David Raab Article Archive

One of the few disappointments in my otherwise charmed life is that my mother has never quite understood what I do for a living. My kids get it (“you help people buy software, right?”). But all Mom knows is I’m not a doctor or a lawyer, and don’t even appear to work for anybody other than myself. Technology consultant? It’s too abstract.

Writing this column just makes things worse. Mom is vaguely pleased that other people care what I think, but she knows I’m not a professional journalist. So it only adds to her confusion about my real job.

I only mention all this because I often suspect many readers of this column also don’t understand what David Raab does for a living. I also realize that most of you don’t care. But it’s still worth clarifying every so often, so you can better understand the perspective from which these articles are written.

This, as my kids have already told you, is the perspective of a buyer. I do indeed make my living helping people to select software, or sometimes service vendors, for marketing and customer-related applications. That means I spend a lot of time talking to users about their requirements, and even more talking to vendors about what their systems can do. Some vendor research is part of live selection projects, but much is done without pay (sorry, Mom) to find products that future clients might find useful.

One result of this–and yes I am meandering toward some sort of a point–is that I occasionally spend considerable time learning about products that never succeed in the market. Since I wouldn’t bother with these products if I didn’t find them inherently interesting, no great harm is done. But I do develop a fondness for certain categories, particularly if I think they offer real business value. Then it can be frustrating to watch them stagnate, and I find myself wondering why they haven’t caught on when it seems they should.

One of these categories is customer behavior monitoring software. By this, I mean software that identifies complex patterns in customer behaviors and alerts marketers when significant events or changes occur. Of course, most types of customer analysis involve some sort of behavior tracking, so what’s really significant here is the idea of complex patterns. Complexity is itself a relative term, but a reasonable rule of thumb might be that complex patterns involve three or more transactions over at least as many time periods. This is enough to look for trends and deviations from trends, as well as other mathematical and chronological relationships. That’s still pretty darn vague, but the point is we’re looking beyond simple sums and averages, and we’re looking at the customer’s own behavior, not how the customer ranks relative to others. Okay, that’s two points. No apologies: this stuff isn’t simple.

Behavior monitoring software has a fairly long history outside of marketing, for generic process monitoring tasks like running oil refineries and for customer-related activities such as detection of fraud and money laundering. The earliest marketing applications date to the late 1990s, with products including Harte-Hanks Allink Agent (now Daily Deposit Builder), Verbind LifeTime (later purchased by SAS) and Elity Insight (now part of MarketSoft). These differed considerably in their origins and technical details, but shared the general characteristics of examining daily transaction streams, detecting complex patterns, and issuing alerts when specified patterns occurred.

The key technical difference between behavior monitoring systems and traditional customer analysis tools was that the monitoring systems stored transaction history in ways that made it easy to detect patterns and deviations from patterns. For example, a bank deposit of the same amount every Friday is probably a payroll check. A behavior monitoring system might set up a record with buckets for each week’s payment, perhaps going back three months. New transactions would be noted as they arrived, and an alarm could be issued if an expected transaction did not appear. This is considerably more efficient than writing a SQL query to select transactions that meet the pattern and to recognize when one is missing. A behavior monitoring system might continuously track dozens or even hundreds of such patterns, although not all would appear for each customer.

In essence, these patterns became new data about each customer. Because conventional analysis tools could not process patterns efficiently, it was data that had not previously been available. More important, it was useful data: typical patterns included declining usage (often an indication of impending attrition), unusually large deposits (often an opportunity to capture money moving from one investment to another), or a sudden increase in purchase frequency (possibly indicating a life change such as new home or baby). And, because the systems were updated daily, new data became available quickly enough to do something with it–change a credit limit, send a new marketing offer, or simply make a customer service phone call to ask what’s up. Timely contact, rather than precisely identifying the best response, was the systems’ real benefit.

Early users of these systems, largely in financial services, reported great success. Payback was often less than one year. It seems that contacting people when they are receptive, even if the message is somewhat untargeted, is tremendously effective. Who knew?

