Data scientists love to create models. They’re always finding new ways to manipulate data to uncover new insights. But their efforts are often wasted, as many models are never deployed. In fact, a recent ZDNET article highlights that only 22% of data scientists say the models they develop to enable processes or capabilities usually make it to deployment. This is a source of frustration for data scientists as well as stakeholders expecting a return on AI investments.
There are many reasons for this, including the lack of an analytics and data platform for AI that can support AI model development at scale. Another major issue is limited interaction among data scientists, business teams, and IT. Data scientists often operate in silos, narrowly focused on specific issues and datasets, without working across teams to find holistic, reusable solutions. As a result, many data scientists feel their perspectives are not sufficiently considered when it comes to planning, building, and deploying AI models, according to a Rexer Analytics survey.
Data scientists want to be the stars in the game of AI, but they can’t do it without teamwork. Take soccer legend Lionel Messi. Why do his goals always make it to the back of the net? It’s not all about creative moves. It’s his talent for leveraging other players’ strengths and his holistic awareness of the field that ultimately win the game.
So, what can data scientists do to get (and keep) their AI models in production? Be team players and focus on the bigger picture, not just innovation.
Step out of the sandbox
Get involved early. Get involved in the early stages of product design. Gain a more informed view of the overall business objectives and the operational constraints so you’ll be better equipped to deliver products that better fit actual use cases.
Focus on outcomes. Understand the real-world value and cost of your work. Use these insights to advise on the best way to deploy and how to maintain and support models in production. When you demonstrate a broader understanding, others are more likely to incorporate your ideas, concerns, and suggestions.
Listen to IT. Be respectful of the role of IT in protecting current systems and working with tight budgets (and the exploding cloud transformation costs they often face). Understand IT resource requirements and constraints.
Understand the output of AI models. Ask business users how models are used in practice to satisfy specific business objectives. If possible, experience this for yourself. Apply these insights to discover ways to improve the models or identify new data sources that can be integrated.
Participate in the whole lifecycle. Instead of simply handing over your innovations to the business to use and the IT team to support, know when to step back in to retrain, refresh, or retire models.
How IT and business teams can help
IT and business teams should make an effort to better understand data analytics and what data scientists do. Today, less than half of managers and decision-makers are knowledgeable enough to make informed decisions on AI deployments. While managers shouldn’t be expected to make these decisions alone, they should know enough to involve data scientists at the right points.
Second, business and IT teams can foster stronger relationships with analytics teams. Sometimes the business team acts like the customer and IT team like the gatekeeper, which creates barriers to communication. It’s important to have ongoing discussions with analytics teams to avoid making unrealistic demands or restricting resources without a full picture of what’s necessary.
For more on the role of business leaders and IT teams in driving successful AI deployment, check out the previous two articles in this series here and here.
Winning at AI
AI implementation is a complex and often challenging endeavor. Like winning in sports, succeeding at AI requires teamwork. This means close collaboration, open communication, and playing to each other’s strengths. Each player must take responsibility for their own expertise while understanding and valuing the roles of others. By building data-centric and team-oriented culture, companies can more quickly realize value from their investments in AI.
