How machine learning can complement mathematical optimisation

Resource optimisations enables the use of resources (people, raw material, equipment, financial) needed to produce a product or service at an optimal level. This process is crucial to perform a task as efficiently as possible and is used in almost all organisations to improve operations. Understanding the mathematical optimal can oftentimes be a great solution, but in many cases machine learning (ML) can become a powerful complement to your optimisation strategies, regardless of your industry.

Why resource optimisation benefits from machine learning

Traditional optimisation through mathematical models is based on finding a mathematically optimal solution. This typically requires making assumptions about how the problem is framed, which make up the goal function and constraints in an optimisation model. To put this (very) simply, you have certain data, a target or a goal, and you frame the problem. The results depend on whether your data was sufficiently detailed and comprehensive enough, and whether you were been able to scope the task properly. However, once the complexity of the production process grows, machine learning can become a great complement to your optimisation models. As inherently in the term, machine learning learns from data. Here again, there needs to be enough good data, but because of the ability to learn from examples, you don’t have to know everything in advance, but you can adapt based on the data that comes in. In other words, you’re not merely optimising based on the rules devised from the data, but you may be able to find out new insights from the data with ML and adapt accordingly. Let’s take an example.

Getting closer to the truth with machine learning

Machine learning is well-suited for making estimates, predictions and suggestions for of complex, multi-faceted situations. Think about employee sick days. The 80% probability that 20% of your personnel will be on sick leave, will change your optimal plan completely. Without the visibility to the potential changed situation, your optimisation model will be useless. Obviously some of these rules might exist even without ML, but the difference is in the way the ML model has been trained and what it then finds in the data that is coming in.

80% probability that 20% of your personnel will be on sick leave will change your optimal plan.By enhancing your optimisation model with machine learning, you decrease the amount of uncertainty your model has to withstand and thus improve the planned optimal. The learning capacity ensures that you don’t need to know everything from the beginning, and you don’t get locked onto your early assumptions. Machine learning assures your assumptions will be closer to the truth. This strength can be applied to any unforeseen situations: sick days, service or raw material delays, or slowdowns in production or supply chain, but also making your services more personalised: creating dynamic pricing, personalised content and product recommendations.

Learn to react faster because you already know the future

When producing any service or product, the process is complex enough that one small change can amplify into a significant delay or decrease of quality. Your optimal plan might not be optimal, supply chain might run into hiccups, your employees might get sick. The way to tackle uncertainty is to develop your organisation’s skills to adapt faster and develop a new optimal plan according to the new situation. The ability to adapt is not only for humans: with machine learning, you’re embedding these skills to your technology stack as well. When starting to incorporate machine learning into your optimisation strategies, there are three things to keep in mind:

1. Start by enhancing your existing optimisation models with ML

Nothing disruptive will be fast and smooth. If you want changes that will be adopted to the current workflows, the machine learning solution should become part of your existing optimisation strategies, potentially as one new input that helps your optimal model derive the best plan.

2. Be one step ahead of your own process but also your competitors

Improving the estimates by taking into account machine learning driven probabilities and predictions helps you devise a new plan and withstand change. As with any new technology and machine learning model, learning is constant – get started among the first, and be always one step ahead on your learning curve.

3. Start from a business problem and start today

Finding the best model and the most impactful opportunity takes time and therefore you need to start exploring whether you have the right problem in mind or if you have all the necessary data available today. In many cases some areas in the organisation are more prepared than others, but you need to get the required level of digitalisation and gather data as much as possible. Starting today enables you to find out what needs to happen first before it’s too late.Silo.AI’s working model starts with an easy design sprint and PoC which aim to 1) find the best AI opportunity within your optimisation strategies 2) find the best AI model to come up with the desired results. Take a look at our view of an AI solution and its different phases below.

This post was created together with our Head of Operations Henrik Nyman and our Account Executive Juha Tiainen. Reach out to learn more.