Don't Predict, Decide: Why businesses need Intelligent Decision Systems.

“Effective decision making in today’s complex and disrupted business environments must be connected, contextual and continuous to drive good outcomes.” 
- Gartner, How to Make Better Business Decisions

As the Gartner report pointed out - and those who have embraced Machine Learning (ML) have learned - ML can be an essential tool for driving better business decisions in today’s complex environment. However, there’s a misconception that building a Machine Learning model alone is enough to achieve these results. It isn’t.

Machine Learning models are predictions of possible outcomes for any given scenario. Creating the continuous, connected, and contextual process that Gartner describes means putting the models in motion by letting Machine Learning models inform intelligent Decisioning Systems, which drive good outcomes. 

Successful Intelligent Decisioning Systems have several characteristics:

• Continuously learn to improve outcomes
• Contain one or more ML models
• Makes a decision or recommendation 
• Includes controls and transparency
  ‍

Decision Systems are continuously learning to improve outcomes.

A Machine Learning model without Decisioning System is just a snapshot of a moment in time. The models are static, based only on the initial snapshot of data. In order to make the models dynamic they must live in an intelligent Decision System that continuously pipes in results data, - teaching the model what it did right or wrong. This feedback loop between ML models and Decisioning Systems is where companies can get stuck. Teams spend a great deal of time and effort tweaking and tuning one model, hoping for an optimum state of predictability. When and If that model is finally deployed (a big IF), it rarely changes. It doesn’t learn and grow in real-time, if at all. This is no different than a decision tree of yore. Machine Learning Models that are continually tested and refined as part of intelligent Decisioning Systems achieve the best possible outcomes on an ongoing basis.

Decision Systems use one or more ML Models.

Good business decisions often have multiple goals, and sometimes those goals compete with one another. For multiple goals to be achieved, you need multiple models predicting the performance of each goal. For example, a Fintech wants to increase Loan Offer Acceptance from prospective customers. If they just built one ML model to optimize for that goal, it would likely lead to unprofitable loans and perhaps even increase charge-off rates. With a Decision System, they can consider multiple models - including one that optimizes for loan profitability. 

Decision Systems make a decision or recommendation  (ie. They made a choice)

So now that you have multiple goals and models, an action has to be taken. In Machine Learning, deciding the action to take, with competing models, is expressed as an Objective Function. This is usually a custom-coded piece of software that requires the developer to run the model(s) predictions for each decision option to determine the best choice in that scenario. In Intelligent Decision System - this this automated. The Objective functions are already included can even weigh goal importance to align with your business objectives. 

Decision Systems have Controls and Transparency.

With a tool as powerful as Machine Learning, controls are essential to making good decisions that conform with the business’s best practices or regulatory environment. Controls and transparency as to how the models behave are also needed to build trust with internal and external stakeholders that can be affected by the model’s results. Just as you wouldn’t want everyone driving in a car without a shared system of roads and rules, intelligent Decisioning Systems provide guardrails and a well-marked road to drive on. ML Models in and of themselves do not contain those controls. They are math algorithms delivering predictions agnostic to your business rules and restrictions. A successful Decisioning System allows stakeholders to design for clearly understood optimal outcomes while staying within their business’s guidelines. 

The world's most successful companies benefit from Machine Learning by connecting and deploying them within intelligent Decision Systems. These systems go beyond single model creation by considering multiple models and feedback loops on ever-changing conditions while maintaining business controls. A model without a Decisioning System is like a Ferrari without wheels - loud and going nowhere.