Join our upcoming online event - "CDO TechVent: Data Observability- Thursday, August 25, 11:30 a.m. Eastern Time.

Predictive Analytics Mediators: The Key To Outsourcing Data Science

So you’ve decided to outsource your predictive analytics.  Good for you! There are a lot of advantages to outsourcing your predictive analytics. Cost savings and access to top talent being two great reasons. But there is one job title that you are going to need when you outsource that you may not realize that you will need. You’re going to need a predictive analytics “mediator”.

What is a PA mediator? 

Now ‘mediator’ is kind of a strange name and it really doesn’t do justice to this role as ‘mediation’ is just a small part of the job function. But it good because it emphasizes a small but important part of the job description. The role itself it encompasses many of the following job titles: manager, envoy, ambassador, mentor, analytics lead scientist, educator, thought leader, project leader.

More specifically your Predictive Analytics ‘Mediator’ will be the person who will do the following things for you:

  1. Translate between the strategic business needs and the brainiacs who will be your Predictive Analytics Professionals (PAPs) who actually do the heavy lifting
  2. Interact with senior management and set reasonable expectations and help secure sufficient resources for success
  3. Mentor the super-smart PAPs (Predictive Analytics Professionals) so that they stay on task for the business
  4. Enforce deadlines and play the role of project manager so that analysis doesn’t arrive in the ‘nick of too late’ to be useful for the business.
  5. Enforce good A/B testing and test and control structure so that the ROI of predictive models can be measured
  6. Push back against marketer requirements that a particular algorithm be used and instead delegate the selection of the best predictive algorithm to the PAPs
  7. Interface and make sure there is efficient communication and well-defined project deliverables between any in-house and outsourced analytic resources
  8. Own profit center and be on the hook for making sure that the predictive analytics investment is producing real value to the business

So here’s the problems if you don’t have a PA mediator:

  1. There is a communications gap between the PhDs and the MBAs. Effectively they speak different languages and have vastly different cultures.
  2. The PhDs may get stuck in a rut and not consider other solutions. A mediator can propose new algorithms, techniques or data sets that the PAPs may not have considered.
  3. The PhDs will do exactly what you tell them to do but they may be lacking in seeing the big picture. It is very rare that you just need a predictive model based on your first specification. Getting the right model (even just defining the right problem) requires a lot back and forth and business interpretation as the analysis begins to gel into something actionable.
  4.  Senior management doesn’t know how to hire PAPs because they don’t know a good one from a bad one. It’s like asking a PhD astrophysicist to hire a great salesman or vice versa. The salesman will hire a great salesman and the astrophysicist will hire a great astrophysicist.
  5. Senior management just read an article in the Wall Street Journal and they need to get some of that “predictive analytics” that they fear their competitors have. Problem is that the smart folks who understand the business don’t really know where PA could be applied or whether it could effectively be applied.
  6. Your analytics team will careful set up a test and control strategy to measure the effectiveness of your marketing initiatives and your ad agency will mess it up and you won’t be able to measure the ROI of your programs.

Hiring a predictive analytics mediator is like hiring a general contractor to build your house

So hiring a mediator whether in-house or as a consultant is going to cost you money.  But would you build your house without a general contractor? If you don’t have experience in home construction would you want to interface directly with the plumber, electricians and carpenters? Could you speak their language?

In other instances this role might also be called and ‘engagement manager’ or ‘program’ or ‘project manager’ and some industries have these roles for analytics that you wouldn’t expect. For instance Ken Rudin, the former VP of Analytics at Zynga, characterized the maker of Farmville as “an analytics company masquerading as a games company”. Zynga would often hire product managers from finance or consulting who had analytics skills as well as business competence and good project management skills.

Unfortunately hiring a general contractor for your house is a lot easier to find than an employee who has the rare combination of business, managerial and statistical skills. If you can find one and can afford one go ahead and hire them. Otherwise you might want to consider using a consultant or possibly creating your own Predictive Analytics Mediator by training an analytics person on your business (it is much harder to try to teach an MBA statistics than it is to teach a PhD in statistics enough about business to be useful).

So here’s the benefits when you hire a predictive analytics mediator:

  1. Business Accountability – you can hold your mediator accountable for business goals like ROI rather than analytics goals like 95% accuracy.
  2. Top talent – your mediator will know top talent when he or she meets them and can also help you find them. Unless you really know the difference between CART and CHAID decision tree algorithms you might want to hand off to the experts.
  3. Stable team – even if your PAPs are not employees you will still be investing in them and training them as they get to know what you really want and you begin to understand what their capabilities are. Your mediator can help to make sure that the same people keep being used for your jobs (even if you have off-shored your predictive analytics team). This is important because every time you get a new consultant you need to retrain them to your business or you risk getting a perfect model for the wrong business problem.
  4. Timeliness – a mediator can more fairly set deadlines because they know what is involve and they can more expertly assess whether a delay is warranted by something unavoidable in the analysis or if the delay is unacceptable. This fundamentally will also help you to build and retain a top-tier team and grow loyalty as your team comes to expect knowledgeable fairness from you.

Here’s how to get started

If you are looking to find a PA mediator you can certainly start on the web or by attending any of the conferences on big data or business intelligence or predictive analytics. These guys will be expensive because it is rare to have both the business and analytics skillsets but they will be a necessary way to get started.

I’d recommend that your mediator initially be a consultant rather than an employee as you get started. First you want the best possible person and it may be difficult for you to attract such a skillset to your company since it will not be your core competency. It is also good to use them as a consultant as a way to try them out. Engage them in strategy meetings about your business. They should be able to quickly spot the areas that predictive analytics can benefit your company and be able to explain to you in understandable language how a predictive analytics project could be accomplished. If your mediator doesn’t get your business after the first meeting or two or talks in technical language that you don’t understand then it’s time to find another. It’s even a good idea to have a small ‘get-started’ project that multiple mediators could try at the same time. Then hire the one that does the best job.  Over time you can grow your core competency either in-house or outsourced depending on the successes you have... no rush but it is easy to get started today!

Thanks to Brij Masand who provided expert feedback on this article. Mr. Masand has over 15 years of experience in the field of data mining with special expertise in building lifetime value models through offshored analytics teams.

Stephen J. Smith

Stephen Smith is a well-respected expert in the fields of data science, predictive analytics and their application in the education, pharmaceutical, healthcare, telecom and finance...

More About Stephen J. Smith

Books by Our Experts