How to Evaluate Service Bureaus (1,440 words)
Current State of Many Service Bureaus
Why have so many service bureaus failed to fully adapt to the times? I think because merge/purge has been perceived as a commodity for years, and therefore has become a self-fulfilling prophecy. In such an environment, service bureaus tend to compete on price. And often, direct marketers get exactly what they pay for in terms of quality.
In contrast, the 1986 service bureau evaluation with which I was involved generated immense industry interest. We were invited to present the results of our study at a Direct Marketing Association trade show. Also, the editor of a leading journal attended the seminar and published our findings in a series of articles.
By the early 1990s, however, the focus on service bureaus and merge/purge algorithms had vanished. The industry had moved onto other topics such as neural networks and open systems data warehousing technologies. Merge/purge had become a perceived commodity not worthy of attention.
Increased Need for Data Hygiene
Ironically, cutting-edge service bureau work is more critical today than it was in 1986, because the quality of the data, I believe, is worse than it's ever been. In the 1980s, most direct marketing orders arrived via the mail. Now, inbound call centers capture the bulk of transactions, many of which contain glaring misspellings and address element omissions. This is particularly true of small to mid-sized firms that can't afford to integrate real-time hygiene technologies into their CRM infrastructures. And Internet data presents its own name and address quality nightmares.
Powerful merge/purge software is essential for correcting many of these problems. Name and address hygiene has significant ramifications for data mining and quantitative analysis. It's important to identify records that correspond to multiple orders by the same individual and consolidate them into a single customer view.
Consider the analytical ramifications of two legitimate duplicates that aren't consolidated. During the building of a predictive model, a multi-buyer will appear to be two separate—and less desirable—single buyers. Likewise, during a lifetime value analysis, one relatively valuable customer will be represented as two not-so-valuable individuals. (The opposite, of course, will be true whenever an illegitimate consolidation takes place.)