What If? Optimizing Scenario Planning with AI&ML

Article
By
MathCo Team
March 14, 2022 9 minute read

Demand patterns, supply chains, and logistics – not only did the past couple of years highlight pressing gaps in global networks, but also exposed the fragility of traditional forecasting and planning tools. Alongside dealing with short-term operational changes necessitated by the pandemic, such as workforce planning, inventory reconfigurations, and transportation constraints, businesses have also had to examine viable methods of ensuring agility and resilience in the longer term.

Creating and iteratively modeling a range of options to understand the impacts of business decisions has become the need of the hour, and, as a direct result, what-if analyses and scenario modeling are now increasingly being used to inform a wide range of business decisions. [1]

The problem with traditional scenario planning methods

Interestingly, several large organizations continue to rely on spreadsheets for scenario planning, leading to multiple assets that must be tracked and maintained. While spreadsheets are known to be an effective and speedy means of conducting basic what-if analyses, they are typically used when a hypothesis needs to be tested quickly, when data volume and scenario complexity are low, when documentation on the model is available, and when businesses have the know-how to address the risks created by this method. [1]

However, multiple spreadsheets to create, edit, and update makes data governance and traceability challenging. Further, with manually listing assumptions, copy-pasting data from various sources, and other time-consuming tasks, data complexity, reproducibility, and explainability become major limitations for stakeholders. Considering the shift toward automation in large businesses, the specific focus on scenario simulation has similarly shifted towards using analytics iteratively and deriving the best insights from available data. [3]

The scenario planning methods creating an impact

While qualitative scenario analyses are commonly used across industries, AI-based simulations that quantitatively explore the causal relationships between drivers and develop contingent plans of action have only recently come to prominence following the pandemic. [2]

Moreover, model-based AI — which uses available data as opposed to historical data — saw rapid adoption in the past year. According to the World Economic Forum (WEF), as more data was generated, data-rich and model-rich approaches were combined, creating hybrid solutions. The WEF further recommended focusing on capturing the inter-relationships of multiple domains (e.g., demand, production, supply, finance) as well as agile data science methods that account for the speed, urgency, and uncertainty of decision-making. [3] Two scenario planning types — system dynamic and agent-based models — allow for this:

System dynamic (SD) models:

These models can be applied to a range of scenarios, including global markets and economies, climate change, product and service life cycles, and supply chain and inventory management [3] to derive insights into business performance. With stock and flow sequences being the two key inputs for this model, decision-makers can get holistic insights on business dynamics as well as how cause-and-effect sequences occur within systems to inform business performance over time, [4] for instance, with the impact of purchases (flow) on inventory (stock) in a given period. This method involves intuitive, graphical representations of models — making it easier to use and understand than complex spreadsheets — and also allows for quality control and real-life insights, [3] enabling truly informed decisions.

Agent-based models (ABMs):

These models follow a bottom-up approach — the actions of individual agents within systems are examined and mapped to effects on the larger system, for instance, in terms of modeling the impact of human behavior on COVID-19 transmission in communities. In moving from simple theoretical models to representations involving real-world data, ABMs are significantly aiding scenario planning and can be used to examine a range of factors, from the impact of harvesting and wood prices on furniture markets to the economic impacts of climate change in the remote Canadian north. [5] ABMs are set to become useful techniques in post-pandemic scenario planning, with use cases ranging from traffic optimization and energy consumption to land use and economic recovery studies.

What if: The advantages

Scenario analyses have become a prerequisite to informed decision-making on contingencies, including supply chain operations, demand planning, and workforce planning. They allow for sensitivity analyses, drill downs, and thorough examinations of data to explore how different scenarios can impact businesses in the short-, medium-, and long term. For instance, a business user can use what-if scenario analyses to examine how project/timelines can be impacted by minute factors such as supply shortages or a fractional increase in labor costs.

In addition to providing specific and detailed views on business information and areas of improvement, what-if analyses also present a range of overarching benefits, including the following:

  • Creating, in combination with data visualization and storytelling techniques, interactive visualizations where inputs and scenarios can be easily changed to see how this affects the model
  • Accounting for variables rather than rows and columns, making data more accessible and context-relevant, as well as factoring uncertainty into the model by building it into variables
  • Considering all potential scenarios and iterating using AI to generate many more scenarios than possible manually
  • Enabling better collaboration and transparency by offering individuals clear insights into the impacts of granular changes for their business, giving them a wider range of insights to work with, and enabling them to view the various sources of data and add their own viewpoints.
  • Reducing human bias through evidence-based approaches. [6] When the models used to generate scenarios incorporate diverse causal perspectives from multiple industries/sectors and scenario generation is automated, the results obtained can be free of bias. Valuable perspectives that could potentially be dismissed by individual users can be highlighted using AI to bring hidden yet relevant business information to the forefront.
  • Improving exploration owing to the speed of insight generation made possible by what-if scenarios. Users can better explore new outcomes, and expansion capabilities, and make use of previously untapped data to make decisions faster.
  • Helping users envision and plan for future risks and opportunities, the impacts of possible policy decisions, and possible future events. With what-if analyses speeding up scenario generation, the duration of the planning cycle can be reduced, e.g., from yearly to quarterly, allowing for more immediate decision-making and responses to changing situations on the ground.
  • Alongside enabling businesses to identify KPIs and key drivers, what-if analyses can ensure both short-term resilience and long-term competitiveness, especially in the case of disruptive events. Mapping out the impact of changes in drivers on business and making information on different courses of action available to business users, actionability becomes a significant advantage here.

Let’s take a look at some use cases where what-if scenarios can be used to improve business decision-making:

1) Customer retention: If a company forecasts $14 million in sales for the upcoming year, based on a deal close rate of 35%, the impact of customer retention can be studied through what-if analyses. For instance, if the close rate is increased to 40%, the revenue projections can improve to $14.5 million.

2) Reducing risk: A Chief Risk Officer (CRO) can use a model that captures potential risks, such as demand shocks, as well as the causal links between these risks (poor advertisement reception leading to a drop in demand). In addition, the CRO can identify potential triggers, such as adverse weather events and drops in workforce numbers (including due to diseases and pandemics). With this information, a range of what-if scenarios can be applied to examine business impacts, for instance, weather events or diseases leading to absenteeism. High-impact changes, as well as surprises, can subsequently be identified using what-if analyses. The CRO can then refine these scenarios by generating additional hypotheses involving other risk factors.

3) Weather impact: Agricultural businesses can use what-if scenarios to examine the impact of weather on harvest quality. This can help them forecast sales as well as explore avenues for future investments, including in terms of infrastructure, damage prevention, and storage.

4) Hiring outcomes: Individual functions can examine the effect of new hires on revenue. For instance, if six people are added to the team over the next six months, what would the impact on revenue be? Such questions can be effectively explored using what-if analyses.

5) Shifts in demand: For an F&B business, tracking changes in consumer demand is essential to business decision-making. Using what-if analyses, newly emerging consumption patterns, such as those for vegan or meatless products, can be factored in to examine the impact on the supply chain and sales, thus informing future production needs.

What-if analyses present actionable insights across functions, helping businesses make informed decisions in the short- and long-term, pivot effectively in the aftermath of disruptions to explore a range of scenarios, and allow them to gain a comprehensive overview of key drivers impacting operations. Asking “what if” questions and making the best use of available data, therefore, becomes an essential starting point in ensuring both resilience as well as competitiveness, unlocking new business opportunities across a range of contexts.

Bibliography