The Belarusian IT industry is famous for the quality of its products in the country and abroad. One such example was the INFINIUM platform, which changed logistics in Belarus, and in terms of cost-reduction efficiency it surpasses the products of foreign IT giants.
The editors of the largest information and financial Internet resource in Belarus, Myfin.by, conducted a detailed interview about the platform for automating transport and logistics processes and cost optimization based on the Data-Driven approach, artificial intelligence and Big Data from the founder of INFINIUM, Mike Dedunovich.
Artificial intelligence has increased the profitability of businesses. And became available
Could you explain in simple terms how AI became capable of bringing money to users?
Thousands of companies have been using our cloud systems for 20 years. This allowed algorithms to compare the behavior of impersonal users with each other, catch successful practices, train the system on them and create “industry intelligence.” Now the system can find the most profitable solutions. Almost instantly.
Has AI learned to provide better solutions than humans and old technologies?
Not only. If you have reliable information, you make the right decisions. When you have 20 years of reliable information, algorithms picking up trends and millions of factors, you make the right decisions for the coming weeks.
Has forecasting become possible?
Yes. But it is also important that the system tells people which action is not optimal and can train them. In addition, she sees really effective employees, which allows the manager to reward them on time so as not to lose them.
What savings figures are we talking about in practice?
If we are talking about road freight transportation, then about 15% of costs are unjustifiably abraded on asphalt. The system sees this by comparing 100,000 drivers with each other: many drive at the same pace in completely similar conditions, but at lower costs. This means that the rest unjustifiably overspend the resources of their companies. And the system provides tools to reduce them.
And if we talk about delivery to stores and e-commerce, then historically there has been a figure for lost logistics profitability of up to 30–50%.
Let’s talk about delivery logistics: 30-50% is a lot.How it works? And what do forecasts have to do with it?
We have GPS data on the movement of 50,000 anonymous corporate vehicles over 20 years. Our Data Science engineers have learned to use them to predict traffic on the roads and queues in hypermarkets.
How does this relate to business profitability?
This allows you to give the driver the most stressful, but guaranteed, feasible task for tomorrow. That is, the most profitable for the enterprise. No driver dissatisfaction.
Then we spent 10 years developing our own algorithms to optimize delivery route networks, taking into account street loads and queues for unloading. So that route assignments for the driver are optimal and created in seconds.
And who will take into account the requirements of buyers and the capabilities of drivers?
The robot takes into account parameters from the price of the car and the flexibility of the driver to the desires of customers and the risks of them refusing orders. This allows you to balance at the peak of financial optimality of logistics. And instead of routine, logisticians got the opportunity to manage delivery strategies.
In practice, our robot reduces the average work time of logisticians by 30 times, and delivery costs by 30%.
How can you make sure that all drivers drive exactly as prescribed?
Intelligent planning and GPS control of drivers – in one system. Therefore, a plan/actual comparison became possible. Automation detects deviations from optimal routes, runs and schedules. You don’t need to watch everyone: pay attention only to violations.
What advantages do your users get compared to foreign systems?
Google, TomTom, Yandex, etc. receive data from navigators and do not know what the user is driving. But a city bus driver and a fast courier in a car are different “planets”. Averaging their speed is the “average hospital temperature.” The forecast of delivery speed is also obtained accordingly.
We were able to solve this problem: one of our companies (BelTransSputnik) has been installing more than 50,000 GPS units on cars with its own hands for 20 years. We know what type of vehicle the GPS data is coming from.
Therefore, we make street speed forecasts separately for cars, vans, and trucks. This precision made it possible to get rid of conflicts with drivers and reasonably increased the pace of delivery. The driver sees in the map of our application how to go, when, where to be, how much mileage he will be paid at the end of the day.
Consequences of AI: employee – is more valuable, manager – is more independent
Maybe logisticians, like navigators in aviation, will disappear as a profession?
The navigator did not disappear. Technology has made him a second pilot.
What does a logistician usually do in delivery? He spends the whole day distributing orders to cars for tomorrow. He must take into account the capacity and price of each vehicle, driver availability and visits to suppliers’ warehouses, customer requirements and return of containers, cross-docking and cargo compatibility… By the evening, he will complete one logistics option and send out tasks to drivers for tomorrow.
But you need to somehow make logistics optimal for your company: take into account the risks of standing in queues, rescheduling, traffic on every street for tomorrow. And it would be good to calculate a dozen alternative logistics options in order to choose the optimal one.
A person competing with a computer is like a pedestrian overtaking an airplane. You don’t need to compete with him, but “ride him”.
So it turns out that the work of a logistician is no longer needed?
Vice versa. The system automates the work of a logistician so much that he can close the computer after 15 minutes. But the system can further calculate different strategies: “customer focus – profitability”, “risks – benefits” and find an even more profitable solution.
For example, a logistician will try to allow some clients to arrive 10 minutes earlier. The system will recalculate and reduce, say, 3-4 cars in the daily delivery. This is a savings of $10-15 thousand per month. If there were no complaints from clients in a month, it means the logistician did everything right. And worthy of a bonus from the money saved.
What is the benefit of a manager, besides saving?
Digitization of processes. The employee wants to get paid more. To do this, he sometimes tries to make the company dependent on himself. For example, a logistician keeps in his head: when do goods need to be delivered to stores, what is the approximate unloading time? If such a logistician leaves, everything will collapse.
During implementation, we digitize the employee’s knowledge. We invest them in the logistics optimization system and teach the logistician to create more profitable solutions and increase the company’s profitability. Yes, nothing will collapse with his departure. But it is not profitable to let such an employee go: it’s about the “chicken” and the “golden eggs”. Therefore it turns out Win-Win.