But, despite documented return on investment at brand-name clients, adoption of behavior monitoring software for marketing has been painfully slow. After years of selling, no vendor has much over a dozen installations. The problem is not the technology itself–fundamentally similar systems from other vendors have hundreds of installations for fraud detection, anti-money laundering, insider trading, and related applications. Defining patterns and reaction rules is somewhat labor intensive, but vendors can now make recommendations based on previous results. Nor is it necessary get everything perfect at the start: a few simple patterns that detect obvious opportunities can more than pay for a system. Vendors have even offered the software on a hosted basis to further reduce deployment effort, speed implementation, and cut initial expense.

So why haven’t these systems been more successful? There are two common explanations.

– timing was bad. Marketing behavior monitoring systems were just appearing when the tech bubble burst and the general economy went into recession. Many companies then avoided non-essential investments, and were particularly leery of new marketing technologies. This harmed behavior management systems, even though their proven return on investment should have given them a solid financial justification.

– execution was difficult. Although behavior monitoring systems could identify marketing opportunities, most companies in the key financial services sector were unable to act on them because of constraints in their outbound channels. Email and mass telemarketing were ineffective because customers did not respond well. Calls from personal bankers and brokers were effective, but companies lacked lead distribution systems or the bankers and brokers had other priorities. Direct mail could also be effective, but personalized mailings could not be produced overnight in small quantities at reasonable cost. With no way to use outputs from the monitoring systems, companies declined to purchase them.

Neither argument is fully convincing. Some marketing technologies have sold well during the past few years, such as email and voice response systems. And while outbound channels are often a bottleneck, it’s hard to believe the problem is so nearly universal.

Experience may yet uncover the correct explanation. Technology budgets are loosening a bit, marketers seem somewhat more interested in experimentation, and outbound infrastructures are increasingly mature. Behavior management systems themselves are attracting more attention: SAS and MarketSoft will add some marketing muscle to the category, and new products have been introduced by firms including Fair Isaac, Synapse Technology, Intelligent Results and Loyalty Builders.

These new products are interesting in part because they take different approaches from the earlier systems. The older systems, from Harte-Hanks, SAS/Verbind and MarketSoft/Elity, all require users to predefine the available patterns during the setup process. This is done by human analysts, assisted at best by semi-automated software. Fair Isaac’s OfferPoint and Loyalty Builders work roughly the same way. But Synapse and Intelligent Results are different: they use statistical methods to identify patterns automatically. This reduces implementation effort and, at least in theory, enables more powerful results.

The new systems also produce different types of output. The earlier products all produce alerts or recommendations by using rules to specify which patterns should generate which results. By contrast, Synapse simply identifies customers with anomalies between the current and past behavior; it doesn’t directly link these to predicted actions like attrition or new purchases. (Actually, Synapse does let users write rules that attach such labels to alerts before they are sent to the outbound channels, but this is an optional, manual process.) The other three new products move beyond alerts to predictive model scores that rank the likelihood of specified events. Intelligent Results builds its models as part of the same process that determines which patterns to track, by training against sample cases with known outcomes. It relies on statistical techniques more commonly applied to text analysis. Fair Isaac OfferPoint and Loyalty Builders first identify the behavior patterns and then use them as inputs to conventionally built predictive models. Table 1 summarizes the approaches of the different vendors:

Every vendor naturally believes its method is inherently superior. But it’s important to recognize that each has had some success, and none has more than a handful of active implementations. Only time, and many non-technical factors, will determine whether any one approach emerges as dominant in the market. It’s quite possible that this will never happen, and different techniques will prove best suited to different situations.

From a buyer’s perspective–and remember, that’s the one I write from–the specific differences between approaches are less important than the fact that users have a wider range of options. More choices means more chances that someone will have found a winning combination. Behavior monitoring for marketers seems a good idea whose time may finally have come. We’ll soon find out.