Every time I speak at a conference or on a webcast, one of the most frequent questions involves the “best” way to organize analytical and big data activity within a large organization. Should the function be centralized or decentralized? Should analysts and data scientists be attached to business functions and units, or in a central pool? To which existing function or organization should it report?

I confess that I have provided a somewhat namby-pamby set of answers to these questions in the past. “There is no best organizational structure for analytics or anything else,” I have intoned. “Any structure’s weaknesses can be addressed through countervailing mechanisms,” I have said wisely if evasively. “Any function can provide a good home for analytics.” All of this is technically true, but unsatisfying.

But I’m not the only party who has failed to make a commitment. Many companies are attracted to small “centers of excellence” (CoEs) that put a small number of people in a central coordination role, but leave the great majority of quants to fend for themselves in highly decentralized environments.  This is appealing if you want to apply a gloss of coordination to a largely uncoordinated activity, but I don’t think it suggests a strong commitment to a well-organized analytical capability. It’s not the worst structure, as I describe below—but it’s not the best either.

So for the remainder of this post I am going to rank my preferred organizational structures in terms of their likely effectiveness. Most of the ranking is based on empirical observations of different structures over the years, but some general logic comes into play too.

  1. Central analytics and data science organization, based in a Strategy function, with analysts assigned to and rotated among business units and functions: This is, I think, the optimal structure and home for analytics and data science. The central function allows for a critical mass of quants and for central coordination of their skill and career development. It should be a shared service, where anyone with the money and the high-priority requirements can get the help they need. The assignment and rotation allows for close relationships between analysts and decision-makers. The location in Strategy suggests to the organization that analytics is strategic, and allows for the analytical team to focus on broader issues of effective decision-making.

  2. Same type of central organization, reporting to IT or Finance or maybe R&D. IT has the virtue of knowing the technology and data structures that work with analytics and big data, and they will clearly have to help implement the solutions in any case. The downside is that IT organizations tend to want to create common, leveraged solutions, and analytics often can involve one-offs. Procter & Gamble is a great example of making analytics work in IT (called Information and Decision Solutions). Finance also has strengths and weaknesses as a home; being located there would help to ensure monetary resources, and analytics would presumably be applied in a way that leads to ROI. The potential downside is that applications outside of the Finance function may get short shrift. Caesar’s Entertainment is one example of a well-functioning analytics shop within the Finance function. Analytics and big data can also fit within R&D, particularly if a major activity involves developing new products and services with those capabilities. GE is a good  xample of this. The downside is that not all applications of analytics will fall into the R&D bucket. GE is great at big data and analytics for R&D, but somewhat less impressive to my mind at using analytics to make better internal management decisions.

  3. Center of Excellence, located in one of the above-mentioned functions. Yes, I disparaged it above, but it does have its virtues. A shallow layer of coordination is better than no coordination at all. A small group of central analytics and data science managers could make some progress at assessing needs, prioritizing projects, and managing analyst careers. It’s a lot to ask of a small group that doesn’t actually own any quants, but some of the desired objectives may be accomplished.

  4. Analysts and data scientists in one function, e.g., Marketing. This is not a totally bad idea if almost all the analytical and big data activity is within one function, and Marketing is a big area of application for many firms. The problem, of course, is that analytical opportunities in Finance or HR or Supply Chain may well go unnoticed and unaddressed. There’s no getting around it—if all the quants report to the Chief Marketing Officer, he or she is going to want to see a lot of their activity, if not all of it, directed toward Marketing.

  5. Fully decentralized analysts with no coordination. This is a bad idea—so much so that if there were any other real options I would rank them before this one. It means that quants feel alone, that they work on only one type of problem, and that innovative solutions won’t be shared across organizational boundaries. Don’t do it.

OK, I have come out of the closet with my true preferences. However, I still have to say that no single structure is perfect—even my #1 pick above. No matter what structure you choose or design, something will go wrong with it, and will have to be addressed through means other than structure. But that doesn’t mean that some structures aren’t better than others.