Role of Decision Science In Digital Transformation

6 min readFeb 12, 2022

Across industries globally, Digital transformation is a major initiative undertaken to create new / transform existing business processes for better customer & stakeholder experiences, driving increased revenue and cost optimization. Decision sciences is a key component of Digital Transformation, enabling better decisions in every aspect of business execution — whether it be pricing, promotions, recommendations, cross/ up sell etc.

While there is a lot of active discussion on Data Science, there is increasing need of Decision Science both for business consulting and digital transformation. We try to distinguish these two, and also see how they complement each other. We also explore some well-known approaches and relevant Python based packages available that simplifies exploring these models in detail for business requirements.

What is Decision Science?

Businesses have a lot of decisions to take, whether strategic (where do I open my distribution center, should I enter this market etc.) or tactical (how many should I staff in my store, quantity to order from my vendor, can I approve this loan etc.). Decision science is about automating and augmenting decision-making process for Business users. It helps them to understand the risks, uncertainties, costs, benefits and compare alternatives to make informed decisions.

Data science vs Decision Science

The last two decades has seen a tremendous advance in the field of AI, Machine Learning and Data science.

Data science, as implied in the name, focuses on leveraging data into making better products. Data scientists are looking to analyzing data, visualizing and extracting insights through statistical models and algorithms. Data scientists take a statistical view of data in terms of quality, correlation, patterns etc.

From lineage, Decision science has evolved from Operations Research and management science. Decision science has placed more emphasis on Business problem, and hence business context on top of data is key. Decision scientists build upon the models and analysis from Data scientists, adding business context and flavor into recommending actions. More often, business is dynamic and there are many inter-relationships between various actors in ecosystem, making decision making complex. For example, price reduction can help in boosting sales, but competitors can choose to react, resulting in lesser than expected sales uplift. Decision sciences delves into such complexities, helping business users to factor in uncertainties and dynamics.

In highly competitive e-commerce industry, while Data Science focuses on predicting a customer order is going to be on time or not, Decision science helps identifying the right option for fulfillment (right Distribution center or store) to ensure customer order is delivered on time at lower costs. In media, Data Scientists focus on predicting what a customer might watch, Decision Scientists are involved in experimenting and recommending programs to offer to customers.

Decision Science Approaches

Decision science uses a number of mathematical models, a few of which are listed below. There are several use cases across industries like Retail, Banking, Healthcare, Travel etc. Many models have been built specific to each industry requirements.

Operations Research

Operations research (OR) uses mathematical modeling and optimization techniques to determine optimal or near-optimal solutions to complex decision-making problems. Emphasis is on framing the problem, reducing the decisions to mathematical model and identify optimal solutions for real life scenarios. OR has been used effectively to solve Production Planning & Scheduling problems in manufacturing, facility location problems in Supply Chain, Yield Management in Travel & Hospitality, Route Optimization in Logistics & Transportation, Portfolio management in finance etc.

Google has a rich suite of open-source tools at https://developers.google.com/optimization . Other niche packages available include CVXOPT for Convex Optimization (https://cvxopt.org/ ), Prodyn (https://github.com/yabata/prodyn ) for Dynamic Programming, PuLP (https://coin-or.github.io/pulp/ ) for LP modeling etc.

Systems Dynamics

Any Business organization or system doesn’t exist in isolation. Its typical behavioral pattern depends on the associated events and structures in the ecosystem. System dynamics (SD) is an approach to understanding the nonlinear behavior of complex systems over time using stocks, flows, internal feedback loops, table functions and time delays. SD has been applied to solve such complex problems as Inventory in a network, Population dynamics (growth, employment, urbanization etc.), Health (disease propagation, vaccination impact), Climate dynamics (global warming), Economic analysis (market growth) etc. SD works by breaking complex systems into individual subsystems, model causal relationships and use simulation to predict the dynamic behavior of the entire system.

PySD is a simple library for running Systems Dynamics models in Python https://pysd.readthedocs.io/en/master/index.html . There is BPTK-Py framework from Transentis available at https://www.transentis.com/simple-python-library-system-dynamics/en/

Process Mining

Process mining is defined as an analytical discipline for discovering, monitoring, and improving processes in reality. Enterprise systems capture event data or digital footprint as they happen across the enterprise. Process mining applies data mining and process analytics, to mine log data from these enterprise systems to understand the performance of the business processes. This can identify bottlenecks, sub-optimal performances, impact of quality rejections, benchmark similar entities etc. Process mining has seen a lot of success recently in helping organizations improve Customer service processes, Procure to Pay, Order to Cash, Employee hiring etc. Process mining is relatively young, and coupled with simulation has good potential to deliver agile and responsive processes.

PM4PY is a good Python library for Process Mining https://pm4py.fit.fraunhofer.de/

Discrete Event Simulation

Business process execution is about a sequence of events in time acted upon by multiple systems and stakeholders. Discrete event simulation applies computer-based simulation methods to model the ecosystem and predicting the performance of systems which are driven by events occurring at discrete instants in time. Some of the common applications in industries include modeling queue and operator performances like in customer arrivals in banks to be served by a teller, customer arrival and serving in retail checkout counters, order processing across machines in a manufacturing plant etc.

Simpy is a discrete event simulation framework based on Python https://simpy.readthedocs.io/en/latest/

Beyond these, there are several other types of models including Goal Programming, Decision Theory, Network Analysis etc.

The following are few top academic journals in this space for reference

· Operations Research (https://pubsonline.informs.org/journal/opre )

· Decision Sciences (https://onlinelibrary.wiley.com/journal/15405915 )

· Management Science (https://pubsonline.informs.org/journal/mnsc )

· European Journal of Operational Research (https://www.journals.elsevier.com/european-journal-of-operational-research )

· Annals of Operations Research (https://www.springer.com/journal/10479)

How can Organizations Leverage Decision Sciences?

Decision science has made a lot of progress over the years, increasingly solving/ predicting things that have been traditionally impossible or too complex. The methodology and approach depend on the business problem being solved. Decision scientists need to develop business contextual knowledge on top of analytical modeling capabilities.

1. Formulating the Problem — This involves laying out the details of the problem including the environment, actors/ systems, decisions to optimize, potential scenarios etc. A clear and adequate understanding and verbalization of the problem is key to ensure the success of the project.

2. Building the Model — This involves breaking the system into subsystems, modeling the subsystem parameters, transforming data into relationships and generating results. Collecting relevant data, validation and cleansing data is crucial for effective models.

3. Testing the Model — In this step, the researcher simulates various scenarios. Overall effectiveness of changes to control variables needs to be tested through sensitivity analysis. The usefulness of the model can be tested with limited proof of concepts in restricted environment, executing the business decisions based on model and comparing it as against current way of working. A/B testing is a well-known way to compare two versions of something to figure out which performs better.

4. Implementing the Model — This phase involves abstracting the model and deploying it in production. Cloud technologies, API and containerization enable consumption of the models through Digital technologies (Low Code systems, Business Intelligence, mobile apps, RPA etc.) for Business users. A critical feature is ability for “What -If” scenario modeling enabling business users to try and experiment. Change management is critical in ensuring business adoption.

Summary

With rapid adoption of digital business, every part of the organization needs to improve decision making at scale. The need today is to leverage the vast amount of data available internally and externally and make high quality decisions based on facts. Decision science compliments Business management in their quest to deliver better products and experience to customers, through a journey of experimentation and learning from impact of decisions.