Analytics programs often flop. And many analytics success stories hide a dirty secret – they’ve been exaggerated, claiming credit for just about anything that went right, whether or not it was really the result of a disciplined data-driven decision process.
Why so many failures? Is analytics a losing proposition? Does it take a brain surgeon to make it work? No.
Analytics failures nearly always have the same root cause: there was never a realistic plan for success.
Take a look at the marketing materials of vendors in the analytics space. The language is so repetitive, so vague. They promise you will “get insights,” “gain unprecedented insight,” “uncover insights more quickly and more easily,” and many other small variations on the spellbinding theme of insight. So, tell me, exactly what’s an insight worth?
Let me give you an example. At an industry event not long ago, one retailer presented an analytics success story. The retailer, a seller of women’s clothing, purchased social media monitoring services and text analytics software. Using these resources, the retailer’s staff noticed that some women said they liked this or that item, but needed a bigger size. The retailer added larger sizes for a few products. Retailer declares analytics success!
Wait a minute. What problem was the retailer trying to solve? Nobody defined that. Specific business problems didn’t seem to come up in the decision to use analytics at all. What was the goal? Umm, no goals mentioned, either. Did the new products sell well? Nobody said anything about sales. Which means that we don’t know if the new revenue outweighed costs of adding additional products, or investing in analytics. In light of these simple questions, the success story looks less successful.
Another thing. Was this analytics investment really necessary to learn that women want clothing in larger sizes? The size distribution of American women is not secret information.
The success story was backfilled. The retailer made an investment, and needed to come up with a rationalization to justify the costs (and maybe a rationalization to justify the costs of attending the conference). But it was not a good case in favor of analytics.
Approach analytics as you would any other business investment, Make a business case, and a plan for success. It’s the surest way to avoid analytics failure.
A good plan starts with something you already understand: a business problem. If you’re new to this, choose an easy one first. Make is something fairly small and well-defined, a problem whose cost is well understood, but isn’t one of the company’s biggest issues. Don’t raise your stress levels with a project that draws a lot of attention from the beginning. You can have attention later when you get great results.
Now, figure out what data you will need to understand the problem and options for correcting it. Is that data available to you now? Why not? Do you need special permission to use it? Is it something the company gets, but doesn’t keep? Is it something you could start collecting now? Can you get it from an outside source? Decide what you have to do to get the data, and do it.
Outline your process. Write things down. Prepare a summary of the business problem and its impact on the organization. Define goals for the analytics project, keeping them reasonable and modest (you can always be a hero by exceeding them). Make a rough plan of the analysis to be done. Check resources – do you have the right data, the right people participating, the right resources for them (computers, software, workspace, and so on)?
Let me say this again: write down every element of your plan, including the goals.
No matter what you call your approach to data analysis (statistics, data mining, data science…), you can get top-notch guidance for planning and executing the work by following the CRISP-DM process. It’s a step-by-step structure for analytics projects (developed by a consortium of 200 organizations committed to using analytics) that helps you maximize the chance of meeting your own analytics and business goals.
There’s a step-by-step guide for using the CRISP-DM analytics process that is available free for download, so get it now. (Or go to your library and borrow a copy of my book, Data Mining for Dummies, which includes a lighter explanation of CRISP-DM.)
As you work, you may uncover facts that lead you to change your plan. That’s normal. But it’s also normal that people don’t make notes about things like that as they work. Write it down, write it down, write it down! You memory isn’t perfect. Keeping detailed documentation preserves your work, helps you show value, and prevents you from going over the same steps again and again.
Why so many analytics failures? It’s because so many people don’t plan for success. Think it through, make a realistic plan, and you’ll have yourself a real analytics success story.
