Creating a predictive model factory


For all of you who have ever been to Heidelberg, location of the headquarters of SAS Germany and the place where I live, you might know that a town called Mannheim is nearby. Some might even know that Mannheim is the place where a genius German engineer called Carl Benz invented the car 125 years ago. In 1886 he filed a patent for a motorcar, which is accepted to be the first engine-driven car in the world.

Mannheim recently celebrated the 125th anniversary of the car - a German success story of invention and innovation. However, what also came to my mind was that the market success of this invention, without which modern society would not function as it is today, was probably more fueled by the innovation of a car maker in the young United States of America. When Henry Ford revolutionized the car making process by introducing the assembly line – the process that is still used worldwide in car making today – he laid the foundation for the democratization of the car. This assembly line reduced the cost of making a car to an amount that made it sellable to a much larger audience and as a result started to provide value to a larger market.

There are parallels with the evolution of predictive analytics in modern business organizations. There is a lot of invention and innovation going on in many organizations, sparked by very smart people, and business value is created in niches through this ad-hoc approach. However, more and more organizations have started thinking about ways of embedding predictive analytics into organizational processes in order to streamline workflows, remove bottlenecks and leverage synergies. Much like the invention of a leaner production process for the car, organizations have started to introduce leaner production of predictive models. That is why a name like “Model Factory” came naturally to describe this process.

With the implementation of a model factory process, organizations can remove bottlenecks  effectively, such as:

  • The preparation of the analytical base table (ABT), which can be seen as the supply chain of material for model production.
  • The efficient development of models as well as the migration of the predictive model score code from development to testing and production, which can be compared to the actual production of the models.
  • The quality control of the models in production, which can be seen as the warranty for the models.
  • The retraining of stale models, which can be seen as service of the models.

The metadata-based integration of the crucial products (SAS Data Management Studio, SAS Enterprise Miner and SAS Model Manager) together create a technology infrastructure that enables organizations to focus on the content of the supply chain of models rather than on logistical and organizational bottlenecks. And, similar to the car, the continuous simplification of the interfaces supports the democratization of the predictive analytics process, taking it out of the hands of experts and involving the masses. This will help further in spreading business analytics throughout the entire organization and reducing the risk of concentrating business-crucial knowledge in the heads of a few.

For the analytical expert this is a promising direction as well. Rather than spending large parts of the day with repetitive tasks that can be done more effectively and efficiently in an automated way, they have more time to be innovative and creative, which is much more fun. Also, the distribution of the predictive analytics process across many stakeholders will create more acceptance and insight into the value of business analytics in an organization, giving the analytics expert more visibility. This is why I see the dawn of the model factory as a true win-win situation for companies that want to become analytical enterprises.


About Author

Sascha Schubert

For more than 20 years Sascha has been helping SAS customers and prospects all over Europe to design, customize and proof solutions for SAS Advanced Analytics in industries such as Banking, Insurance, Telecommunication, Retail and others. In his daily work he applies SAS predictive analytics and machine learning to Big Data to make business processes more efficient and more effective in a digital world. Sascha has a PhD in Statistical Climatology from Humboldt-University of Berlin.


  1. Great article. I like to talk about the need for "industrialized analytics" but I like the factory concept too. A couple of comments:

    Enterprise Miner and Model Manager are certainly important but so is Rapid Predictive Modeler - the ability to broaden the base of people building models and still manage them as part of the "factory" is going to be really important as companies step up the number of models being built.

    I also think that the cloud has a lot to offer in this regard as we found in the recent Predictive Analytics in the Cloud research/> of which SAS was a sponsor.

    Finally, in my recent book Decision Management Systems, I talk about the integration of predictive analytic models into production decision-making systems. The need to wrap business rules around the predictions to turn them into decisions should not be underestimated. As YouSee found (a SAS customer I interviewed in this great video) you need the rules to make the predictions truly actionable.


  2. Nice blog Sascha! I fully agree that the trailblazers in innovative predictive modelling will set the stage for those organizations who are successful in industrializing analytics to reap the big benefits (much like Benz did for Ford...though I suspect no one is starving in the Benz family either 😉

  3. Great parallels between auto industry and the analytics industry in its adolescence. However I think true democratization of analytics will occur at the consumption level when micro and small businesses can actively start using the power of predictive models in planning/running their activities. After all this is what the motor car has enabled.

  4. The beauty is to discover that my research work is quite like creating a predictive model and the inspiration is to find a new advantage of my research work: by solving common problems in a programmed and structured procedure effectively and efficiently, the human experts can have more time and more fun to solve more difficult problems in much more innovative and creative ways.

    My research work is to develop an Expert System(One kind of Artificial Intelligence) on water distribution network decontamination to help the water utility managers or operators make their decisions by offering them guidance on optimization of the currently available and possible technologies and recommend a thorough and expedited solution to contamination.

    The Expert System (called Decon) initially aims to help water operators in reducing the guesswork and preventing them from falling back to heuristic solutions that may not be adequate for the event in hand. This advantage is quite similar to quality control in production.

    I totally believe that creating a predictive model factory can be a promising enterprise and not only in business area but also in engineering and other wider fields.

Back to Top