3 point framework for Enterprises to lead a successful Decision Automation vision
Most people agree that algorithms for self-driving cars are sophisticated, and once they are developed, the algorithms might have a universal appeal. The Big Techs have made significant advances in algorithms and the industry sees decision automation as the next big opportunity. As the enterprises embark on the journey, they are beginning to realize that automated decision making in the enterprise could be far more complicated than self-driving cars.
This article draws parallels between self-driving cars and autonomous decision making in an enterprise. The goal is to provide a framework for enterprises to learn from a rapidly growing industry and to identify the challenges upfront.
The autonomous car algorithm considers obstacles, signals, STOP signs, distance between objects and more. In comparison, autonomous decision making in the organization has higher number of variables and more importantly, the signals continuously vary.
The challenges in automated decision making are two folds:
(a) Planned (eg. known STOP sign for self-driving car analogy)
(b) Unplanned (a pedestrian or basketball landing in front of a car)
3 point Framework for Autonomous Decision
Planned and unplanned decisions are fundamentally different. The algorithm for self-driving car automatically makes a decision to stop for a pedestrian (react quickly, unplanned) or stop for a STOP sign (planned and proactive act). In this example, a self-driving system:
- First, read the pedestrian and processed the signal immediately,
- Then, ran several calculations and scenarios to figure out how to best avoid hitting the pedestrian (deflating the air bag versus slow stop versus sudden stop) and,
- Finally, took the action to stop.
Drawing parallels, for businesses to take a decision quickly, they would require to:
- Get information and process it rapidly,
- Analyze scenarios and understand how it impacts for both planned and unplanned events,
- Then, take a decision based on most appropriate scenario
There are many solutions in the enterprise to address point (1) and present the data in a nice visualization like Tableau or Power BI. Similarly, there are integrations to transactional systems to address point (3) for example, connectors to ERP systems.
The real challenge is to address point (2). Can there be a system where a decision maker evaluates all the possible outcomes and takes the next best action? How do we take actions for something which is not predefined? Can the machines learn from history to recognize a pattern, simulate impact for various actions and then take (or at least recommend) an action?
Items to consider upfront
It is important to remember that business process automation may not lead to decision automation. Conventional process automation systems are rules based and they can work on predefined conditions. In reality, many business decisions require quick, often unplanned actions where the outcome is uncertain or there is a range of possible outcomes. In an ideal case, an autonomous system can learn from these actions over a period of time, and recommend the next best action to the decision-maker.
Every organization has underlying challenges — such as, quality of data for analysis, different industry customizations, unknown interdependencies, regulations — whose knowledge typically resides across different functions (IT, data scientists, business, engineering, legal). Therefore, for successful decision automation, technology enablers must be intuitive and accessible to diverse teams for collecting inputs and ongoing operations.
Conclusions
Decision-making with business uncertainties and the range of possible outcomes is the biggest challenge for every industry. It took several years for autonomous vehicles to deal with uncertainties and become somewhat accepted. The time for autonomous decisions in enterprises has just begun.
In order to succeed, the solution needs to take contextual inputs from different stakeholders in a centralized system and learn to quickly react to the events, similar to how the sensors of self-driving vehicles are connected to a control plane for algorithms to take inputs and perform actions.
