Free will is the core of a human being. Free will implies making decisions. But how are decisions defined? Decision making is choosing an alternative from many possibilities. It is a process that results in the selection of a set of actions among several alternative scenarios.
But what about business? What types of decisions are there, and what helps us to decide? We have business rules and other control logic to guide us to select the best choice for the best action. But that does not mean the decisions are easy to make. And in the age of digitisation, we have to make decisions all the time.
So what is the extent to which we can outsource the power of decision making? And is that even possible? The answer lies in automation.
Decision making as a problem-solving activity
First, I would like to point out two contradictory decision scenarios: the engineering and the business perspective. From an engineering standpoint, decision making is a problem-solving activity to identify and analyse the available set of actions, then determine the most appropriate option given the existing/expected conditions and constraints.
From a business perspective, increasing customer expectations and easily changing perceptions, unforeseen competition from disruptive players (which didn’t exist a few years ago), and operating in new business models are all bringing new challenges to the entire value chain. Organisations have to accommodate that complexity and speed of change without sacrificing productivity, quality and timely delivery. It’s all a matter of coping with the speed of change while remaining competitive.
Both perspectives understand that they can automate decisions. In the business world, we call this decision management.
Decision management systems capture, automate and govern frequent and repeatable business decisions. In a digital economy, business decisions – such as those related to the increasing focus on delivering great customer experience at all points of contact (omnichannel campaigns, special offers, dynamic pricing, buy online and pick up anywhere, etc.) – are subject to frequent changes and have a short life cycle.
The SAS standpoint
There are several high-level capabilities you need for performing intelligent decisioning:
- Diagnosis and identification of the situation.
- Application of business rules.
- Defining viable actions with AI and machine learning models and other decision logic elements.
- Determination of the optimal or best action set – or at least an acceptable task to resolve the situation.
But intelligent decisioning requires even more. You must be able to trigger the determined action set and capture the actual results. And, finally, you need to ensure traceability, auditability and explainability of the entire decision-making process.
Diagnosing and identifying the situation
Let us begin with data collection. The key is to gather as much data as possible. In manufacturing, for example, you need to get data from each machine or production element on the shop floor. This includes operational and execution information like the arrival of materials, start times of each programmed task, tooling used, task progression, stop time, completion time and, finally, departure of the finished product.
You would also need to collect sensor information from tools, including machine temperature, vibration, tension, stress and energy consumption. Next, you stream all this data for processing and analysis.Intelligent decisioning requires capabilities for diagnosis and situation identification, business rules, AI and machine learning models, and other decision logic elements. Click To Tweet
Detecting critical events
Within the stream, the system normalises or standardises the data to prepare it by filtering, combining and aggregating. Then we can assess and analyse the data by applying advanced analytics, AI and ML models to identify and infer the situation. From there, critical event detection and prediction is all done at subsecond speed.
If a critical event is found, that information is sent to the decision engine. The system also stores model information, including input data, past data (lag size), model details and outcomes. And all this happens while analysis and processing continue.
Taking a decision
The decision is the result of a decision flow in which the system applies a set of business rules and models (advanced analytics, AI, ML) to evaluate event criticality and impacts. Within those decision flows, the system can call in other data sources – for example, a knowledge base – to assist in the decision-making process. The input data for the decision flow should be stored, as well as all the relevant input and output information of each decision step, including model details and outcome. This is all linked to the triggering events to ensure explainability.
So, now that the best decision was determined, this step is about what to do. Thus, it’s the time to automatically trigger the execution of the appropriate protocol, a set of actions or tasks meant to address or resolve the detected situation. All the relevant information should be recorded to ensure explainability.
Capturing the actual outcome of the applied protocol is key to monitor the decision flow relevance and fitness, and for refeeding process autonomy.
The abovementioned pace of change requires more and more complex decisions to be made in short time frames. Some of those decisions could even be required to be made at the event moment – for example, continuous quality control/assurance will reduce production defects and rework or product return.
Thus, organisations must evolve from automated to autonomous processes to improve the efficiency and flexibility of their operations. They should take advantage of all available data (coming from customers, suppliers, enterprise IT and OT systems, production equipment, logistics, and many other internal and/or external data sources), embed AI/ML models in the decision-making processes, and intelligently drive automated decisions and actions.