The goal at this stage is to match likely consumers with likely product choices. The program’s “rules engine” — the heuristics it follows to identify the most potentially valuable consumers — knows, for instance, that visitors browsing expensive jewelry are more likely to buy if they come from Beverly Hills than if they come from a less-affluent area. Just as important, however, is the ability of the program to learn from past transactions. Over time, as Michigan’s economic fortunes have eroded, for example, the scoring model might note that people from tony Birmingham, outside Detroit, are now more hesitant than they once were to buy pricey items, especially compared to, say, individuals from less-hard-hit suburban Boston.
Once a visitor is identified as a hot lead, another filter determines whether to invite him to chat — that is, the program analyzes whether talking to him is virtually the only way to convince him to make a purchase. Think of a floor clerk in a Sears major appliance department sizing up several customers and approaching the one who appears most certain to buy, using intuition drawn from experience. On one level, deciding who to invite for a chat is a simple scheduling problem: Are there enough agents available to handle the chat? Increasing the number of agents means increasing the number of invitations to chat, which in turn means approaching colder leads who are less likely to end up making a purchase. The colder the lead, the lower the potential profitability. On a more strategic level, the software must determine the number of agents that will maximize profitability. Further statistical modeling is needed to select the right agent for each consumer, depending on such criteria as the best-performing agent for the product category that individual is looking at. Even a great used-car salesman isn’t likely to make much money working at Tiffany.
Now, it’s time to chat. Here the goal is simple — to translate the art of selling into a science. Once that Sears clerk approaches a prospect, he has to use his experience to make dozens of instantaneous judgments, based on any number of visual and linguistic cues: Is the customer detail-oriented, or does he prefer a softer touch? Am I pushing too hard, and is he beginning to resist? The customer appears to be losing interest — is now the time to begin offering discounts? The 24/7 chat format, of course, does not allow for all the nuances any decent salesperson picks up in a face-to-face conversation. It does, however, perform analyses of thousands of chat transcripts, through text mining and data mining, to perfect the techniques that human customer service representatives use to close the sale.
Text mining, for example, can offer insight into how a salesperson should talk to consumers to achieve the greatest degree of success. Some of these insights are based on extensive research in neurolinguistics, which argues that people can be classified as aural, visual, or kinesthetic, depending on how they perceive the world. That classification, in turn, can provide hints of the most effective communication strategies for convincing them to buy an item. Aural consumers listen for product details, so an effective sales approach might be, “Let me tell you how many megapixels this camera has.” Visual consumers want information about the product’s appearance. Thus, the salesperson might say, “This camera comes in three exciting colors and will fit into your shirt pocket.” And kinesthetic consumers respond to pitches that tap emotions, such as, “You’ll love how this camera balances in your hand, and it’s perfect for taking pictures of your granddaughter’s nursery school graduation.”