Why you'll never find a perfect attribution model

In partnership with

Attribution modeling has always been a hot-button marketing issue. Simply put, attribution is your ability to credit your marketing initiatives with success or failure in driving key outcomes.

But because different attribution models assign credit to touchpoints in the consumer journey in different ways, each one — even the most complex and expensive to implement — will give you a different version of the “truth”.

Why the concept of attribution matters

The more you understand the success of your marketing initiatives, the better you can understand:

• What your customer path to conversion looks like

• How to allocate resources across marketing initiatives

The more insight you have into your customer journey and how the dollars you’re spending are driving your desired outcome, the better and more effective your marketing strategy can be. Simple, right? 😉

Of course, it’s never quite this straightforward. Think about your typical decision making process and journey to making a purchase, and how that compares with each of the different attribution models below.

Different attribution models, defined

At a high level, there are two main buckets an attribution model can fall into: single-touch attribution and multi-touch attribution (MTA).

Each have their merits, and some models are better than others depending on the complexity of your customer journey.

For example, a single touch attribution model may work just fine for an ecom product that’s made as an impulse purchase, whereas a multi-touch attribution model may be more appropriate for a purchase that has a longer sales cycle, requires more research, or has an omni-channel component.

Single touch attribution models

The basis for single-touch attribution models is that there’s a single touchpoint in the user journey that’s the most influential in driving a conversion.

• First touch attribution: The customer’s first touchpoint gets credit for driving the conversion

• Last touch attribution: The customer’s last touchpoint before a conversion occurs gets credit for driving that conversion. This is probably the most commonly used, lowest-lift attribution model, particularly if you use UTM parameters for ad and link tracking

Honorable mention:

• “How did you hear about us” or HDYHAU: Not so much an attribution model as an additional datapoint, this is a question asked immediately post-conversion with preset options for the user to choose from. A few notes:

  • Since this data is self-reported by the user, it’s sometimes undeservedly seen as “less true” than UTM-based tracking. But it’s actually pretty trustworthy — if you’re skeptical, add some dummy options (like TV, if you’re not advertising on TV) and you’ll see how virtually no one thinks they heard about you this way.

  • It’s precisely because this information is user-reported that it’s so valuable in helping you understand what initiatives are memorable to your users

Multi-touch attribution models

The basis for multi-touch attribution models is that there are multiple touchpoints in the user journey that influence a conversion.

• Linear attribution: Equal credit for the conversion is given to all touchpoints. E.g., if you have 4 touchpoints, each is responsible for 25% of a conversion

• Position-based attribution: More credit is given to specific touchpoints at certain parts of the user journey, depending which are most important

• Time-decay attribution: More credit is given to touchpoints the closer they are to a conversion

• Data-driven attribution: Credit is assigned to the most influential touchpoints, according to machine learning

  • These will typically be touted as the “best”, especially with all the advancements in ML made available to marketers in the last year or so, but the downside is they’re a total black box

A note on marketing mix models (or media mix models)

Marketing mix modeling / media mix modeling / MMM, as I’ll refer to it from here, is more of a statistical model than an attribution model, but it’s still worth mentioning here.

MMM takes into account all potential factors that lead to a conversion — not just touchpoints in the user journey — and assigns relative credit to them.

Funnel has a great explainer video on MMM that includes a helpful sports analogy:

Think about trying to assign credit to individual players in a soccer game based on the outcome of that game. Obviously the player who scores a goal doesn’t get 100% credit for that goal, or for the win. This calculus is what an MMM tries to solve for, taking into account as many factors as possible to appropriately credit them for the “win”.

MMMs can be built by in-house data scientists, specialized third party agencies, or leveraged through external SaaS platforms.

Note that while MMM is an option within the total universe of attribution possibilities, it’s not a practical option for a small startup. For sake of comparison, The Knot recently published a great article about building an MMM practice, and their annual revenue is around $400m. You need a LOT of data, and money, to build an MMM, and I bring it up to demonstrate how deep and complex you can go on attempting (operative word) to understand attribution if you wanted to.

Taking it back to reality

This past week I saw an influencer ad for a pair of pants I decided I absolutely must have. I clicked on the ad, and immediately determined, based on the gimmicky site (compare-to pricing! countdown timers! spin to win!) and lack of fabric composition info on the product page, that this particular pair of pants was going to be low-quality and I would look elsewhere.

I was then absolutely bombarded with ads from competing brands, quickly realized this was all the same pair of dropshipped pants at varying low price points, and resolved to look for something similar from other brands I trust. I narrowed it down to a few pairs and ultimately ordered from a brand I’ve purchased from before.

In the interim, the brand I purchased from had sent me a few emails, and I’d been to their site directly a couple of times to browse for the pants I was looking for. But those touchpoints — the emails and my direct visits to the site — were certainly NOT the driving forces behind my purchase. See where this gets complicated?

Maybe I’m more neurotic than most in my purchasing journey, but the fact is there are MANY factors that influence a user’s decision to convert, and, most importantly, those factors don’t happen in a trackable vacuum.

No single attribution model can take into account every external factor, particularly those that happen inside your customers’ brains! MMMs, as they’re custom built, can do a better job of incorporating the macro, but you have to have LOTS of data, and $, to make building and employing an MMM worthwhile.

Know your size

If you’re not doing at least $10m in annual revenue, I wouldn’t recommend spending any of your budget on advanced attribution modeling, and certainly not on building an MMM.

You should be able to get 95% of the way there with a combination of data sources, like your ad channels, your source of truth database, UTM tracking, and post purchase surveys. The exact combination is where the art and the science comes in.

Marketing to humans

While it’s always worthwhile to understand the world of possibilities when it comes to attribution, and have something to aspire to ($450m in annual revenue? Yes please), it’s also important not to lose sight of the fundamentals.

Reminding ourselves that we’re marketing to our fellow humans is a theme I always like to come back to.

Some attribution models are going to be better than others depending on the nature of your business, but none alone will ever provide a complete version of the truth. That’s because the humans on the other side of those models are making countless decisions and consuming tons of information on a daily basis. Think back to my pants example, and for such a simple purchase!

Supplementing your quantitative attribution data with qualitative customer insights will always paint a richer picture than even the most advanced attribution model can.

My tactical advice

To do this, the approach I typically take with small startups (under $10m annual revenue) combines a few different methods to triangulate a plausible version of the truth. We get a clearer picture of performance this way, when we know in our heart of hearts that last-touch alone isn’t enough.

This method is basically free, and gets you 95% of the way there, even if you’ve started to incorporate less trackable marketing, like audio / video / out of home: