Marketing Attribution is one of those absurdly challenging problems that impacts many levels of a company. There are finance teams that will use it to set budgets, marketers who allocate those budgets, and analytical teams measuring the data and how effective the marketing spend is. Selecting a proper attribution methodology can be an intimidating project, but it’s worth the legwork and research to ensure you get it right.
Put simply, attribution, when done right (or done better than you are doing it now), allows you to more effectively allocate your marketing budget and provides much greater visibility across your marketing and testing activities. You can trim spending on channels that aren’t driving incremental sales, while reinvesting that money saved in scaling channels and campaigns that are more efficient.
Having worked on a variety of attribution problems for different companies, there are 5 key things to consider prior to getting started.
1. Just say no to last click or first click attribution.
2. Accept the fact that no attribution model is going to be perfect. There are always going to be tradeoffs.
3. What is your business model?
4. Who are your customers? What segments of customers are driving the most $?
5. What is the customer journey / sales cycle like?
I’ll address these five points in order below:
1. Just say “NO” to last click or first click attribution.
Friends don’t let friends use last click as their attribution methodology. Last click attribution isn’t just a flawed way for measuring sales, leaning on it too heavily can be downright bad for your business. The key reason last click is such a poor method is that it’s winner take all; it awards one hundred percent of credit for a sale to the channel that drove the last touchpoint prior to purchase. This completely ignores the impact that other channels may have made in the customer journey. Furthermore, there are a whole slew of vendors and affiliates out there that exist solely to interject themselves between you and the customer, just prior to purchase, where they can claim credit for a sale that was going to take place anyway.
Last click tends to under value channels that play a big role in customer discovery and awareness (such as email), and over values “closing” channels (Direct, Affiliates). If you are still running last click attribution, you could do worse by arbitrarily picking one of the more straightforward attribution methods (such as position based or time decay) and just switching to that. It won’t be perfect (and no attribution method is) but it will almost certainly be an improvement from last click. Which brings me to point number two….
2. Accept the fact that no attribution model is going to be perfect. There are always going to be tradeoffs.
A lot of attribution projects fail because businesses become obsessed with finding a “magic pill” for attribution that will give them crystal clear insights and an intelligent way to allocate sales credit that perfectly aligns to their business model. I hate to break it to you, but a perfect attribution model (mostly) does not exist. There are simply too many variables involved and too many nuances to different businesses that nearly no solution will be perfect right out of the box. Some brick and mortar businesses have challenges collecting data about their customers, some businesses have really high ASPs and a much longer sales cycle, some deterministic click stream based solutions work well for digital but not much else. If you are doing TV advertising, how are you accounting for the impact of the broadcast? The list goes on and on.
The good news is you don’t need to have a perfect attribution model for it to be effective. A well thought out, logical attribution solution that is customized to your business model can still add tremendous value. It’s helpful to make a list of your “must haves” for an attribution solution. For example, it must be able to estimate a customer’s propensity to purchase (incrementality), it must account for the impact of TV broadcast, and it must provide data down to the campaign level, not just the channel level.
3. What is your business model?
This seems obvious, but it’s important to consider your business model and make sure that your attribution method works well with it. For example, a daily deals site with a $40 ASP that sends out a huge email blast to everyone at noon everyday is fundamentally different from a retailer selling high end jewelry with a $500 ASP. A pure e-commerce business is much different from a brick and mortar retailer with an e-commerce presence.
So before you just blindly stick an attribution algorithm on top of a stack of data and call it a day, make sure you understand what makes your business model different. Businesses with a longer sales cycle or higher ASP may want to prioritize customer discovery, as frequently your products will be more of a considered purchase as opposed to someone who bought a $25 T Shirt from a promotional email. If your business model is especially unique, consider how any of the common attribution methods may or may not fit your business model.
4. Who are your customers? What segments of customers are driving the most $?
What percent of your revenue comes from your “best” customers? For many (but not all businesses), you see the classic 80/20 pattern – the bulk of your sales are being driven by the top 10-20% of your most engaged customers. Consider how these customers (and by extension, the marketing channels they use most frequently) will impact an attribution solution. For example, your best customers are the ones who order the most. Therefore, those customers are receiving the most transactional emails (since they are doing the most ordering). Since they are your best customers, they are more likely to open/click emails, and also less likely to opt out or unsubscribe. Your best customers are also the ones most likely to be visiting your site the most frequently. The end result here is that channels that are weighted disproportionately towards your more “best” customers (email in our example) will be weighted more heavily.
Depending on what your objectives are, this may or may not be desirable. For example, just how much are these touchpoints driving base vs. incremental buying activity? That’s a separate question, and one best answered via testing and holdouts but the key takeaway is that understanding who your customers are and which ones are driving the most sales and traffic is vital. If you aren’t already aware of the ins and outs of your customer file, you definitely will want to be prior to rolling out an attribution methodology.
5. What is your customer journey / sales cycle like?
This is similar to #4, but the typical customer journey, especially via clickstream analysis or mapping online to offline commerce, often gets overlooked when considering an attribution methodology. You can start to dive in here by pulling the data around a few important questions / metrics (if you haven’t done so already)
Addressing the questions above will start to fill in the picture on what the customer journey and a typical sales cycle is like for your business. This is important for a few reasons. First, if the typical new customer takes 5 or more sessions from first visit to conversion, you know for sure that an attribution method like last click will miss out on a TON of engagements in the customer’s journey to purchase. Conversely, if the typical new customer converts very rapidly, say within one or two sessions from their first visit, this may not be as big of a deal (but you still should not use Last Click :)). Second, getting a sense of the amount of time customers take leading to conversion can be helpful for establishing reasonable parameters for an attribution model. For example, if you are using some sort of time decay model you’ll likely want to set a “lookback window”, or how far back prior to purchase you want to capture customer touchpoints. If the typical customer takes 30 days to conversion and you set the lookback window at 3 days, you’re missing a lot of the touchpoints along the customer’s path to conversion.
“Day Count” is an interesting measure because it can provide a bit more insight around visit “loyalty” than simply counting sessions per cookie/customer. For example, measuring the number of unique days a customer visits the site vs. the number of sessions a customer has will give you a better sense of visit “loyalty”. A customer who has 10 sessions in one day is likely different, from a loyalty profile, than a customer that visits once a day for 10 consecutive days. The former may have just found some product(s) you happened to be promoting that day that interested them; the latter is a customer returning each day to check out what you have to offer. This opens up some further segmentation possibilities and questions you can ask your analytical team. How “loyal” are our best/worst/new customers, from a visit persepective? How many visit us at least X days a month? What marketing channels are they using?
If you’ve been thinking about changing your attribution methodology, or just taking a closer look at how your existing model fits, it can seem intimidating at first (because it is). There’s typically a lot of work involved, and the potential impact to the business and its marketing investments is huge. Two important resources are the data around your business, and a technical/analytical team that can pull, clean, and summarize it to fit an attribution method with the realities of your business. The questions and points I listed above are a good place to start, but they are not by any means the ending point. Attribution is a tough problem, but if you can get it right the juice is worth the squeeze.