Want more Insights from SAS? Subscribe to our Insights newsletter. Or check back often to get more insights on the topics you care about, including analytics, big data, data management, marketing, risk and fraud.
Marketing attribution: Web analytics won’t work, but predictive analytics will
By Frederic Thys, Systems Engineer, SAS Belgium
Turning leads into gold is the eternal pursuit of the marketers, like turning inexpensive metals as lead into gold was the Great Work that alchemists tried to achieve for centuries. The alchemists ultimately failed in their quest for a very simple reason: they did not base their approach on facts, but on mystical and arbitrary assumptions. As a marketer, don’t make the same mistake. Don’t assess the effectiveness of your marketing efforts based on assumptions and arbitrary models like last-touch attribution or linear models from your web analytics tool.
Marketing attribution is about giving credit where it is due and no matter how fancy the custom weighting model used in your web analytics platform, assigning credit to marketing touches based solely on presence, frequency and order of touchpoints is wrong. The frequent presence of a touchpoint in the customer journeys is not enough to ensure that this touchpoint will lead to a customer purchase. This is confusing correlation with causation.
Based on this definition, any approach that does not consider the likelihood of a given customer type to convert independently of the marketing channel will fail to accurately estimate the sales uplift of those marketing channels.
This is because marketing simply does not account for 100 percent of new sales. And marketing will account for a different proportion of the new sales depending on which customer segment you are looking at. For example, highly loyal customers with a high likelihood to purchase are less likely to be affected by marketing on any of their usual channels.
Marketing attribution is about giving credit where it is due and no matter how fancy the custom weighting model used in your web analytics platform, assigning credit to marketing touches based solely on presence, frequency and order of touchpoints is wrong.
Two steps toward better marketing attribution
Attracting a first-time customer is not the same process as loyal customers repeating sales. So, before starting an attribution modeling exercise, you should evaluate:
1. Their likelihood to purchase independently of your marketing efforts.
2. The incremental effect of your marketing, on top of this likelihood.
This means you’ll want to implement a two-step approach because for nearly all your customers, the sum of your marketing effort won’t account for 100 percent of the purchase value.
Unfortunately, web analytics won’t be able to tell you the purchase likelihood of a specific customer or customer group, or how you should segment your customers before employing marketing attribution. If you want to know that for Segment A, the marketing channels should receive only 20 percent discount, for segment B only 40 percent. For this, you need to establish customer profiles based on advanced analytics.
Why is marketing attribution crucial?
Your mission as a marketer is communicating the value of your product or service to customers for the purpose of selling or promoting it. However, in your daily work life, you may spend more effort in communicating the value of your marketing activities within your organization.
Why is that? In my experience, our internal quest mostly goes wrong because clicks, registrations, impressions (in other words, our own marketing jargon) are often meaningless outside the marketing department. Because of this self-imposed exile, they are difficult to link with the sales metrics driving the whole organization.
Make no mistake, the story about the value of your marketing that hooks the CEOs, should be: “Which channel should we best invest in and how much should we invest if we want to achieve a global 10/20/50 percent increase of our sales next month based on our sales forecast?”
This channel/metrics connection is crucial in a multitouch world, where your channels and customer interactions with them are multiplying. You can’t optimize your marketing investments properly unless you can accurately assign credit for the new sales to your channels.
What are the really hot leads within the customer journeys that you are tracking? Certainly not the ones that will convert to marketing or sales qualified leads but rather the ones that will translate into new sales – into gold. Attribution, by pointing out the touchpoints that really count when talking about incremental sales, establishes a behavioral profile and designates the right leads among the usual suspects.
A bottom-up approach
Customer journeys are comprised of the sequence of events or touchpoints between your customers and your marketing channels. Digital touchpoints are central in this view, but offline interactions like event registrations, call center interactions, touchpoints where exposure is consistently being tracked in your CRM, might be integrated.
The following visual analytics screenshot illustrates part of the most common journeys among customers and prospects for a specific timeframe.
This is the “what happens” part of your story – your plot; the customer– your protagonist. Your goal is to understand how customer behavior changes as a result of a specific customer journey or sequence of touchpoints with your brand.
That's what your attribution story is actually about. It is about predicting the likelihood of future purchases as a function of past behaviors (focusing mainly on the touchpoints). You’ll use regression techniques such as decision trees, logistic regressions that segment your customers, and isolate the ones that will convert because they’re replicating behavior that led to purchases in the past.
Step one: Establishing a baseline
The game of marketing investments is about precision. To maximize precision when measuring the sales created by your marketing efforts, you need a first step where you identify the sales uplift created by all the other factors at your disposal, excluding marketing.
For existing customers, you can use information about previous purchases to quantify their intrinsic likelihood to purchase. For prospects, you would typically use a global engagement score to quantify their intrinsic likelihood to purchase for the first time.
For example, you can create a decision tree using visual analytics to segment your customers and prospects as a function of their likelihood to buy:
On the first chart, the decision tree prioritizes the factors other than marketing that are contributing to a likelihood to purchase. We have in decreasing order of importance:
- Business segment.
- Type of product purchased.
- Premium flag.
- Total amount.
- Loyalty score (at the bottom).
On the next chart, with the leaf diagram, the decision tree distinguishes 20 segments with a different intrinsic likelihood to purchase.
Step two: Measuring the incremental effect of marketing on new sales
So for each segment (thanks to our decision tree) you now have customers’ likelihood to purchase (without marketing influence) as a baseline. You now can include the marketing channels to see the incremental effect they have on the baseline for the intrinsic likelihood to buy.
The following example shows the results of a logistic regression for the segment with the highest likelihood to purchase excluding any marketing influence. In this example, it’s a baseline of 10 percent.
The screenshot below shows that website engagement, social media and organic search contribute to a significant incremental effect on the likelihood to purchase:
- If a customer from this segment demonstrates a high website engagement, then his or her likelihood to purchase increases from 10 percent to 35 percent.
- If a customer from this segment demonstrates high website engagement and converted from social media, then his or her likelihood to purchase increases from 35 percent to 75 percent.
This is a great insight because you can now measure the incremental effect of your marketing in terms of sales uplift. The second takeaway is how more accurate you can be at identifying your most attractive leads when including marketing activities. The top 10 percent of marketing leads within this segment are 10 times more likely to convert as compared to randomly targeting the entire segment.
These are our leads transformed into gold. This is the second chart of the following screenshot.
Putting it all together
With advanced analytics applied to marketing attribution, the marketers might not have discovered the Philosopher's Stone of the alchemist, but with the ability to credit the right marketing channels for their impact on the new sales, it seems that marketers have finally discovered a way to ensure their prosperity inside their organizations.
Free white papers
Getting to Why in Omnichannel Marketing Attribution explains how advanced analytics can help you better understand customer behaviors and optimize your marketing strategies.
Marketing attribution: Giving credit where credit is due explores how marketing attribution helps you to make the best marketing investment decisions.