When people hear attribution, they immediately think about channel-level attribution. And while that's often the most important and drives the largest budget decisions, there's another type of attribution that can derail your growth.

Geographic attribution.

It’s the lesser-known and often ignored sibling.

Below I'll cover:

Geographic attribution operates under the same principles as channel attribution, but instead of where do you assign “credit” to a channel, you need to assign “credit” to a customer’s geography.

It sounds like a straightforward concept, but as we know with any attribution, it isn’t. And the more complex the business is, the more complex Geographic attribution becomes.

In addition, most people only think about attribution for acquisition but it’s also an important concept for remarketing.

Example 1 (Acquisitoin)

You have a business where the conversion window is quite long (common in Travel Tech). You run ads in the US but the person takes your service in Europe.

Your main ads run in the US.

Conversion happen in Europe.

Is that customer “American” or “European?”

Example 2 (Acquisitoin)

You're running a marketplace business and have a structure in which General Managers (GMs) run individual regions.

You’re the GM for New York and you see ad spend in your market but the use case has most users using the product in LA. The New York GM sees that Marketing hurts their P&L and says “Fuck that…I’m not spending here”. They turn it off. The LA GM now sees their market deteriorate too and the whole business is slowing down.

How do you ensure fair distribution of PNL ownership here?

Example 3 (Remarketing)

You’re the Head of CRM at a company and you send localized comms. Now a person signed up 3 years ago in New York but they’ve now moved to LA. Unfortunately, your Marketing data says they’re still in New York and you keep blasting them with New York comms and they unsubscribe. Womp womp.

What do you do?

In all 3 examples, the wrong Geographic attribution would lead you to the wrong marketing decision which ultimately would hurt growth.

This is exactly what Geographic attribution tries to solve:

Once you solve these, many challenges fall into line .

There are 2 times when you need to understand Geographic attribution:

At Uber, how we handled this was quite straightforward: the city that the first trip happened in got the credit for that trip.

What's more important is the context and the reasons for why we were able to do this.

First, from a customer acquisition perspective, the window between intent and conversion was short. Most people were signing up and booking within the first day, and almost 80% within the first week.

Second, you had a use case that was primarily domestic where users used it in their home city for work or social travel. Of course, there are other use cases like to/from airport but the frequency of those is less.

We knew that when we served an ad in that market more often than not the customer would take the Uber in that market. We didn't have to worry too much "what if they see me in one country or city and then go and take another."

That of course did happen, but this is where knowing your customers / product behavior is important. If a user saw an ad in New York then took their first trip in LA then there’s a decent population of the reverse happening too where they saw an ad in LA then took the trip in New York. Effectively these “edge” cases cancel each other out.

Given the above, this meant that we were comfortable assigning credit to the city where the first trip happened.

This aligned Marketing spend with Marketplace operations where the GMs felt comfortable with the spend / credit relationship.

Now, what happens to that user as they move forward in time?

Imagine how frustrating it might be for you to get a communication when you live in New York after you move from San Francisco.

How do the teams know where you now live? And how do they update it?

Geographic attribution is hugely important, and that's how we approached Uber was to get smarter and smarter with more data we have. We wouldn't nail every use case, but you're not aiming for a hundred percent perfection; you're aiming for 80.

Uber followed a simple waterfall (and eventually got smarter with machine learning). We knew:

For each user then we could create a column in our Marketing database that identified their “Marketing” city. Naturally, we’d also be able to translate this column into Marketing Country, Region, and Mega Region. This was our best approximation of where they lived and thus when we should target them with geosensitive comms.

When a user signed up → Marketing city = signup city

When a user took a trip → Marketing city = first trip city

Their past 5 trips → Marketing city = most common city

Sometimes, a user wouldn’t take many trips but would session a lot and so we could replace trip city with session city.

Over time, we could get smarter by looking at credit card address and billing zip to really triangulate where our user was from in addition to other patterns like excluding certain airport trips or other behaviors that represented non standard behavior.

So, how should you think about all of this?

Below is a framework for how you should think about Geographic attribution.

The simples way to think about this framework boils down to my all time favorite philosophical concept: Occam’s razor.

So, using that as a guiding principle, here’s how to use the framework.

Step 1: What does your growth model look like?

If your growth model does not require geo specificity then do not introduce it and you likely don’t need Geo Attribution. You can still capture this information but it likely won’t have a huge impact on your planning / ad strategy.

This often happens in early stage startups that have one geo or products that don’t necessarily involve needing to know where the user is from. For example, Facebook doesn’t really care what geo you’re from.

If your growth model requires geo specific, proceed to step 2.

Step 2: How is your business organized?

Remember, Geographic attribution is about assigning “credit” not actually understanding where a user comes from.

Here’s what I mean. If you’re an Ecomm store that’s available all around the US then knowing a person came from city X isn’t that important.

If you’re an Ecomm store that has two different offerings in the US for some reason then knowing where a person came from is important. It’ll help with marketing and operational reasons.

Your geographic attribution should match the level of complexity of your business. If you’re running your business at a level that creates multiple geographies then you will need geographic attribution. Proceed to step 3.

Step 3: What channels do you use + what is the customer journey and inter-purchase period??

This is where it starts to get complicated so explaining this succinctly is going to be a challenge but here I go.

The goal with geographic attribution is to find the path of least resistance that allows you to make smarter marketing decisions.

If your inter-purchase period is short then in all cases use transaction demographic to assign Geographic attribution to the user. Where they buy is where they’re “from”.

Now with longer inter purchase periods (think 1month+), it gets challenging. If you’re primarily on Organic Social then there’s no real way for a user’s geographic to help you make smarter marketing decisions. It also doesn’t help with planning too much. So here , you try to identify a user by their user demographic (think credit card, phone #, etc.).

If you’re running paid channels with a long interpurchase period, congrats you’re playing on hard mode!

You need to identify the user demographic AND the transaction demographic.

User = transaction demographic?

Then the geographic attribution can be the same and you can feel confident your ads are being deployed in the right place.

User ≠ transaction demographic?

You need to adjust how your Marketing spend is calculated and how it shows up on PNL. Specifically, you’ll want to give transaction credit to the place where it happens BUT you need to adjust the marketing PNL by also assigning that Marketing spend to the same transaction location (and thereby adjusting from the original location).

Example

You run $100 of Ads in the US. 10 customers globally are from the US. 5 Customers buy in the US and 5 in Europe.

US gets credit for 5 transactions but all of the spend is in the US. So, you adjust it and say “Hey Europe, even thought they saw the ad in the US (or so we think), they converted with you so you’re going to have to take the Marketing spend”.

As long as you’re consistent with the reverse (Europe ad spend, US conversion) then from a PNL perspective it should be okay and everyone will be happy. The hardest part is of course building a system that can do this in real time and is consistent.

If you’ve got questions about how to do this at your company, reach out to me! I’ve talked to 3 teams in the past 5 weeks about just this.

This one is a bit more straightforward but you have to find the most representative information for a customer and assume that’s their location.

Here are some data points that can help you:

You have to build a triangulation across this and accept that some people will slip through the logic and then just continue to get smarter. Honestly, there isn’t more to it then that so don’t complicate things.