This paper has nothing to do with GenAI! A framework for approaching decision problems to maximize the chances of making it to production – and making an impact.

Take Our Content to Go

Prescriptive Analytics in Decision-Making Introduction

In a rapidly evolving business landscape, the role of advanced analytics has become pivotal. This whitepaper delves into the realm of “Prescriptive Analytics in Decision-Making,” illustrating how Mosaic Data Science harnesses this powerful tool to drive strategic business outcomes.

The Hype and Reality of GenAI

Let’s take a moment to step back from all the hype around GenAI. It’s not that GenAI tools aren’t fascinating or even mind-blowing; they are. Mosaic started talking about the potential for GenAI and LLMs back in 2018. It’s also not that GenAI tools aren’t useful, they (generally) are. However, everyone in the AI world knows we are in a hype cycle on our way to the “Peak of Inflated Expectations.” Avoiding some common temptations before we get to the top might be helpful.

Figure 1. 2023 Emerging Technologies Hype Cycle | Gartner, 2023

Temptation #1: Look at this shiny new tool. I want one!

Hint: This thinking is not a recipe for effective problem-solving.

While the hype is exciting, it is easy to fall into the temptation of being attracted to shiny objects. When we get lost in the glare of the solution, we often lose sight of what real problems are worth solving, like how we want to upgrade to the latest and greatest smartphone with every new release when the real problem is figuring out how we can better use the technology in our lives to support meaningful experiences. The more time we spend in the glare, the less time we spend making an impact. And the more time we spend in the glare, the more time we spend falling into the second temptation.

Temptation #2: This will solve all my (business) problems!

Hint: No, it won’t; it actually won’t solve most of them.

One obvious mistake we have seen time and time again is the over-extension of technology tools into problem domains they are ill-suited for. Tools have their use. Screwdrivers are used to drive screws, saws are used to cut wood, classification algorithms are used to classify, and GenAI models are used to generate content. If I said screwdrivers cut wood everyone would think I am nuts. Why? Because everyone understands what screwdrivers are and what they are for. A fancy way of saying that is that screwdrivers are in the “Plateau of Productivity” phase in their hype cycle.

Let’s face it, GenAI tools solve a specific subset of common business problems, but they do not solve all business problems. While the latest models excel at a great many multimodal generation tasks, they do not currently have the ability to recommend effective solutions to highly complex decision problems. These are problems that are characterized by:

  • Many possible solutions: Problems where the number of possible solutions is so mind-boggling that the decision-maker cannot find an optimal decision path on their own.
  • Highly constrained environment: Problems so constrained by external factors that it is extremely difficult even to find a feasible solution.
  • Conflicting objectives: Problems that require humans to weigh the tradeoffs between multiple, often conflicting, objectives.
  • High degree of uncertainty: Problems where following one path could lead to a domino effect resulting in sub-optimal states down the road.

Those types of problems can be classified as decision problems and require a toolset specifically aimed at prescribing explicit, feasible, and optimal recommendations to decision-makers.

Real-world problems often involve complexities and constraints beyond current AI model capabilities. Decision-making in business requires a nuanced approach, considering a myriad of solutions, constraints, and conflicting objectives under uncertainty.

Mosaic’s Framework: Prescriptive-First Analytics

In our engagements with customers, Mosaic often approaches problems with a prescriptive-first approach. What does this mean? Many are familiar with another famous chart from Gartner outlining the maturity curve in analytics organizations. Predictive modeling is used to make predictions about unknown values (e.g., how long it will take to drive home along my typical route). Prescriptive modeling is used to make recommendations (e.g., which route to take home). 

Figure 2. Analytics Maturity Curve | Gartner

This visual implies that organizations follow the maturity curve linearly, with most organizations currently building solutions at most up to the predictive analytics space. Prescriptive analytics in decision-making, for many, feels like a far-off possibility, not least because they see so many of their predictive analytics efforts fail. Why do they fail? In large part because predictive models are not tied directly to the actual decision problems that they are intended to address. Teams build predictive models that fail because the actual decision problem that they are there to support is not fully understood – they are accurately predicting the wrong thing, or predicting it at the wrong time, or without any clear path for a decision maker to understand and use the predictive outputs.

Prescriptive-first analytics approaches problems by first looking through the prescriptive lens and using predictive models to quantify the decision process directly. We approach problems in this way because we have seen, across industries, that following the analytics maturity curve linearly often results in predictive models that do not take adequate regard for the ultimate decisions that need to be made.

What does our approach look like?

