Sramana Mitra: Let’s double click down into this particular use case. What happens? You take the customer database of this logistics provider, and you do a characterization of each of their leads and then you do clustering?
Jim Swift: We have as much data on companies as anyone I have ever seen, if not more. We have insight into things like purchase behaviors. We know [there is] about approximately $1.7 trillion in spending by U.S. companies on an annual basis. We put it into 45 categories, and we roll it up into different levels. This helps us get a more granular view into what companies are doing. We also bring in other behavioral data about hiring, news events, and all kinds of other things as well as the traditional demographic data, public record information, and others. We append all that information to the client’s customer base. So, they send us a file with their customers, we append this information and then use modeling techniques and define what equations, clusters and relationships best describe their customer base, using the information we have on them. Then we take that and project it against all the companies we know about that aren’t their customers. We come back with various confidence levels and ways to score the prospects they should go after. They then upload [these projections] into their sales force automation system, and they proceed to make the calls.
SM: You are providing them with targeted leads?
JS: That is correct.
SM: They are buying the leads from you, not just the segmentation information?
JS: That is right, because we have the data. We have some customers that will pull more raw information and do their own modeling. What I found, coming from the consumer world, is that there are more companies in the consumer world that have their own modeling teams and are doing their response models, risk models, and other things. In the B2B world, the data hasn’t existed. Any segmentation they do today is based on demographics, and mostly that is sized in terms of number of employees, sales volumes, and industry codes that the businesses are in. You can build some interesting things with that, but we can take it to another level of detail.
SM: What are some examples of the level of detail or characteristics that you can model on, that are interesting and non-standard?
JS: Purchase behavior is one of the big ones. We know a lot about what companies are buying. The way I think of it is companies tend to do three things. First, they have a capacity to create things – whether that is physical plant or headcount depends on the business, but they have a capacity. The second is production. The third is sales. That usually involves shipping if it is a physical product. We align our purchase behavior information around those categories. Then I want to know how these things interrelate. If you spend money on certain things that are tied into creating capacity – you might buy land, you might build a plant, you might lease a new office, etc. and I have different clues to identify those behaviors.
The second category is the production, so you have people involved in supplies and equipment, raw materials, etc. The third is the delivery of the service. When it is something a company manufactures and distributes, then transportation-related things are really useful there. We track 45 categories within those three broader categories. I want to be able to look at the relationships between those – if I see one area growing versus another, the model can pick up different things that might be happening in the business. If you look at other things related to people – it might come from purchases or also from hiring, it might come from things you do which tend to indicate the number of people you have, news, online information, unstructured and structured information that we can pull in. All of this gets brought in to this framework, and we organize the data in a way that we can model on it.