return on spend is another join, against the marketing analytics tables.

None of this is hard for a data team. It is a small project with a maintenance tail, and it is the part that tends to become one person's spreadsheet.

Attribution tells you where to move the budget. Sending the conversion back tells the ad platform's bidding who to find more of. PostHog does this with realtime destinations.

• Meta Ads Conversions sends hashed email, name and the Meta click and browser identifiers with the events you filter for.
• LinkedIn Ads Conversions sends the conversion with an optional value and currency.
• Google Ads uploads offline conversions matched on gclid, and PostHog labels it an experimental destination it does not officially support. Google itself needs 6 to 48 hours to process them.

• Attribution next to behaviour. The same person carries their first campaign, every product event, their session replays and the flags they saw, so you can ask which channel brings users who activate, not only users who sign up.
• Product-led conversions. When the conversion is a signup, a trial start or an in-app upgrade, first and last touch on a PostHog event is exactly the right place to measure it.
• Spend in the same tool. The marketing analytics beta removes the spreadsheet for cost per signup per campaign across eight ad platforms.
• Price. A million events a month are free, and there are no seat charges, so the whole team can look.
• Everything queryable. When the built-in view is not enough, the SQL editor and the warehouse mean you are never locked out of your own data.

The pattern is the one you would expect from a product platform. Wherever the conversion lives in the product, PostHog is excellent. Wherever it lives in the CRM or on the phone, PostHog can hold the data but the attribution becomes something your team builds.

Teams in that second situation rarely replace PostHog, and they should not: the product team keeps its analytics, replays and experiments. What they add is an attribution layer owned by marketing, which tracks the website server-side, keeps every touch rather than the first and the last, reads deal stages and values from the CRM natively, and sends the closed deal back to the ad platforms without anyone writing an event for it.

That is what LeadJourney does, and the comparison page sets it against PostHog row by row, including the rows where PostHog is ahead.

• PostHog vs LeadJourney: the full comparison, fifteen rows and both price lists.
• Multi-touch attribution: what the models do with the touches first and last touch leave out.
• Offline conversion tracking: sending the closed deal back to the ad platforms.
• BigQuery marketing attribution: the same warehouse problem, worked through on the GA4 export.
• HubSpot Marketing Hub attribution explained: what the CRM side can credit on its own.