Tweet This
A lot of business people express interest in predictive analytics, but don’t know how to spot a good opportunity to use it. But it’s not a secret science. The characteristics of best-bet analytics applications are pretty simple.
By a “good” opportunity, I mean one that gives you a fighting chance of solving a real business problem and making a positive return on investment.
First, a little on the bad bets in predictive analytics. Let’s start with an easy problem to spot. If you can’t explain how you expect to make money from your analytics investment, you’re in trouble already. Don’t let anyone convince you to buy expensive stuff on the promise that you’ll gain “insight.” Nope, don’t do it. You’ll end up unhappy, or worse yet, you’ll convince yourself that some little tidbit of information justified the purchase.
If you have to convince your management to make a significant investment to pay the costs of doing analytics, don’t start with projects that focus on increased revenues. Wait a minute, doesn’t management like increased revenue? Yes, but that doesn’t mean your case will be convincing. Remember that every manager has heard such claims before, and has doubts about results. Plus, your manger may not have much authority to follow through on your plan to increase revenue. So, focus on analytics projects aimed at decreasing costs first.
Don’t look for drama. Investing a fortune on one big decision is risky, whether the decision is based on analytics or not. You’ll lower your risk and get better results by choosing cases where you can make lots of small, low risk decisions. If you can make the right choice a little more often than you do now, you’ll make money.
Direct marketers have been making money this way for decades. Mailing advertisements to prospects costs a lot of money. So good direct marketers have learned to test every element of their pieces, from the color of the envelope to the price to the words used in the postscript (the “PS”) of a sales letter. Based on these tests, they know what’s most likely to appeal to any given prospect. Now, “most likely” is not a guarantee that the person will respond. In fact, most don’t respond, no matter what the offer. But by experimentation, it’s possible to discover enough to create profitable marketing campaigns, and to do so consistently.
Let those direct marketers be your role models. Look for these characteristics to spot a good predictive analytics opportunity:
- Appropriate data is available. This means data that is relevant to your business problem, high-quality (correct and clean), and available in sufficient quantity for the use you have in mind.
- Many opportunities to predict. Think lots of little interactions. In direct mail, thousands of prospects. Internet advertising, perhaps millions of interaction where ads may be served.
- Benefits for correct predictions, no big loss for incorrect. Serve an ad and make a sale, great! Serve and ad, but don’t make a sale? No big deal. Get it right a little more often and you’ll make some money.
Here are some examples of good predictive analytics applications. Read them over, and you’ll get a better sense of how to spot a good bet for predictive analytics in your own work.
- Buy or don’t buy. Should you make an offer to this person at this moment?
- Close account or not. Is this person about to stop doing business with you? Should you try to save the customer?
- Share post or not. Will this person click “like” or “share” right now?
- Spend how much? How much is this person likely to spend today? In a lifetime? On a particular class of products?
- Match document to query. Which document is the best response to a search?
- Route inquiry to proper department. Where should this service query go – to customer service or tech support?
Don’t try to be a hero by choosing big, high-risk projects to start off in predictive analytics. That approach is a great way to experience your first analytics flop. Instead, choose the low profile, good-bet projects. The results will make you a real hero at the office.
