I- Decision-Making AI: From Predictive to Prescriptive

A. Multiple AI, not just one

The use of Artificial Intelligence (AI) in the supply chain is multifaceted. It can perform functions such as forecasting, simulation, alerting, or decision-making. This last function, decision-making AI, is the least common but has the most significant impact on operational performance, both economically and environmentally.

However, all these forms of AI are complementary and can feed into each other. For example, a high probability of stock storage calculated based on historical data can enhance a production planning decision by integrating this risk. Serving different purposes and objectives, the implementation of these AI can be prioritized based on your maturity and challenges (see Part II).

B. Prescribing an Action Plan rather than Predicting a Scenario

Most AI models are based on more or less sophisticated learning techniques, often related to Machine Learning. They are used to process, query, or simulate historical data to describe or diagnose a past situation or predict a future one.

Unlike these commonly associated approaches, decision-making AI falls under prescriptive analytics. It takes those insights to the next level by suggesting the best way to manage a future situation. Based on mathematical optimization technologies, its output is not an estimate or scenario but a concrete, directly implementable action plan.

C. Optimization Solvers: How do they work?

In practice, prescriptive decision-making AI takes the form of optimization solvers, which can be described as software engines for solving complex business problems. Their operation can be illustrated by the data they input, the solution they output, and the constraints and objectives modeled within the engine.

Since they do not need historical data to work effectively, optimization solvers require the least amount of data — only the data strictly necessary for solving the problem:

• Input including a list of tasks to be accomplished and available resources.
• Output producing an optimal planning of allocation of tasks to resources, in other words “who does what, when, and in what order.”

Within the solver, a digital twin of your problem is integrated, taking the form of a mathematical model of its parameters. This model, customizable as desired, can then be solved by powerful algorithms producing the optimal solution in seconds.

Thus, an optimization solver is a genuine decision-support tool. With each variation in the problem (data, constraint, objective, parameter, etc.), it can generate the optimal solution among billions of possibilities.

D. Applications of Optimization Solvers in the Supply Chain

The applications of optimization solvers in the supply chain are numerous, at each of its stages independently, but also (and especially) in the synchronization of the supply chain within a systemic optimization logic.

Here are some use cases for solvers in the Supply Chain:

• Transportation and distribution routing plan (RouteSolver)
• Palletization and 3D loading of trucks/containers (PackSolver)
• Picking, Batching, and Slotting in warehouses (PickSolver)
• Production scheduling (PlanSolver)

Optimization solvers’ exceptional configurability makes them as powerful as they are modular. Depending on the input data transmitted, the same solver can be used for strategic, tactical, and operational decision-making.

For instance, Atoptima was approached by the shipping company CMA CGM and its branch CEVA Logistics to design its deployment plan for a fleet of zero-emission heavy trucks (electric and hydrogen). The decision-support tool deployed enabled:

• At the strategic level, to optimize the positioning of additional charging stations.
• At the tactical level, to position the zero-emission fleet across various warehouses.
• At the operational level, to plan driver routes by synchronizing their breaks with vehicle recharge times.

II- How to Integrate Decision-Making AI in the Supply Chain

A. Which AI to Choose for Decision Support?

AI based on learning logic and decision-making AI are totally complementary. They can be used separately or together, the former enhancing the model of the latter, for example. The choice mainly depends on the application. While those based on learning mainly serve to “relieve” humans of redundant and low-added-value tasks, decision-making AI — embodied by optimization solvers — is particularly well-suited for generating solutions for complex and structuring decisions.

In particular, there are four main differences between mathematical optimization (on which Atoptima’s optimization solvers are based) and Machine Learning (commonly associated with learning logics). Therefore, when choosing to use AI to enhance your decision-making, it is fundamental to question the underlying technology.

B. What are the Criteria for a Good Optimization Solver?

Once the type of AI is chosen, the moment comes to select the tool. Here, three fundamental criteria for selecting an optimization solver can be retained.

Functionalities

The range of functional decisions that a solver can account for not only determines its ability to evolve with your changing constraints and objectives but especially its ability to tackle your actual problem comprehensively, and not a simplified version of it.

Should I set some decisions in advance before submitting the problem to the solver? Should I divide the scope of my problem into multiple sub-problems (by area, resource, period)? Should I make approximations on key problem data? Should I modify the solver’s output in post-optimization? If the answer to some of these questions is “yes,” you’d be better off continuing your search for a tool that can handle all these realities in an all-in-one solver, as the productivity gains will be significantly higher.

Adaptability

Firstly, adaptability in modeling your issue before deploying the solver. This point relates to the capacity to model your real problem, with all its nuances (hard vs. soft constraints, for example) and in a global optimization logic (integrating decisions on palletization, loading, and routing into an all-in-one solver, for example). Adaptability also appreciates the ability of the same solver to handle different decision levels: strategic, tactical, and operational.

Secondly, adaptability when using the solver, mainly due to the flexibility with which it can be parameterized. It is in this sense that a solver can truly be a decision-making tool at the service of human. The full power of optimization intelligence lies in a paradigm shift in the way the solver is set up and mastered. This involves moving from a mode where ‘I want the solver to produce abc, so I set xyz’ to a mode where ‘I want the best solution to achieve my objective abc while respecting my constraints xyz’. In other words, moving from a medium paradigm to a fine paradigm.

Performance

Performance is measured in two ways: the quality of the solution and computation times. The quality of the solution is assessed by its ability to produce feasible solutions that respect all constraints and, of course, at lower cost. Even on “simple” problems, significant differences can exist in the cost associated with the solution produced by different solvers.

The concept of computation time is associated with scalability. In combinatorial optimization, the difficulty increases exponentially with the size of the problem. On large data sets, highly advanced algorithmic strategies will be necessary to maintain reasonable computation times.

C. Step-by-Step: Key Steps in Implementing Decision-Making AI

1) Identify the use case where decision-making AI will have the most impact. 2) Standardize the data strictly necessary for optimization (list of tasks and resources). 3) Free up decisions fixed by hand to allow the creation of synergies through decision-making AI. 4) Integrate an optimization solver according to the criteria above. 5) Achieve productivity gains of around 30%.

Implementing decision-making AI in your supply chain can be done progressively, without immediate software integration. A possible first step is to conduct a study where Atoptima realizes the mathematical modeling of your problem and runs its solvers internally to produce solutions. Then, the tool can be deployed in your organization and enhanced by integrating more and more decision levels to leverage the full potential of optimization intelligence through a systemic approach.

If you have resources (whatever they may be—vehicles, containers, production lines, humans) to dispatch and plan, then decision-making AI will undoubtedly add value to your operational performance. The economic and environmental interests are aligned, and such decision-support tools can even fully alleviate the additional costs of environmental transition.

Expert in mathematical optimization for +25 years, Atoptima stands out for its performance and scalability with tools 40 times faster than market solvers. It tackles the most complex business constraints, where no other alternatives exist.