Defining and embracing ModelOps: Scaling up data science
Shawn Rogers is an internationally recognised strategist, thought leader, speaker, and author specialising in analytics, business intelligence, Big Data, Cloud, IoT and social media technologies. He has founded and sold two Internet/media start-ups and is presently VP Analytic Strategy for TIBCO Software.
Traditionally, data science teams are full of brilliant people working solo, tapping into their own data sources, running things on behalf of a department, and not the entire business. But a transition is afoot, and it can be seen in the adoption of ModelOps, driven in part by the need for data science teams to be more sophisticated and grown-up, combined with the desire for better, reusable analytics frameworks, that leverage the power of a group versus the power of an individual.
Why now? Why focus on data? Analytics has proven itself to be recession-proof. If we hark back to the economic downturn of 2008, those companies that were able to find an extra couple of points of margin or customer satisfaction, generally had a better analytics strategy they could execute on. So, they survived because they knew the combination of data analytics and the ability to act on insights gave them a survival edge. Today we have a different downward trend based on the global pandemic. So analytics is once again being used as a survival tactic, but it needs to be better managed, and that is where ModelOps comes into play.
The notion of ModelOps is on everyone’s radar as a proficiency or a skill set needed to scale analytic practices. Companies face a challenge in that they may have the ability to build models but come undone when they need to deploy them, monitor them, test them for accuracy and performance, and move them off the workbench as a proof of concept and finally into production.
With ModelOps you coalesce or aggregate and bring together the data science teams and the models that come with them, to better test AI and ML-driven logic, automate decision making and ultimately be more competitive. If you want to scale AI and ML into the rest of the business and infuse it into other critical applications, you need to focus on centralising a lot more of these practices.
You cannot scale from a handful of models used in pockets of excellence to hundreds if not thousands of models without adopting an altogether different approach to modelling, design, and data science. This is where ModelOps becomes critical.
Every data scientist has a preferred tool, whether that is R or Python, which is why it is essential to view them as data artisans or artists. These are inventive people who like to use their imagination as much as they need to apply math to a problem. It is this combination of math, algorithms, and creativity that drives analytic practices forward and propels them quicker.
It is this sheer number of tools that are driving the need to bring together data science assets and move them into an Analytic Centre of Excellence (ACE). Instead of allowing data scientists to work in silos inside of departments or far-flung organisations, we must bring together the tools and the people, so we benefit from a community approach – and it really does take a community to drive data science and analytics.
An ACE allows us to scale in ways that we couldn’t before by eliminating redundancy and standardising our approach to analytics which is key to ModelOps and scaling analytics practices. There is no sense in having multiple data scientists working cross purpose in different departments but on similar projects – you need to be able to rinse and repeat. Take the picture they have painted and let someone else use it, no matter what paintbrush they used. This way, you save time, resources and money at the end of the day.
