Hybrid AI Technologies
The unique property of INFORM's 'Hybrid AI' technology is that it uses both, data-driven algorithms based on advanced analytics, machine learning, deep learning, etc. as well as knowledge-driven algorithms based on mathematical optimization, Operations Research, and human know-how. Data-driven AI techniques can harvest large amounts of data to detect hidden patterns leading to new insights. They also have the ability to learn, i.e. improve the algorithmic decision making over time. On the other hand, mathematical optimization is typically better capable to solve complex planning puzzles. It also provides much faster response to situational changes and process disruptions, elevating the agility of organizations. Hybrid AI helps customers get the best out of both worlds, as leveraging computer algorithms with human expertise yields results significantly superior to both.
Real-world HybridAI Applications
Managerial decision making requires pinpoint accuracy in capturing the right input factors and decision logic for each individual type of planning puzzle. Therefore, INFORM offers dedicated algorithms for different types of management challenges, such as forecasting and demand planning, inventory allocation, supply chain optimization, production scheduling, shift planning, workforce management, transport optimization, spare parts distribution, etc. Today, these AI algorithms drive daily decision making at more than 1,000 companies (often multiple operations) in different industries. This huge work experience is constantly distilled into even more tailor-made decision logic with respect to specific company types, such as airlines, airports, aerospace manufacturers, automotive OEM and suppliers, building materials logistics, chemical plants, container terminals, financial institutions, food manufacturers, insurance companies, hospitals, machine tool makers, pharmaceutical, retailers, seaports, steel mills, transport fleets, warehousing and logistics, etc. to name just a few.
INFORM integrates the dedicated AI software with any existing corporate IT systems such as ERP, warehouse, supply chain, order management, and so on. Based on existing data and user guidance (e.g. order priorities), AI algorithms do the actual planning. Results are presented to users for either an optimization re-launch with modified inputs, or approval with subsequent data integration into the transactional IT landscape. “We are an add-on rather a replacement for anything which is currently in existence,” says Adrian Weiler, CEO of INFORM. As turn-key solution provider, INFORM offers not only software but also consultants to hands-on facilitate the AI collaboration in the actual client organization.
Agile Optimization for Improved Decision Making
“Purely data-driven systems are not agile enough for many applications,” says Weiler. In online banking, for instance, fraudsters are always on the prowl for new fraud schemes. Neural networks identify known patterns even if they are very subtle. But when fraudsters come up with a novel idea, neural networks need weeks to identify the fraud, separating signal from noise in the data. Humans are much faster.
INFORM’s hybrid AI helps customers get the best out of both worlds, as leveraging computer algorithms with human expertise yields results significantly superior to both
“We do not believe in AI running wild,” remarks Weiler. He believes that human control on a meta-basis is essential for AI to function properly. Being a private pilot, Weiler makes use of the autopilot "basically for 90 percent of the time". But depending on the situation, he firmly takes control. He believes that this should also be the case in management decision making. “We don't believe in the total autonomy of the AI in management decision making,” adds Weiler. This is is why AI software needs an interactive user interface.
“What we call 'Agile Optimization' can be explained in three words—smart, rapid, and interactive,” says Weiler. The platform is smart because it addresses large complex decision making problems, often involving scheduling, sequencing, and so on, while producing results very rapidly. But Weiler asserts, “Interactive is an indispensable part because we try to integrate the human decision maker as much as possible.”
In November 2012, the New York Times did a study on timely flight departures in which Delta Air Lines emerged as one of the airlines with the best performance. "We played a significant part in that improvement," ascertains Weiler. Likewise, optimized spare parts distribution makes for happy customers. In materials handling and logistics, costs are typically cut by 12-18 percent. And manufacturing companies enjoy up to 20 percent more orders completed on-time while saving millions in reduced inventories.
The Road Ahead
“Currently, we are aggressively targeting the U.S. market,” says Weiler. Although they have been growing steadily over the past 20 years, especially in Europe, Weiler says they intend to increase their focus on the North American market owing to its size. On the technological forefront, they envision amalgamating better machine learning technologies into traditional mathematical optimization on which a couple of research projects are underway. INFORM usually earmarks three-quarters of corporate profits into research for this purpose. "There are a couple of exciting developments right now in this particular field of integrating machine learning into conventional mathematical optimization, a convergence of sorts," concludes Weiler.