Strength – is in ease of use. And the risks are easily eliminated
How difficult is it to move from traditional processes to using AI? What percentage of employees succeed in this?
Simplicity is the key requirement. After expert configuration of the system for the company’s processes, logistics routine work is done with one button. If someone says: “You need to change processes to fit our system,” it means that the developers have not made settings for any business processes. Or they want to get money from you for “personal” improvements.
So, the system will “take off” and increase profitability by pressing one button?
If you use a strong system, this will not happen. The more the system can “bend” to your processes, the more settings it has.
You can lease a Boeing. Your friends fly on it and are happy. But go into the cockpit: there are hundreds of knobs and buttons. You can fly in any conditions. But if you don’t hire a pilot instructor for training, you won’t take off.
The same goes for the cost optimization system. In addition to the right to use the “aircraft”, you need to pay for the “instructor pilot”. He will delve into it and configure the “plane” himself to suit your tasks. It will teach you to “fly” and have fun.
By the way, Boeing is not difficult to fly. If everything is set up correctly before takeoff, then flying an airplane is no more difficult than a bicycle. The automation does the rest for you.
How can a company switch to using AI without stopping its business?
If you have employees who want to learn how to manage AI and become more valuable, then there are no risks of transition.
If your employees do not want to master new technologies, then we recommend ordering a logistics audit before implementation. It will show what additional profitability is at stake, how you can deliver using our system with fewer vehicles, miles driven and fewer hours of your employees.
“Give up” the audit findings to your logisticians. Let them try to find errors. And when you see that there is nothing to argue with them, and the savings found are 30%, you will understand that you need to start.
Don’t worry: you won’t have to break anyone. If employees are unwilling to learn new technology, we will create an IT channel for parallel execution of logistics work. When it turns on, you will pass 5% of orders through it, then 20%, 50%, and start working in a new way without any risks. And the best employees themselves will want to move to a new department.
Driver retention through Big Data and Data Mining has become a reality
Let’s return to motor carriers. They are pitted against each other at tenders, where every percentage is important. And you said about 15% excess losses. Where does this figure come from?
We have been collecting data on the work of more than 100,000 anonymous drivers for 20 years. We developed algorithms that compared drivers with each other, and saw that many drivers perform the job with the same productivity, but at lower costs.
One person drove the section and spent 26 liters of fuel, and the second spent 30 liters. If the conditions were the same, then the second driver burned 4 liters in vain. Because the first one proved with his passage that you can spend 26 liters.
Where did the excessively burned 4 liters go? They additionally wore out tires and wheels, increased the accident rate, etc. This is the law of conservation of energy. Therefore, multiply the cost of wasted fuel by 2–3.
How do you identify drivers who were actually in comparable conditions?
We compare drivers with each other on the same type of cars, roads, with the same load, weather, traffic density, route travel time, etc. But even if we group all drivers on each section of the road under completely comparable conditions, we can see a large spread in the costs they create .
It is like an “X-ray” that determines the hidden qualities of a driver.
But today there is a shortage of drivers in the market. Sometimes there is no one to choose from.
This system is not against the driver, but to help him and the HR service. It shows the driver his suboptimal actions on the road, teaches him to save more fuel, and keep more of it for himself as a bonus. Offer the driver not a “stick”, but a “carrot”. Help him earn more without relocating his family to Poland. And then the issue of driver retention will not be so acute for you.
And the company receives no less: by extending the life of the vehicle and its consumables, and reducing accident rates. Add marketing advantages to this: this is how clearly we monitor the careful transportation of your goods, how we reduce the risks of damage and CO emissions.
Why did you mention the HR service?
Because it is an input filter. Already after the first 50–100 km, the assessment system will show who is in front of you. And not unfounded: it will take into account the complexity of the route, provide a set of evidence on the map with suboptimal actions of the driver in these conditions. You will not allow people into your team who will create unnecessary costs.
In addition, for each of your drivers, the system shows how they rank in your industry. What is the probability that the best candidate is behind the gate?
In advertising there are driver rating systems. Why don’t they work?
They are based on formal approaches: the percentage of driving time on cruise control, increased speed, pressing the brake, compliance with speed signs, etc. But maybe the driver braked the engine correctly – that’s why the increased speed. Maybe he was driving through the hills and correctly turned off the cruise control, which interferes with fuel economy? Maybe he entered the city at 69 km/h and drove through it without braking at that speed at night? That is, he is a professional. You will punish him in vain, he will slam the door, and you will lose an excellent employee. Try to evaluate any driver with such a system, and he will be indignant: “I had a difficult route, you can’t get there better.” And you will have to adjust the rating system so as not to offend the person. What kind of ruler is this that needs to be bent to suit the person being measured?
How can you take into account the complexity of the route and how is the objectivity of the assessment achieved?
We have spent 20 years painstakingly collecting individual Big Data from the work of drivers in various industries: from e-commerce to international transportation. Under each section of the road, a base of up to millions of passages along it was accumulated. And now we compare on a section of each complexity: how did thousands of different drivers drive through it in completely similar conditions? We do not know of any system that could solve this problem as successfully. And we are glad that we managed to do this in Belarus and help our companies.
I would like to understand: why didn’t you take on solving all these problems before?
There is a law about the transition from quantity to quality. While our Big Data “matured”, until it became so large and versatile that it became possible to apply Data Science tools to it, we could not solve these problems. And mathematical methods for searching for patterns in Data Mining have appeared quite recently.
But now we can handle more and more problems that previously had no solution.