Behavioral Insider: How does transactional data stack up against other behaviors and data points in predicting future behavior?
Mark Klein: We talk about the continuum of customer data. We classify them on a scale from passive to active. Demographics is very passive data, psychographics is a little more active, then you get to browser clicks, which are more active still, and then you get to transaction data. We have found, with a lot of data to support this, that the most accurate predictions come from transaction data. We do use Web analytics, but we say around here that customers vote with their wallets, not their zip code. We are very accurate. We analyze all kinds of transactions for Microsoft, people buying everything from Xbox to SQL Servers, and we can predict what a customer is going to buy next.
BI: In a down economy, your company and approach is seeing more business. Why?
Klein: Companies are looking for more revenue from their existing customer set. People's marketing ratios have been way out of whack. We look at a lot of client data, and invariably the ratio of revenue from existing clients to new customers can be on an order of ten to one. Yet, most of the spend is directed at new customers. And we think that balance is starting to shift a bit.
BI: Any retailer knows to remarket to their own customers -- but what additional insight is your analytics bringing to it?
Klein: They know who their customers are. I see signs at trade show booths claiming they will help clients know their best customers. If you don't know who your best customers are, who are the easiest ones to spot, you are in deep trouble. The best and the worst are the easiest, but you have little insight into that broad middle spectrum. Fundamentally, we collect four pieces of data. It is a customer ID, a transaction date, a dollar amount, and some SKU descriptor. From that we are able to get to our loyalty model.
We calculate a loyalty score for everyone. And that model looks at a lot of the things you can derive from those pieces of data. We look at lot at changes. We look at derivatives rather than value. We look at change in recency, for example. We say who is going to buy next? What specific products are they going to buy? Our clients download regularly a list of customers that meet certain criteria and they will typically get the top three or top five products the customer is likely to buy in the near future.
BI: What sorts of things can clue you off to a cross-sell opportunity? It seems that part of this process must involve knowing more about the customer's other profile data.
Klein: We look at higher order affinities. We look for patterns of behaviors, so we can say, if you bought so many products from a set we created, you are likely to buy another from a bundle we have created.
For example, one of our clients uses postcards to market to their customers with four images of the four products that the customer is likely to buy next. Each of [them] will get different sets. And they can use Longbow to push a button to have those postcards in customer hands in a week. Targeted promotions are one of the best ways to improve marketing response.
BI: How do you calculate when customers are at risk of leaving? Are there telltale signs?
Klein: We do it for Microsoft, Verisign, and Second Life. We look at the transactions occurring in Second Life and see which customers are not likely to keep spending money with them every month. It is not that there are telltale signs. In analyzing data you learn that the derivatives of values are often more telling of what is going on. So we look for changes in behavior. Those changes pop up very easily. So we can put up a scatter graph of customers, one dot for every customer. And the ones in danger show right up in the region of the chart, like reading an MRI.
BI: One of your blog posts referenced the issue of demographics. Would you place transactional data and this predictive modeling in relationship to the other types of targeting in terms of strengths and weaknesses?
Klein: Transactional data is obviously used more with existing customers and intent based click stuff is more on the customer acquisition side. On the other hand, we also use all of that data. One of the things on my desk right now is an engagement metric we have been calculating for a couple of our clients. One of things that goes into it are the Web clicks. This client sells medical supplies, and when they find one of their customers is under-engaged, compared to others they will typically direct one of their field staff to find out why and see if they can raise their engagement.
When we dig into some esoteric challenges we use some of that other data. We work for a large B2C company selling diabetes supplies to over a million customers. We help them identify customers for products they are just bringing to market, so we look at some aspects of patient data like which of them is likely to have high blood pressure or heart disease and things like that.
BI: Like a lot of data-driven service companies, you have your hands on data from competitors. What are the policies for segregating intelligence but also sharing the learnings?
Klein: We never let one client's data cross over to another client's even though in some market areas we have several clients, as in technology. We work for CDW and Dell and others. We can't let what we know from one creep into another.
We work on pay-per-click and customer acquisitions, but it is hard for us to understand what are normal click-through rates, for example. So what we are doing is aggregating data from about 20 of our client set and will start publishing aggregate statistics like the average ratio of existing customer revenue to new customer revenue or what are typical response rates for upsell and cross-sell and win-backs.
People quote response rates, but unless you tell me what kind of campaign that is I can't take it very seriously. We've actually had clients cooperate with one another. CDW sells a lot of Microsoft stuff. And we know everything that CDW sells from Cisco and HP, etc. CDW would have fits if we told Microsoft that kind of information., But they don't have trouble taking the CDW and Microsoft sales and talking to Microsoft about that. So they got together and do a promotion that puts Microsoft product they want to push on specific kinds of hardware, so we identified people who would buy those bundles by using both company's data and not letting one company know what they other company knows.