Business people sometimes wonder if commercial data vendors offer the key to their financial success and consumers worry that those vendors are selling their most personal secrets, but often data vendors know less than you’d think.
For decades, it seemed that every conservative political group in America mailed advertisements to my mother. That was a big, fat waste of ad money. Mom never spent a dime on politics, and she spit at the mere mention of Nixon, Reagan or Bush. So why all that mail? She subscribed to investing magazines like Forbes, Fortune and Barron’s, and to some direct marketers, that implied conservative ideals and cash to support them. The data flagged her as a good prospect, but on that count, the data was dead wrong.
That’s how it is in predictive modeling. Even with the best of data, models only produce correct predictions a fraction of the time. If the data is less than the best, don’t expect miraculous results.
If you’re considering buying data for your business, start small, and test one or several samples to assess the quality of the data. For example, if you’re buying a list of likely prospects, use a test campaign and track responses, or, if possible, fact-check the data sample. Even if it’s just a spot test, it might be worthwhile to pick a few names and verify them. Does the person exist? Is the employer, title and address information correct?
When I wrote Data Mining for Dummies, a book aimed at business people without data analysis backgrounds, I wanted to give readers a sense of what kinds of data commercial vendors provide, so I got some commercial data about myself and shared it as a detailed example. Some of the information was absolutely correct. Some of it wasn’t.
A few months later, I bumped into an executive of the data supplier, and mentioned that I’d used that data in the book. “How’d we do?” she asked. I wish she hadn’t asked. I didn’t want to be the one to tell her that her product wasn’t that great. But I didn’t have to; she clearly knew that there was room for improvement, and was working on it.
Some vendors have direct access to highly personal data sources. For example, credit bureaus collect information on consumer debts and payment history directly from creditors. These reports can go on for many pages for a single consumer. Every consumer can check the reports, and has good reason to correct errors. But these reports are neither relevant nor available for every marketing use. Most vendors of marketing data have far more limited access to personal information. They may struggle to develop and maintain accurate sources for information that marketers want.
It’s easy to understand the difficulty of providing information such as income; that’s private. But I wondered why, for example, the vendor had no information about my home, since that’s in public records. Still, public records are area vast and disjointed collection of sources, and governments are not always nice about sharing information. In fact, the difficulty of obtaining public records is a major driver of the commercial data industry. It’s often cheaper (and certainly easier) to pay a vendor for data than to deal directly with a myriad of less-than-helpful government sources.
Want a sense of what kind of consumer data you can get from commercial sources? Here’s what’s on offer from Acxiom, one of the largest commercial sources. Acxiom provides six types of consumer information:
- Characteristic: Age, gender, education, marital status
- Home: Own or rent?
- Vehicle: Make and model, insurance renewal date
- Economic: Income, estimated purchases with various credit cards
- Purchase: Number and amount of purchases in categories organized by channel (on and offline, mail order, and product types)
- Interests: A list of your interests, from a long, long list of possibilities
Some of this information, such as home and vehicle ownership, is a matter of public record. But you might wonder how they get data on private matters such as income, spending and credit card use. Banks and credit card issuers are not handing over all your financial records. Instead, this information is estimated, using sources such as surveys, loyalty program data and warranty records.
But, is it accurate? Sometimes yes, sometimes no, sometimes… kinda, but not really. At least, that’s what I see when I look at my own data. Looking at this from a data analyst’s point of view, I’d be reluctant to use this source for predictive modeling, since I can see a number of significant errors in the data. But all real-life data sources have flaws, so if I did not have a clearly superior source, I’d test some of this data in modeling and see what results I could get from it.
You can see what information Axciom has about you, and even correct errors, if you’re motivated (I’m not.) at https://aboutthedata.com/. And here’s a fresh sample (Yes, this is Axciom’s image of me, warts and all. I’ve disguised only the date of birth and income fields.) to give you a more concrete idea of what’s on offer:
Characteristic:
- Date of Birth 1/23/45
- Gender Female
- Education Completed Graduate School
- Marital Status Single
- Number of Adults [household] 1 Adult
Home:
No information found
Vehicle:
Auto Policy Renewal Month July
Economic:
- Estimated Household Income Range $50-75,000
- Presence of Credit Card Credit Card Holder - Unknown Type
- Number of Purchases - Cash 2
- Number of Purchases - Credit Card 1
- Number of Purchases - Amex 20
- Number of Purchases - Discover 1
- Number of Purchases -
Visa 1 - Number of Purchases - Other 11
- Online Purchasing Activity True
Purchase:
- Number of Orders - Housewares 2
- Number of Orders - Linens 3
- Number of Orders - Home Care 4
- Number of Orders - Garden 1
- Number of Orders - Jewelry 1
- …
- Weeks Since Last Order - Apparel 199 View/Remove
- Weeks Since Last Order - Men's Apparel 199 View/Remove
- Weeks Since Last Order - Men's Big and Tall Apparel 215
- [This continues with dozens of additional categories.]
Interests:
- Fashion Interested
- Children's Items Interested
- Computers Interested
- PC Owner PC Owner
[Again, the full example includes many, many more categories.]
