What to do with all that machine learning data
For many Fortune 500 companies, Decision Optimization (DO) is the technology of choice to convert Machine Learning data to better, faster, smarter decisions. In his book “The Optimization Edge,” Steve Sashihara shows how companies including Walmart, Google, Amazon, UPS, Marriot, and McDonalds secure their leadership position by incorporating Decision Optimization into their businesses. But you don’t have to be a Fortune 500 company to combine Machine Learning (ML) and Decision Optimization, and thereby move up the leadership chain.
Recently, I met two clients at an industry show. Each had in-house data science teams with deep expertise in either Machine Learning or Decision Optimization. Each had a fair amount of success applying these technologies in isolation, as did many of their competitors. They asked me how combining Machine Learning and optimization can help them gain the competitive edge, and especially how all the newly-learned data can improve their decision models. To answer this question, I’ll first take a step back.
How does the interplay between ML and Decision Optimization typically work today in businesses such as these two clients? ML models are used to generate better, more accurate data — for example, a demand forecast gets better when it not only considers historic numbers, but also learns from weather and social networking data. Decision Optimization models then ingests those ML-generated forecasts, and automatically recommend which actions to take — for example, which products to produce, as well as where and when to produce them, to fill the demand. A single optimization model is capable of trading-off multiple business goals, millions of possible decisions, and millions of business constraints. The output of a Decision Optimization model is typically a set of recommended actions, such as price points, plans, and schedules.
The previous paragraph describes the baseline of how Machine Learning feeds into Decision Optimization to generate optimal actions. But that is something these two clients already knew. What else could they do?
In fact, where the flow between ML and optimization has traditionally been a one-way flow, in state-of-the-art decisioning solutions this becomes a cyclical process for even more benefit: once an optimization model has recommended an action plan and that plan is operationalized, the data on the execution of that plan can be used to once again learn how to improve the forecasts, how to automatically make the decision models more accurate, and how to better hedge against risks. For example, in an organization where a central model recommends decisions to a distributed team or satellite businesses, those distributed teams can provide feedback or onsite knowledge via an app, and that feedback can then combine with other unstructured data to further improve forecasts and to adapt decision-models in real-time.
In this way, not only do the forecasts “learn” and improve over time (a typical application of Machine Learning), but the decision optimization models learn and adapt in parallel: at the same time as ML techniques help businesses generate more accurate data, Decision Optimization helps businesses to use data to generate even better, faster, smarter decisions.
While the smart data, models, and algorithms are at the heart of such combined ML and Decision Optimization solutions, the line-of-business decisioning environment is the soul. With Decision Optimization applications line-of-business users can quickly generate and visualize plans and schedules in a collaborative environment. Hundreds of planners can collaborate on trade-offs, scenario analyses, and visualize business-wide impacts. Whether on-prem, cloud, or hybrid, IBM Decision Optimization will accelerate your decisioning processes for competitive advantage.
Learn more about IBM Decision Optimization and how it can add value to your business.