There have always been multiple ways of making sense of data and solving business problems with analytics. Perhaps one of the earliest classifications was the descriptive/predictive/prescriptive one, which I can’t remember whether I invented or borrowed (I feel like I invented “prescriptive,” but I could be wrong, and I certainly didn’t invent the other two). There are other important distinctions as well, including whether the analytics are for internal decision support or customer-oriented products and services, and whether analyses use “traditional” analytics or artificial intelligence.
Whatever your favorite taxonomy or category of analytics, I’m here to argue that it doesn’t matter from an organizational standpoint. That is, I think all the categories should be supported by one organization. I have certainly observed less inclusive approaches in organizations—for example, a “business intelligence” group and an “advanced analytics” one, and others in which analytics and AI are supported by separate organizations—but in general I believe these are not good approaches to supporting analytics. I’ll spend the rest of this post arguing why an inclusive approach is more effective.
I’m assuming that in your organization, analytics groups do multiple things. They consult with internal groups on their business issues and decisions, and how they might be addressed with analytics or AI. They recommend or use particular tools to pursue those objectives. They either generate the analytics themselves, or assist the business user in doing so. Then—if they are doing a good job—they help to deploy the analytics in production systems and processes.
Categories of Analytics Do and Should Overlap
Perhaps the best reason for dealing with all types of analytics through one organization is that the different categories often overlap, and should do so even more than they do today. An analyst hoping to do prediction should, in many cases, explore his or her data with descriptive analytics first. Companies planning to push analytics that make recommendations out to customers or employees—in other words, prescriptive analytics—first need to develop predictive models. And, as we’ve argued before on this site, the differences between predictive analytical models and supervised machine learning are subtle at best.
In many cases, analytical work is based on a discovery process over time that involves the use of increasingly sophisticated methods. If an internal customer for analytics has to move among multiple support organizations for help with these different methods, the discovery process isn’t likely to work smoothly. For example, if a business user consults with a “business intelligence” group to address its analytical needs, it’s unlikely to hear much about more advanced forms of analytics and AI. And in the opposite direction, chatting with an AI group about your problem is unlikely to yield a recommendation of, “All you really need is a bar chart.”
How an All-Inclusive Analytics Organization Works
Of course, not all quantitative analysts and data scientists have the same skills for all forms of analytics. If they’re combined in one organization, they need to be able to flex to deal with a wide variety of internal customer needs. That means, among other things, that each member of the analytics group staff needs to be classified in terms of what kinds of analytics they can do or support. The classification can change over time, of course, as the analyst learns new methods and skills. In this type of all-inclusive organization, analysts and data scientists should be encouraged to acquire new capabilities all the time.
It also requires someone at the front end of the process who can understand their problem and oversee the customers’ needs for analytical or AI solutions. Others have called this role an “analytics translator,” and it’s a very important one. The role may be part of a centralized analytics organization—with assignment to specific business units or functions—or may be decentralized to report into the business. Companies like Procter & Gamble have had such roles for over a decade—initially centrally structured but embedded into business units, and now reporting directly to them with an indirect link to the central analytics group.
The translator can advise business users of analytics and AI on what methods and tools are most relevant to their problem. If there is a need for deeper support from an analytics or AI specialist, the translator can recommend the right one for the job. To succeed in the role, translators need to be familiar with typical business problems and decisions in the units they support, and also with the range of analytical solutions available to address them. Their expertise is broad rather than deep, but it is just as valuable as detailed knowledge of a particular analytical method. I’ve known a few companies that disparage people in the translator role as “light quants,” but criticism of this critical role isn’t justified or appropriate.
An all-inclusive analytics organization does presume a fairly high level of central coordination of analytics, and a desire to match the best people and methods with the analytical problems they are prepared to solve. It doesn’t require full centralization, but it is probably incompatible with full decentralization or a loose community of practice. However, I think it’s the most effective way to pursue analytics and AI within a large, sophisticated organization.