Future-Proof MTA Data Collection
Instrument first and last UTMs on day one so you can move to multi-touch attribution without waiting.
- Difficulty
- Moderate
- Time to result
- ~days to results
- Steps
- 4
- Confidence
- 92%
Most companies start with first-touch or last-touch attribution and only later want multi-touch (MTA), but they lack the historical data to do it. Hay's fix is to design the data collection for MTA from day one, even while using a simpler model, so the data is already there when you graduate. It requires no data science, just capturing the right parameters as both user attributes and events.
Origin
Austin Hay's recommended best practice, developed across his martech consulting and roles at Branch, mParticle, and Ramp.
Core principles
- 01Businesses evolve first-touch/last-touch to 50/50 split to MTA, and by the time they want MTA they lack the data.
- 02Third-party b2c tools only have two objects, a user and an event, so you must deliberately design what you stuff into each.
- 03Storing first and last UTM values on the user, plus firing events for the middle steps, lets a warehouse coalesce the full journey.
- 04This is not complicated to set up; most people just skip the early work.
How to run it
- 1
Capture all inbound parameters on arrival
When a user hits the site, collect the referring URL, all UTMs (campaign, medium, etc.), and every ad-network click parameter (Google Click ID / gclid, Facebook fbp/fbclid, TikTok ID, Microsoft ID). Keep a canonical list of these parameters.
Pro tip Maintain a standing list of the ad-network parameters you want to capture so nothing is missed as you add channels.
- 2
Store first and last locally
Persist the values on the device (browser storage / first-party cookies) as UTM-first-campaign and UTM-last-campaign. Every time the user returns, overwrite the 'last' values with the newest source while preserving 'first'.
- 3
Write to both the user object and events
Collect the attribution info as a user attribute AND fire a page-view event carrying first and last, so every touch in the middle is recorded on an event.
Watch out Third-party tools only give you a user object and an event object (only Snowplow lets you control schema), so you must consciously decide what goes where.
- 4
Coalesce in the warehouse when ready
Your data team coalesces over all the last-UTMs seen across the user's events to reconstruct first, middle, and last touches, enabling any attribution model on demand without a waiting period.
In the wild
A user arrives at Lenny's newsletter from a Google ad (first touch), then later returns from a Facebook ad (last touch). With first and last UTMs stored on the user and page-view events firing throughout, the warehouse can reconstruct which channel gets credit under any model.
→ The company can switch from first-touch to last-touch to full multi-touch attribution instantly, because the underlying data was captured from day one.
Common mistakes
Only collecting UTMs and calling it done
Capturing UTMs alone, without first/last handling, the middle-step events, and writing to both user and event objects, leaves you unable to build MTA later and losing the historical data you would have needed.
Waiting until you want MTA to instrument for it
Because attribution needs historical data, deciding to build MTA only when you want it forces a long wait to accumulate data; instrumenting from the beginning removes that delay entirely.
Is it for you?
Best for
Growth engineers and martech owners setting up analytics for a new website business who anticipate needing multi-touch attribution later.
Not ideal for
Mobile-app attribution under iOS 14 ATT, where deterministic one-to-one tracking is largely unavailable and probabilistic methods apply instead.
From the transcript
“most businesses either start with first touch or last touch and then eventually they want to move to a multi-touch attribution model”
“you've got to store those parameters locally on the device if there's a browser or whatever you've got to store it as UTM first campaign…”
“you're going to collect it both as a user attribute and as an event that way what's going to happen on the back end for…”
“this isn't actually that complicated to set up most people just like don't do the work early on and then when they want to go…”
From the episode
The ultimate guide to Martech
Austin Hay (Reforge, Ramp, Runway)