Organizations make lots of decisions every day. Each is an opportunity to improve their bottom line, reduce fraud or improve lives.

Today, businesses are collecting more and more data they could use to make more data-driven decisions. Better decisions. But data-driven decisions aren’t keeping pace with the volumes of data being collected.

Why? Because it isn’t just about the data. It’s about learning something important from that data — deriving insights that can guide the thinking of a decision maker.

The Foundation of Effective Decisions

Think about it as requiring the three P’s:

  1. Process. For decisions to create value, somebody needs to act on them. Understanding the key elements and mapping out the decision process is a vital part of becoming more effective and repeatable.
  2. Prediction. Organizations need to take advantage of insights derived from predictive models to determine what is likely to happen next in order to influence the future.
  3. Policies. All organizations have formal or informal procedures, policies, regulations or even laws that govern how they operate – what some call “business rules.” And they all have resource constraints.

So, what gets in the way of that effective decision-making?

  • Data: Too much, not enough and an inability to identify patterns or trends
  • People: Too many touching the process, personal biases, politics, emotions, a schism between business and IT
  • Process: Poorly documented, ad hoc, too many handoffs, compounded by lack of standardization across the enterprise

Overcoming these issues requires a structured decision-making process.

Automating the process and minimizing human intervention with a decision engine based on the three P’s ensures the process will be consistent and repeatable.

Combining prediction and policies gives decision makers the confidence to act. All three P’s need to be integrated, working in harmony like the engine of a car, for maximum performance.

Decision Making Evolves Over Time

To illustrate the kind of decisions we’re talking about, let’s consider how marketing professionals might evolve their decision making for customer targeting.

Organizations are trying to personalize interactions with their customers, patients or citizens. They don’t treat them all the same. In order to facilitate that personalization, they try to categorize them into logical groups that should be handled similarly. High-value customers may be handled by a customer service representative, while low-value customers may be served by a kiosk or sent to the web.

First step: business rules

A common way to group customers is based on purchase behavior. Recency, frequency and monetary measures (RFM) are often used.

With RFM, each customer scores from 1 to 5 for each of the three measures. For example, those who purchased in the last 30 days score 5; those who purchased between 30 and 60 days ago receive a 4. That’s the recency part. Higher frequency of purchases in a certain time period earns more points. So does greater-than-average spending – the monetary measure. Customers with the highest combined RFM scores are deemed “high-value.”

In this case, the business rule defines the point at which the score changes: 30 days, 30 to 60 days, etc.

Business rules are statements of business logic that specify conditions to evaluate and actions to take if those conditions are satisfied. They come in IF, THEN, ELSE statements: “IF last purchase within 30 days, THEN assign Recency Score = 5.” Or “IF RFM score ≥ 12, THEN send offer, ELSE do not send offer.”

Business rules are typically step one for codifying the decision process.

Moving to models

As an organization matures and markets become more competitive, businesses usually evolve to use more precise, analytically driven models. Specifically, predictive models anticipate which customers are most likely to respond to what.

While these models often include RFM measures, they also incorporate many more pieces of information: demographics, lifestyle, location, income, product purchase history and more. The models could be derived using popular algorithms like regression and decision trees. Or they might involve more sophisticated machine learning algorithms, such as neural networks, support vector machines or random forest algorithms.

Ultimately, the combination of business rules or predictive analytics is more powerful and flexible than either alone, especially when merged into business processes for consistency and efficiency. This is the next step in the evolution of effective decision making.

What Kinds of Decisions Require Analytics?

Traditionally organizations used analytics to make strategic decisions: Should we acquire Company X? Where should we build our next retail store? While they arise infrequently, such decisions represent very high risk or very high reward.

Today, a transition is underway, as organizations turn to analytics for day-to-day decisions. Marketers are executing predictive models in call centers to identify the next best product to offer an inbound caller. Manufacturers are predicting equipment failures, scheduling routine maintenance prior to the failure.

As pressure mounts to eliminate waste, reduce errors or identify fraud more quickly, a systematic approach needs to be implemented for automating and improving high-volume operational decisions. Decision management is an emerging discipline that needs to incorporate the three P’s to realize these benefits.

Spanning Industries, Crossing Functions

Understanding the decision process is key to improving it. Here are a few examples of how organizations might combine policies and prediction within a process.

Marketing. Let’s go back to the RFM example. Business rules defined the RFM scores. But each segment likely behaves differently and we’ve identified the “best” predictive algorithm to predict buying behavior for each segment. So when a customer is mapped to a value segment through the RFM score, that determines which propensity model is applied.

decision making model

Hospital readmission risk mitigation. Readmissions after hospital discharges – a too-common occurrence – are costly for both hospitals and patients. Analytics can help.

First, a predictive model determines the probably of unplanned readmission for each patient. “Readmission” is defined as a patient readmitted to the hospital within 30 days of discharge. Predictors might include gender, demographics, comorbidities, laboratory results, medication orders and social history.

predictive model

The predictive model calculates a probability score for readmission of each patient. The business rule says that IF the probability is greater than 65 percent, THEN look at other attributes (also business rules) to determine the best action to mitigate readmission.

Resource optimization. Health care providers often devise intervention strategies or outreach programs they can offer patients to improve health outcomes. Some patients may be more responsive than others to these programs. And some interventions may be more effective with certain patients. Appropriately allocating limited resources to patients where they will be most effective is a challenge.

True mathematical optimization can point to the “best” use of resources, combining predictive models and policies. That means maximizing overall health outcomes across all patients, given probability of outreach effectiveness, while considering rules and constraints such as programs, costs and other factors.

SAS image

The Internet of Things

The evolving Internet of Things will provide lots of opportunity for automating and improving high-volume real-time operational decisions. There are endless possibilities.

  • Farm equipment manufacturers could adjust settings remotely on machinery, based on real-time operating conditions (temperature, humidity, soil moisture), for optimal performance in the field.
  • Roadways could reroute drivers based on an anticipated traffic jam.
  • Manufacturing plants could reschedule maintenance plans to optimize production.
  • Energy providers could trigger corrective actions in the smart grid to maintain grid reliability.

And it will all happen automatically as part of regular operations.

There is no shortage of executive surveys that highlight the need for better decision making within organizations. And the Internet of Things will only exacerbate the need for a more formal automated and systematic decision management process.

Organizations that incorporate the three P’s of effective decision making can realize the benefits of faster, more frequent, more accurate and consistent decisions — leading to reduced expenses and risks.

Title image "arrows" (CC BY 2.0) by  Dean Hochman