Deep Understanding

  • Our approach requires the technical team to dive deep into the decision problem faced by the business before any other activity. At this point in the process, the focus is not on reducing the problem to fit a specific modeling framework but on truly understanding the problem. Here, we ask and answer questions to understand the underlying business problem and decision context, such as: 
  • What are the objectives?
  • How will the success or effectiveness of a decision be determined?
  • How urgent or fast-paced is the decision-making process?
  • What downstream or other processes are impacted by the decision?
  • How do you look at tradeoffs between conflicting objectives? Do different business stakeholders value objectives in different ways?
  • What complexities make this problem difficult for decision-makers to solve on their own?
  • What data and other supporting information need to be considered when weighing options?
  • What would be the benefit of having better/faster recommendations?
  • What value is there in having the best possible solution vs a really good solution?

Formulating Prescriptive Models

At this point a prescriptive model can begin to be formulated. Time spent in formulation is key, because it is the starting point for translating the human problem to a math problem. If the problem is well understood, then an experienced data science team can quickly decide on the ultimate prescriptive modeling approach (e.g., heuristics, mixed-integer programming, simulation, dynamic programming, reinforcement learning, etc.). However, at this stage of the game, heuristic approaches are very attractive for several reasons:

  1. They are easy to implement.
  2. They quickly give the decision-maker good recommendations without significant investment.
  3. During development, they help the team to understand the decision problem even more.
  4. They are flexible enough to allow the team to pivot as needed.
  5. They may be good enough, avoiding more complex solutions for only marginal gains.

Integrative Predictive Modeling

The effort to formulate a prescriptive modelwill be the backbone for all predictive modeling decisions. Predictive models are no longer seen as siloed solutions that may or may not help the decision-maker. Instead, they tie back directly to the decision model. From Mosaic’s perspective, predictive models serve as support to prescriptive models.

Case Study: Operational Optimization in Renewable Energy

Mosaic helped one of the world’s largest renewable energy producers kickstart their advanced analytics program by executing operational use cases. The customer approached Mosaic with a challenge they face in forecasting the actual production at each of their plants on a daily basis. As Mosaic began collaborating with the team, we uncovered the deeper challenge this forecasting problem fed into making optimal raw material supply decisions at each plant.

The production forecast problem was real, but the broader challenge for the customer was making daily raw material supply decisions to get the right type of materials from the right suppliers to the right plants at the right time for the lowest cost, a common challenge.

Prescriptive Analytics in Decision-Making Solution

As the team grew in their understanding of the broader challenges, a simple mathematical optimization model was formulated. The model minimized the cost of supply while meeting demand and respecting available supply constraints. Because this was a planning problem, each of those components was unknown in the future and needed to be predicted. With this formulation, the team had a clear framework by which to view all the other models that would need to be created to support the decision process.

  • Prescriptive Model: Supply Optimizer
  • Predictive Models:
    • Cost of Supply: How will the supply market prices shift in the near term?
    • Supply Availability: What will the availability of raw materials look like in the near term?
    • Inventory: How do we expect inventory draw down to behave?
    • Production: How much will the plant be able to produce? *The original problem

Customer Benefits

Taking this prescriptive-first approach, the customer realized a number of benefits that they would not have if the team treated the original predictive problem as a siloed use case. First, executives clearly understood what modeling activities were on the table. There were no surprises about additional models that would need to be developed in order to make the original model useful. Second, the project team had a common language to communicate about different model components. This helped facilitate the back and forth between the technical and business teams, minimizing the chances of requirements being lost in translation. Finally, each model had an obvious use that tied back directly to the decision process. Even if the customer had developed each model individually, the risk of siloed models would have been high. By taking a wholistic view of the decision process, the customer received the benefits of a wholistic solution

Conclusion: Prescriptive Analytics in Decision-Making

While we do a significant amount of GenAI/LLM work for our customers, we do so when it is the right tool for the job. We spend more time up front, truly understanding the problems our customers face to be prudent in what tools we pull out of the analytics toolbox.

To help us decide what tools to use, we try to approach every problem with a prescriptive mindset. This helps everyone to level set on the decision-making process that our solution will integrate into. This approach maximizes the chances that predictive and prescriptive models alike will move to production and truly help decision-makers solve the problems they are faced with.

“Prescriptive-first analytics” is more than a buzzword at Mosaic – it’s a philosophy that drives our approach to solving business problems. We leverage this analytical approach to offer businesses not just data insights, but a roadmap to achieving their strategic goals.

Embrace the power of prescriptive analytics in decision-making with Mosaic Data Science. Contact us to explore how our bespoke solutions can transform your decision-making and operational strategies.


  1. Gartner. (2023). What’s New in the 2023 Gartner Hype Cycle for Emerging Technologies. Retrieved from
  2. Gartner. (2018). Analytics maturity curve [Adapted illustration]. Retrieved from Adapted by [Mosaic Data Science].