Artificial intelligence in practice: Four AI projects with great value in leading businesses

Antire has been working with machine learning and artificial intelligence (AI) since 2013. Over the years, we have completed several hundred projects across various industries and with some of the leading enterprises in the Nordic region. 

What most of the projects have in common is that they involve few people and take a short time. It usually takes 1–2 months from start to implementation and production. The projects typically last no more than six months. The AI solutions solve specific tasks, providing excellent value and paying off quickly. 

Machine learning and AI have followed Moore's Law in terms of development, with capacity doubling approximately every 18 months. In the last couple of years, however, growth has accelerated, with capacity doubling every 3–4 months. This means that the AI ​​solutions we deliver do more and have greater value for customers.

The projects can be made available in several ways, for example, as independent apps, dashboards, and solutions that are integrated into the customers' ERP systems. The value depends on use. It is therefore essential that they have a good interface and are easily accessible to everyone who will use them. If customers wish, they can add their factors afterwards. 

Below, we present four concrete AI projects that we have developed recently.

1. Helping a large power company balance the energy market
2. Halving power consumption in mobile towers with AI
3. Precise and inexpensive blocking of fraud SMS messages from leading mobile operators
4. Finding the right place for hydrogen factories and wind mills within hours instead of weeks

Customer: One of the largest power players in Norway

Challenge: In Norway, electricity is traded in two ways. Financial trading of future contracts and derivatives takes place on Euronext, while Nord Pool has the balance market. The purpose of the balance market is to ensure a balance between power production and electricity consumption.

In the Norwegian power system, which is part of the Nordic and European power market, production and consumption of electricity must be in balance at all times. Electricity cannot be stored on a large scale; therefore, imbalances between production and consumption must be addressed immediately. The power system is continuously monitored to ensure that reserves are available to adjust imbalances in real-time.

This balancing market comprises various types of reserves that are activated over different time scales, which can be activated automatically or manually, depending on the urgency of the imbalance. The players in the market, typically power producers and large consumers that may have flexible consumption, offer capacity in these markets, and are compensated for making this available. For companies involved in power production, trading, or managing industrial loads, the balance market presents both opportunities and risks, where precise forecasts and automated decision-making systems can provide significant benefits.

When imbalances occur in the system, for example, when consumption suddenly exceeds production, traders can offer up-regulation, i.e., increased production or reduced consumption, to contribute to balance. Similarly, if there is too much power in the system, they can offer down-regulation by reducing production or increasing consumption. Correct timing and a precise understanding of the market mechanisms are essential to ensure profitability and stability in the power system.

Solution: We developed an advanced decision support platform for the balance market that combines machine learning with physical flow models – what we call a digital twin of the power system. The model has been developed in close collaboration with the traders and should be a valuable tool for them, not a replacement. By avoiding overtraining and ensuring understandable predictions, we have developed a solution that enables people and machines to work together effectively.

The model is composed of several machine learning models that together simulate the power system, of which the Nordic market is a part. The models are designed to be a valuable partner for traders who closely follow this market.

Result: The platform is under continuous development and provides the traders with an essential tool in their toolbox. The first version of the model is designed to provide traders with useful information and better insight than their competitors. It is in the plans that the platform should be able to make independent trading decisions when there are no traders at work. This not only requires a strong focus on delivering perfect predictions, but also a strong focus on risk management and handling.

By being correctly positioned in this market, the customer, one of the largest power players in Norway, can generate significant revenue, or, just as importantly, avoid incurring substantial losses by not being able to deliver registered power. 

Customer: Large telecom operator

Challenge: Base stations for mobile communications account for a significant portion of electricity consumption in the telecom industry. In Norway, there are around 25,000 such masts, and globally over 13 million. Each station uses between 2,000 and 6,000 kWh of energy per year. More efficient use of the base stations saves electricity, money, and the environment. In total, the telecom industry uses 2–3 percent of global electricity consumption.

Solution: We developed a machine learning-based prototype that analyzes the power consumption of base stations and identifies savings potential. The project is about adapting power consumption to actual mobile use. The solution takes into account factors such as traffic volume, weather, location, consumption pattern, and a number of other factors that influence whether the base station is used as efficiently as possible. For example, base stations at winter sports venues have large variations in mobile traffic throughout the year. By analyzing and predicting needs, energy use can be reduced without compromising coverage or capacity.

Result: The solution has demonstrated that it is possible to almost halve the power consumption in the base stations. This enables the optimization of operations both economically and environmentally, without compromising customer experience or safety.

Would you like to learn more about how you can leverage artificial intelligence and machine learning to drive value in your business?

Customer: Large telecom company

Challenge: Daily, over 45 million text messages (SMSes) are sent in Norway. Many are sent from computers, and large amounts of these are attempts at fraud. Almost everyone has experienced receiving a fraudulent SMS. In a recent survey, over 70% of respondents reported having been defrauded. At the same time, there is zero tolerance for legitimate messages to be stopped. This creates a demanding need for precise filtering to distinguish between fraudulent and legitimate incoming SMS messages, primarily because the selection of fraudulent SMS messages must occur with virtually no delay.

Solution: We developed a machine learning solution that analyzes and filters SMS traffic in real time, based on more than 100 million log lines daily. The model identifies suspicious messages with high precision and continuously adapts to new fraud patterns. At the same time, the model ensures that no legitimate SMS is stopped.

Result: Our solution was implemented and delivered excellent results for the customer. The telecoms company nevertheless decided to replace the solution with commercial off-the-shelf products to increase IT standardization within the company, even though the solution cost more than six times that of Antire's solution.

However, the change yielded significantly worse results, along with the high price, that Antire's AI solution was implemented again. This results in significantly fewer fraudulent SMS messages for the telecom company's customers as well as a lower error rate. The result is far greater customer satisfaction. The system protects mobile users against fraud, without stopping genuine messages.

Customer: One of Norway's largest energy companies

Challenge: Hydrogen production will be a crucial component in the future energy mix because it efficiently stores energy and can be produced with surplus power. Hydrogen production is most efficient when it occurs close to both power sources and consumers. At the same time, there are several key considerations to take into account when setting up a hydrogen factory, including electricity prices, area, regulations, logistics, emission possibilities, and local interests. Finding the best locations requires analyzing vast amounts of data. Typically, this task is performed manually. Those who do the work spend hours scrolling through map services, hundreds of hours reviewing various documentation, and attend numerous meetings each year with both landowners and municipalities. Most of them to no avail. 

Solution: We developed a tool that analyzes over 100 factors to identify optimal locations for hydrogen production, down to areas as small as 10 x 10 meters. Users can weigh factors, enter requirements, and make adjustments in real-time in dialogue with stakeholders. The tool quickly ranks areas according to their suitability, considering both proximity to potential customers and investment factors, as well as finances. 

The tool can also be used in meetings with stakeholders, whether they are municipalities or landowners, to quickly identify areas to focus on. 

Result: The efficiency of the work with location assessment has increased by almost 90%. What used to take days manually clicking in map tools is now done in minutes. Our customer, one of Norway's largest power companies, also has a department that focuses on wind power production. When this department saw the tool the hydrogen gang had obtained, they wanted a similar solution. Thus, we also developed a model that calculates the expected annual production of electricity for wind turbines of different sizes, as well as the expected development costs. Perhaps even more importantly, you can quickly identify areas that are not suitable, and can thus focus all efforts on the best places for wind power production.

Would you like to learn more about how you can leverage artificial intelligence and machine learning to drive value in your business?