Defining processes for how ML and AI models get created, deployed and managed may be just as important as building “one more model”
But without it, large scale transformational projects like ML and AI adoption are doomed to fail. In conjunction with our customers, Open Data Group has seen time and again that defining the roles, responsibilities and processes supporting ML and AI adoption may be just as important as creating “one more model”. That’s why we’ve developed the Model Development Life Cycle to support teams as they transform themselves, and their businesses, with ML and AI.
Let’s ask a simple question of the business leadership: who in your organization owns an ML model in production? We’ve seen this question answered with “Data Science”, “IT” and even worse (and maybe more honestly) “I don’t know”. Companies are building applications, driving important business outcomes like customer retention, upsell offers, credit approvals and risk decisions, much of the time without clear roles defined in supporting underlying ML and AI models. This has to change - as business become more dependent on advanced models to make decisions, teams need clear ownership guidelines to ensure repeatable accountable success. The MDLC can help.
One of the most important values of a Model Development Life Cycle approach comes from surfacing what needs to change, and when. Simple example: production data is not the same as development data - teams need to plan for this type of change as the model moves to production. While many examples can be considered, a well planned life cycle can bring understanding of expected change, and allow management of that change to ensuring smooth operations for the teams involved. Learn more from our recent webinar here.
A Model Development Life Cycle helps teams understand the critical integrations and touch points between technologies used for digital transformation. Teams can consider which tools are critical path for production, and how best to create processes and hand-offs to help manage these critical paths. Surfacing these connection points allows important adjacent teams to the data science pipeline like security and governance to have a seat at the table and help plan and support production success. This whitepaper takes a look at the connections from the “create” side of the house - data science.