KWS wants to arouse enthusiasm for artificial intelligence, such as here at the Digital Transformation Day in Berlin.
Technology
AI for KWS
There’s no getting around artificial intelligence (AI) these days. Some examples from the world of KWS illustrate where we’re already using it and what improvements AI delivers.
Hypatos
Transferring invoices made easy
First things first: Reading and entering data from invoices is nothing new – but the new Hypatos system integrated in SAP goes one step further. The software from Germany uses a mix of machine learning and AI to reduce the workload on our colleagues in Berlin and speed up invoicing processes. “This is a lighthouse project for us,” says Kariem Almazahy, Automation Lead Finance. Hypatos has been in use at KWS since November 2024 and can make data-driven decisions on its own. Previously, hundreds of e-mails a day had to be processed manually by Accounts Payable or Accounts Receivable and the data had to be entered. “The challenge was that we needed to be fast, of course, so that invoices were paid on time, but we also had to be absolutely accurate in what we did,” states Kariem. “We had to make sure that no part of an IBAN was incorrect, for example.” Hypatos now performs most of this work. The software was trained with thousands of documents and invoices and then adapted to the special features and structures at KWS and honed further with special prompts (input commands). While OCR (optical image recognition) – the recognition and reading of data – had already been used for a long time, Hypatos can also understand the context. Hypatos makes a decision based on that. If important information is missing, for example, Hypatos notifies our colleagues. “So far, we’ve rolled out the project in Germany and Croatia,” adds Kariem. “We’re very proud that it’s working so well – and other countries and projects are already being planned.” |
Salesforce Einstein
Clever: New ideas for e-mail marketing
We’ve all experienced it: Lots of advertising sent to our private mail addresses throughout the day, often at the same time and one after the other. The consequence: Information is quickly deleted and mails go unread. Yet such advertising is one of the most effective marketing channels, with several billion users – and a direct way to provide information about our products, make KWS even better known and get in touch with customers. That’s where Einstein comes in. It is embedded in Salesforce and is the first AI platform for customer relationship management (CRM). “Einstein is a great solution for us as a company and also offers users a better experience that we can tailor directly to them,” explains Bettina Sannemann, Lead CRM. How is that done? Einstein uses machine learning to determine the ideal time to send an e-mail, for example – individually for each user. The software also creates predictions about interaction behavior: Does someone always look at all e-mails from KWS, or only certain ones, and if so, at what times? Based on that, Einstein can then decide whether someone is receiving too few or too many e-mails so as to generate as many interactions with our farmers as possible. |
KWS MAIA
KWS knowledge in your back pocket
MAIA informs farmers about our crop varieties and digital services within seconds.
Do you need to have the right knowledge about our crops or digital services on hand quickly when in the field? That’s no problem for KWS MAIA, the digital assistant. First launched in Bulgaria and Croatia in May 2024 and now also used in Serbia, Romania and Hungary, the AI-assisted expert helps farmers find out more about our products and services – in just seconds. MAIA can currently be used with the messenger services Viber and WhatsApp. “In Bulgaria, as well as in the whole Southeastern Europe, we have many large farms we offer consulting to directly through our sales and product management teams,” says Petko Vasilev, Head of Bulgaria. “But how do we reach the thousands of smaller ones? They can communicate at all times thanks to MAIA.” The service uses an extensive general agricultural database plus the information fed in by KWS. “Our goal with MAIA is to get even closer to our customers,” says Petko. “We also use the service to show how innovative we are at KWS – and by analyzing the questions and search terms, we can see what farmers are interested in or where we can still work on products in order to get a head start in sales.” MAIA also enables farmers to be actively contacted – in a personalized way. MAIA supplies information about interesting campaigns two to three times a year if the data shows that that might appeal to one of the more than 2,000 registered customers. In this way, marketing campaigns can be specifically tailored or the sales team can be better prepared – with direct information from the market in question. And for those who prefer personal contact, MAIA connects them directly with a local sales employee. “But MAIA already makes the contact very personal,” says Petko. “It speaks the language of each country – you hardly notice you’re talking with a chatbot. KWS MAIA is very attractive for meaningful marketing campaigns, through virtually every possible channel, thanks to the efforts of Ivana Gligorijevic and her team to develop a very good communication package – films, visuals and a number of teaser materials are available, ready for use.” |
Named entity recognition
Tracking down genes with linked information
Artificial intelligence reads scientific texts and establishes connections. Anyone searching for something now gets results quickly.
Finding what is known about the relationships between genes and traits is cumbersome. Most information exists only in scientific publications, and more than a million papers on all topics are published every year. In medicine, structured databases with information from these papers exist, but in agriculture and plant science that is not common. The common way of finding it was to conduct a manual search with keywords across different journals.
“Using AI to discover these connections is not that easy, but it’s also an ideal use case to automate this search,” says Bjoern Oest Hansen, Research Lead Knowledge Discovery. “For example, I can use machine learning to read a text and find traits, but the system also has to recognize the context behind it – when I talk about sting, do I mean the gene sting, a bee sting or the singer?”
To optimize this, the Knowledge Discovery team has trained an AI that uses a method called named entity recognition (NER) – the system searches through texts and sorts proper names or terms into predefined categories (for example, abscisic acid under chemical products or drought under trait). The system needs to be well trained for that to work and a lot of effort was put into getting it right together with colleagues from different departments. This is no easy task, as the same term is often spelled differently or abbreviated – and the AI has to recognize that correctly. Afterwards it has to be processed so if the same gene name appears in multiple organisms, it is matched to the right organism as well. The model is fed up to one million abstracts a year and the output is now stored in a database together with other internal and external information, such as orthology between genes in different organisms. This allows the transfer of knowledge across organisms and connections to internal data, so this can be mixed and matched. A web interface to access this data has been created and if you now search for “damage by insects in corn” there, you obtain all the information available on the topic and can thus continue looking into important candidate genes. One practical aspect is that the AI also provides a summary between specific genes, what we know about them and the traits, and automatically derives a hypothesis. That’s a real benefit for our research, because only with the latest knowledge can we make data-driven decisions for future projects. Discover it now on discovery.kws.com. |
Agentforce
Test phase: Specialist support around the clock
In the future, Agentforce could offer specialist support on our website – around the clock.
Agentforce is an autonomous AI application that could offer employees and customers specialized support around the clock. We are currently testing the AI assistant for our website. In an initial test run, we analyzed whether it can be integrated into our system landscape. If that continues to go well, the service-oriented agent will offer customers several advantages: It can take orders, change delivery addresses and open requests or cases quickly and easily – a real benefit for our colleagues in the back office, who can thus be relieved of routine tasks. The next potential steps are to add further test scenarios to the agent and then gradually integrate it fully into our system landscape. |
INFO
Community for those interested in AI
In 2024, we founded a group of colleagues who are enthusiastic about AI so as to enable them to share ideas and report on the current status of their projects. Now we are opening this group as a community on the intranet. Have we whet your appetite? |
© KWS SAAT SE & Co. KGaA 2025
