Why mathematical optimization will help your organization
Machine Learning may not be enough to get the best out of Data Science
I will tell you the story of Adam*. Adam is a truck dispatcher, working in a distribution warehouse. His daily job is to assign a few hundred daily orders to trucks, so that they can be delivered to their customers on time. He has been working on this for 10 years and it is very hard to replace him (even when he is sick) as he knows the customers, orders and trucking companies quite well.
Adam needs a secondary monitor, as he needs to continuously work with order data and truck data simultaneously while checking distances and driving durations on the map. It costs the company €200, but helps Adam work much more efficiently, reducing the time he needs to switch between windows on his computer.
To get a secondary monitor, in many organizations, you would need to fill out a form, explaining why you need this purchase. Then it has to be approved by your line manager, his/her manager and people from procurement/the IT team or both. Quite a bit of paperwork, right?
Back to Adam’s job… He dispatches 20 trucks per day, and depending on the load type, number of stops and total kilometers it might cost around €300 to €500 for domestic deliveries in European countries. So he spends approximately €8.000 per day without the need of asking anyone. It makes €2 Million per year!
In many organizations, a secondary monitor requires only a few decisions: yes or no and maybe a selection between different sizes or models.
In his job, on the other hand, he has trillions of different options he can choose from to assign the orders to the trucks. And he usually has to do it within only a few hours! Even a 5% cost difference between his decision and an optimal assignment would costs the company €100K/year, much more than Adam’s annual salary!
Adam’s decisions among the trillions of options can lead to an increased cost of €100K/year without anyone noticing, while the decision of a €200 monitor need many pairs of eyes.
Here is where Mathematical Optimization / Operations Research (OR) comes into the play. We can empower Adam with a tool that builds efficient tours Mathematical Optimization. Such a tool can easily help him build more cost efficient tours, in much shorter times so that he can focus on other critical tasks as finding cheaper transportation companies or troubleshooting: which multiplies the overall savings. I will not go into details here, but one can look into on topics like Vehicle Routing Problems to understand how this is possible.
Back to the title… Look at your organization and your biggest spend buckets in your balance sheet. It can be transportation, it can be staff related costs, it can be energy or whatever fits to your business (model).
Can you identify 'Adams' in your organization, who are managing this big bucket of your costs daily and giving very complex decisions by themselves?
Most Data Science work in organizations these days are dominated by Machine Learning type of projects — and for a good reason: There is quite an impact that you can uncover by better understanding today or the future using ML. On the other hand, there are people out there looking at your data or your ML results and taking actions that might cost your organization millions. They might be called plannerd, staff dispatchers, network design specialists, pricing analysts or something else, depending on what your organization does.
Identifying these people and understanding how they give their decisions, can be the first step to unlock savings through Mathematical Optimization / Operations Research.
*The story of Adam – who is a fictional character – is inspired by a story I heard from a very experienced manager I worked with. The crux of the original story told by him was as follows – and it is as impressive as the story of Adam:
a regular white-collar employee needs multiple signatures for a €50 expense, but can call for a meeting of 20 people for 1 hour and spend €1000 without asking anyone.
Author: Baris Cem Sal
Source: Towards Data Science