Many companies have been waiting for years to finally reap the benefits of their multi-year AI investments, and the necessity to respond to the COVID-19 pandemic has provided the impetus. It has triggered many industries to improve their data management, operationalize and update their model production process, and reassure customers that their AI can be trusted. They are using MLOps, RPA, and the full array of tools in DataRobot’s enterprise AI platform to help their businesses thrive in 2021. 

MLOps (Sivan Metzger – Managing Director, MLOps & Governance at DataRobot)

In terms of value realization from AI, 2021 will be a breakout year where many companies finally start reaping the benefits of their multi-year investments in AI after so many years of growing expectations and fruitless investment. More important, firms will also start to better understand where their AI trajectory could or should be headed.

Two parallel trends will emerge: 

1) There will be rising pressure from businesses and CFOs to finally deliver measurable value from AI.

2) IT departments will take an increasing portion of ownership over AI-driven services and applications. 

IT and Operations are typically the groups that own and are accountable for all services running across the company, making them the best-equipped to establish production-grade processes for managing mission-critical services and solutions. 

Once this begins to happen, data scientists will not only start seeing their work come to life, but they will also find that these trends actually free up a lot of their time, which can now be spent on the next wave of data problems. Data scientists will also see much-sought-after actual results, which will help them understand how to adjust and modify their work to serve their businesses even better in the future.

Trusted AI (Ted Kwartler, VP Trusted AI at DataRobot)

In 2021, the trusted and responsible use of AI will become embedded throughout the AI pipeline, not just in a silo or as an afterthought. Trusted functionality, including the process, people, and technology, will be implemented throughout the data-to-value pipeline. In the early experimental machine learning days, companies used to ignore trust altogether. The ensuing, infamous AI missteps in the private and public sectors, such as résumé parsing bias or predicting educational outcomes, led data scientists to think more broadly about trust and bias.  

Data-driven organizations will employ trust to protect against missteps and harm to their customers, employees, brand, and other societal stakeholders, and two facets of trust in 2021 will become commonplace: model evaluation methods and debiasing techniques.

First, robust, formalized model evaluation—beyond KPIs like Logloss or RMSE—will be implemented. Some regulated industries have model risk management groups already in place which review compliance documents or ensure that certifications like ISO are secured. These groups will expand their scope and become implemented in unregulated uses, recognizing that the potential risk-reward-benefit ratio of AI must be understood holistically. Organizations will start to employ a more thorough review of models from a diverse and inclusive perspective. Solving model evaluations in an interdisciplinary manner entails a set of agreed-upon functionality and roles. Trust can only be solved using both technology and people.  

Additionally, a continued focus on bias mitigation will grow in popularity. Data scientists will expand their knowledge of mathematical bias definitions, learning when and how to choose the correct bias evaluation. Organizational leaders will take steps to ensure that bias—particularly relating to race and gender—is understood, documented, and mitigated before any model goes into production. To ensure that the model is behaving as desired once it’s in production, bias will be tracked as data drifts and historical patterns may no longer hold the same significance. Thus, the injection of fairness measures into MLOps will become paramount.      

RPA (Andrew Pellegrino, Director, Intelligent Automation at DataRobot)

In the coming year, traditional automation will see vast opportunities for growth with the addition of machine learning and data science. With machine learning, organizations are able to make high-value predictions to help them act faster through improved efficiencies. Building and then deploying machine learning models into production allows automation teams to focus on high-visibility use cases and build more end-to-end solutions that deliver greater AI-driven ROI. 

Machine learning will also empower automation teams to train their people to become citizen data scientists, helping further the ability of teams to handle more complex use cases. Overall, machine learning is poised to become the main driver for intelligent automation and is going to be a primary focus in 2021.

Product (Richard Tomlinson, Sr. Director, Product Marketing at DataRobot)

The COVID-19 pandemic has transformed how we think about predictive modeling and has put more pressure on businesses to demonstrate real value from their models. The era of doing AI projects for the sake of AI is over. AI projects need to deliver results with actual returns—and fast. Next year, within the enterprise, we finally expect a wholesale move from “Experimental AI” to “Operational AI,” as organizations must move out of the lab and beyond pure experimentation. 

Machine learning models and AI applications need to be operationalized, trustworthy, and ready to deliver value across all of the AI lifecycle. Business leaders and executives must educate their organizations, focus on the most feasible and high-value use cases to build momentum and maximize their chances of success, while giving guidance regarding the cases they should start with.

Continuing on the theme of enterprise AI value, in 2021 we also expect budgets to be consolidated, as organizations will be looking to minimize the number of AI software vendors they deal with. The market has moved from point solutions and towards full solutions with end-to-end value. It is no longer acceptable or even feasible to have multiple disparate products solving multiple disparate problems. The automation and democratization of data science is necessary to keep up with demand and competitive pressure. Despite increasing investments in data science talent and technologies, few companies actually deploy their AI into production.

We also know that, now more than ever, models become stale and need to be constantly retrained or replaced. This further reinforces the demand for end-to-end AI platforms that can continuously learn and automatically adjust to constantly changing conditions. From an operational perspective, teams need to trust the AI that’s driving their business decisions. Data scientists need to defend their models and need tools to tell a story that business leaders and mere mortals can actually understand.

Conclusion

As we look at the future of AI, 2021 is shaping up to be a make-or-break year. Models need to be functional, deliver real, AI-driven ROI, and do more with less data. For industries working to  navigate the post-COVID landscape, avoid AI bias, and deliver trusted AI to customers, the right tools are essential. The time to democratize and deploy robust AI solutions with RPA, MLOps, and Trusted AI is now. 

About the author

Richard Tomlinson

Sr Director, Product Marketing, DataRobot

He works closely with product, marketing, and sales teams to drive adoption and enablement of data management and data engineering capabilities in the DataRobot AI platform. Richard has been working in the data warehouse, BI and analytics space for over 20 years with the last eight years focused on Hadoop and cloud platforms. He is based in Chicago but is originally from the UK and has a degree in statistics from the London School of Economics.

Meet Richard Tomlinson