How to Make a Smooth Transition to Self-Service Predictive Analytics

Who’s Afraid of Self-Service Predictive Analytics?

In speaking recently with Doug Cogswell, the founder and CEO, of ADVIZOR Solutions I began to realize that, as usual, the tools and the technologies of predictive analytics were way out in front of what the average business person could easily consume. Most notably the latest and greatest ideas about making predictive analytics self-service and consumable by the every man or woman.

The solution offered so far has been to simplify the interfaces to these tools and make them more self-service. Unfortunately they are a little bit different from making pumping gas self-service (clear win – except for New Jersey who won’t let you) or from self-service checkout at Costco (not yet a clear win). The assumption is that if you can make the tools easy enough to use that business people will just be dying to use them. 

This hasn’t been my experience and it doesn’t seem to be what is currently happening.

It is Hard to Change a User’s Behavior

In my past lives I have built predictive analytics tools that were very easy to use – at least I thought that they were very easy to use. They took complicated decision tree models (CART) and made them accessible in simple segmentations and easy to use reports. They were very functional and intuitive but the business users did not take to them. 

Instead what seemed to work was to package up the tools into complete solutions for particular problem classes and for particular industries. So for instance we had little success in selling our very cool and easy to use “Discovery Server” tool (from Dun & Bradstreet / Pilot Software) but had great success in our business solution that predicted customer churn in the mobile phone markets. What was paradoxical is that customers would pay considerably more for a much more limited tool that did much less but was targeted at solving one particular problem extremely well.  But this was only true if it was designed in a way that made sense for a particular business. You can see this approach is currently being taken by Salesforce with their predictive analytics and artificial intelligence apps.

This idea of embedding the predictive analytics and the tool into a very narrowly defined solution space makes a lot of sense. It basically provides a ‘safe sandbox’ in which business users feel comforted by familiarity and where there are no ‘sharp edges’ where they can hurt themselves. As long as the business users stay in the sandbox they can wield the powerful (but sometimes dangerous) toy of predictive analytics.  This is where I think the self-service and automated predictive analytics offerings will continue to head in order to be successful.

But… there is another simple solution!  Business users can use offerings that wrap services around their predictive analytics offerings (whether they are self-service or not).

When Pretty Good Predictive Analytics is Good Enough

If you read the literature on predictive analytics solutions you invariably come across the Netflix case study. In 2009 Netflix handed over a $1 million prize to the best and the brightest data scientists who had successfully competed to improve movie recommendations by 10%. Teams from around the world slaved and battled for almost three years before they could achieve that small improvement. And they did it by combining the best solutions from several teams. It was the pinnacle of the best possible predictive analytics algorithms being wielded by the greatest data scientists of our time.

The Netflix prize was a huge success for Netflix but it was not that helpful in solving the problems found in other industries and companies. The reality is that most industries are not high tech like Netflix and aren’t built around petabytes of clean and useful customer data. Most industries, for example non-profits in healthcare and higher education, do not have the time or resources to build core competencies in predictive analytics or sometimes even in basic descriptive reporting. For them a solution that comes with some handholding is just what they need and they are more than willing to accept a pretty good solution (that is still way better than what they currently have) that is safe and robust as opposed to battling it out for that last 1% improvement with the most optimal predictive analytics algorithm the way that Netflix did.

New Concept: Service-Wrapped Solutions

ADVIZOR Solutions represents an important class of solution providers that have self-service predictive analytics offerings but also provide services. Together they provide business customers with a complete solution that is easy to use and safe.  Let’s call these the ‘Service Wrapped’ predictive analytics solutions.

The nice thing about this class of vendor is that the business person can start off with a lot of hand-holding (aka ‘service’) but then begin to use the tools directly over a period of time. Thus they smoothly transition from complete dependence on the vendor’s consultants to being able to move to ‘man the controls’ by themselves – to whatever degree they feel comfortable and at whatever speed they wish to do it.

Cogswell notes that there are two main characteristics required for business users to become direct users of self-service predictive analytics tools. (And by the way, there is nothing wrong in continuing to use the ‘service-wrapped’ solution if that works and is cost-effective).  To be good candidates for evolving to directly use these tools the business users need to be curious and they must have some sense of ‘causality’ in their business. By causality, Cogswell means that they must have common sense about what causes what to happen. The reason for this is that the models might sometimes find things that are not causal – i.e. they don’t actually do a good job of predicting what is going to happen next.

To be successful this approach also requires that the predictive analytics tools be as simple as possible (similar to the quote often attributed to Albert Einstein: “to be as simple as possible but no simpler”). So these tool vendors may well have only a few different predictive algorithms to choose from (perhaps tried-and-true multivariate regression and just a few categorical prediction algorithms). The tools must allow the user to specify some simple measure of robustness for the model and also be able to interact with the predictive model to validate what key predictive characteristics it is using to make its predictions. This goes back to Doug Cogswell’s statement about the business user having some common sense about causality. Self-service tools must allow the user to confirm and explore this causality.

Key Differentiators for the Service-Wrapped Solutions

We have seen that wrapping services (aka consultants) around a self-service predictive analytics tool can provide a rapid and smooth way to move from the “No-PA” state to actionable models being used in mainstream business to actual business users comfortably using the self-service tools. This ‘service-wrapped’ solution approach is applicable in companies where:

  1. They don’t have existing data science resources
  2. A very good solution is good enough and unlike Netflix they aren’t yet going for the perfect 1% optimization
  3. There is a desire to eventually use self-service predictive analytics tools

Four Years from Now

Currently the customers who could make best use of such self-service predictive analytics solutions are a bit confused. Their skepticism and fear of these new techniques seems to increase in lockstep with the seemingly never ending appearance of new terms and acronyms (e.g. “Big Data”, “Deep Learning”, “Artificial Intelligence”, or “Machine Learning”). The reality is that it doesn’t have to be that complicated if you take small steps rather than trying to take one giant leap towards predictive analytics perfection. A pretty good solution is way better than no solution and smoothly leads the way towards a very good solution – as well as enterprise autonomy in using self-service tools.

Cogswell believes that within four years we will see much more penetration of predictive analytics in the marketplace. The main driver for this will be the successful stories that business users share with each other when they are better able to deliver value to their customers and prospects. Wrapping services along with self-service products is an excellent way to get there successfully and smoothly.

Experts interviewed. Many thanks to Doug Cogswell who provided expert feedback on this article. Mr. Cogswell is founder, president and CEO of ADVIZOR Solutions Inc. www.advizorsolutions.com

Further reading and related blog posts:

 

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