Is multi-touch attribution good for anything at all?
No. Multi-touch attribution platforms record all of the deterministic digital touchpoints they can and then divide credit for any tracked sales among all of the touchpoints that reached that customer before converting. Feels pretty sophisticated right? To be fair, the tech itself is sophisticated, but for marketing purposes, it's snake oil. Stay with me.
Hot take: platform attribution is more valuable to your business than MTA.
Tracked Marketing Touchpoints Are Ridiculously Flawed
First, the vast majority of touchpoints recorded in MTA models are from clicks on ads. Clicks are worthless because the most effective advertising on the planet derives all of its value from its ability to tell a story and evoke an emotion.
Clicks don't matter and there's no reason to treat them like they do. In fact, focusing on clicks is actually harmful. They severely decrease your reach (an actually helpful metric) and multiple studies have found no correlation between clicks and sales (Comscore, Nielsen).
Plus bots click at a higher rate than humans and algos optimize for them.
So, multi-touch attribution takes all of these clicks - as long as Safari's Intelligent Tracking Prevention didn't truncate the click ID or the user did have an ad blocker running or there was no ID handoff or cross-device issue - look at all the situations where the user that clicked on that ad converts at some point in the future.
Then, it takes all of the touchpoints that it was able to track against each conversion and divides credit based on some proprietary model to give you a situation where multiple channels that contributed to each sale can share the credit.
Sounds ok, right?
But, what about the fact that branded paid search gets tons of clicks and shows up in tons of conversion journeys? On a good day, it's maybe 10% incremental, but it gets credit nonetheless.
If display, search, and social all touch a sale, what's the difference if they each show 1 sale (platform attribution) or if they each show ⅓ of a sale. It's an order of magnitude shift, but not a new insight.
Plus, at least in-platform, you can see some of the conversions the campaign touched on a view-through. I've run several analyses across accounts on Meta and often saw around 60% of sales paths had no click.
Some people could argue it's not giving ⅓ credit per channel, that it's dividing credit in a smarter way than that. Fair point, but not valuable here because even if the model is better at crediting valuable tactics, it's still handing out 100% of credit for each sale, split up between the touchpoints that contributed, incremental or not. It never asks "what would have happened without the media?"
Why is it the opposite of incrementality?
So we've got this technology doing some voodoo stuff based on clicks from people and bots to split credit in a way that is marginally worse than platform reporting. To many people, it feels business-oriented. It feels sophisticated. It's not.
They are moving toward "true value," but they're not because they're missing most of the high-value non-click exposures and giving all the credit to low-incrementality click-based tactics.
It's almost the definition of snake oil, but unlike a placebo, it doesn't actually help 30% of the time.
What You Need Instead of MTA
The framework, at a glance:
- Platform outcomes: directional signal, not gospel
- Total sales: your actual source of truth
- Experiments: cheap, in-platform incrementality reads
- MMM: the real system of record once you're spending enough
Platform Outcomes
Knowing that a channel is reaching someone on their path to conversion is valuable. Platform data is the only thing the all-powerful AI platform algos can optimize for, so you better get a hold of them anyway. If you can't afford incrementality tools, use some common sense. Don't buy your branded paid search on keywords where there's no competition. There are tools for this. Look up some proxies for incrementality by tactic and apply those to help you get closer.
Platforms double-count, meaning more than one platform will take credit for the same sale. Do not use the sum of the platform data as your final delivery number - proceed to the next step instead.
Total Sales
You need to know your total outcomes. Total overall, not media-attributed. GA is fine for total DTC website sales/leads. Even better if you can combine that with Amazon and wholesale sellthrough. Total sales tells you how the business is doing and gives us something to model our platform spend against.
Day to day, I'd rather run on total sales plus platform metrics as a directional read. The next step isn't dividing credit better. The next step is incrementality. That's testing and MMM, not MTA.
Experiments
There are several useful experiments that you can run inside platforms and there are reasonably priced platforms that you can use to run experiments. You could probably even get an LLM to run a fairly accurate causal impact analysis for you (sorry data science friends).
MMM
This is a massive cornerstone of an incrementality-focused measurement system. If you're spending $3M+ in media annually across all paid channels, you should probably have an MMM.
The math is pretty simple. $3M in media is driving anywhere from $0 to $4 in incremental revenue per dollar. Probably on the lower end if you're not doing incrementality, so let's go with $1. That means your $3M is driving $3M in incremental revenue. MMM should drive 10%-25% of incremental revenue in Y1 ($300K-$750K). It costs $150K, so you're at least doubling your rate of return and probably closer to 4x-10x if you pull spend out of a lower-incrementality channel and end up a little higher in the MMM return scale.
It's negligent to not do it. Check out this MMM ROI calculator that I made and see what you're missing.
But Aren't Platform Conversions Mostly Click-Based?
Yes, but at least you get some view-through, it's free, you can upload 1P conversion data to round it out, and businesses already have a healthy skepticism about platform metrics vs misguided faith in a marginally worse MTA model.
What About MTA Tools That Are Adding an MMM or Incrementality Experiments?
Not far enough. Experiments and MMM together are so much better that MTA actually drags them down. I'm constantly asking myself where those click-based touchpoints are corrupting my data. Drop any and all MTA backbone and we can talk.
I've been harsh here because I'm hammering keys and letting my ADHD roam free. If you are feeling hurt, I'm sorry. I respect the people who built these platforms. I just do not think multi-touch attribution has a place today.
FAQ
Can we keep MTA because finance wants one attribution story?
It's basically the equivalent of lighting your entire ad budget on fire. If you can live with that, more power to you. If not pull together your platform data, bump it up against your overall outcome trend, layer in context about your category performance and share the story with them. If you want to work this out, get in touch with me. I can explain better over the phone.
Is GA enough?
Useful for total website outcomes. Not useful as the system that decides how much credit each touchpoint deserves.
What replaces MTA?
There is no silver bullet source of truth. It's about having a few data points that help you paint an accurate picture of what is happening. MMM is the best foundational piece to understand the incremental contribution of marketing overall. Incrementality experiments help you understand incrementality at a more granular level. Platform metrics help you understand if each tactic is doing better or worse at achieving the goal you've set out for it and serves as your real-time optimization lever. Total business outcomes (e.g. total sales in GA, Amazon, etc.) acts as your barometer for whether the overall business is moving in the right direction.
What's the difference between MTA and MMM?
MTA divides credit for a sale among whatever touchpoints it could track, based mostly on clicks. MMM works backward from total sales and total spend to estimate what actually moved the business and by how much each thing was actually responsible.
I don't spend $3M a year. What should I do instead of MTA?
Run on total sales plus platform metrics as a directional read, use in-platform experiments where they're available, and apply incrementality proxies by tactic. Explore incrementality testing platforms (prices come down every day).