Smart Assistants: Why Many in-house AI Chatbots have Problems

There are plenty of test projects with ChatGPT, Gemini and co. However, many companies are finding that AI projects are not as easy to implement as they had hoped.

Individualised AI chatbots or smart assistants offer great potential. They enable companies to communicate quickly and comprehensibly with employees and customers automatically. All available data can be used for this, including internal information that ChatGPT, Gemini and co. have not been trained with. However, according to a Bitkom survey, only three percent of German companies are working with generative AI in their company, with a further six percent planning to use it in the current year.

What are the stumbling blocks in the introduction of smart assistants?

The experience of Lufthansa Industry Solutions (LHIND) from projects with AI chatbots shows this: Companies face the same challenges with smart assistants as with other AI projects.

Data must be available and of high quality

This is often less of a problem when an AI chatbot is fed with information from the company’s internal wiki or website. However, when it comes to more sophisticated and number-based data from the data warehouse or SAP system, many companies still need to take action.

Functioning data governance

Then there are also no problems with document rights management. This is because the company’s own sharepoint, which is often the basis for the AI chatbot, contains a wide variety of data that not everyone is allowed to access, such as sensitive or personal information. Access to such data must also remain prohibited via the AI assistant.

Employees must also be trained

“Many projects show that people are not trained to communicate with a chatbot,” says LHIND Manager Julian Staub. “They are expected to work in a completely new way and this has to be learned.” For example: How do I talk to an AI? How should I ask questions? How should I structure the sentence? How much information do I need to provide? Employees therefore need to be picked up and trained, otherwise there will be a lack of acceptance for the smart assistant.

It is important to involve the specialist departments in the development of the AI chatbot right from the start. The circle should not be limited to IT experts. Companies should also start small. “Just uploading ten terabytes and then simply testing away doesn’t work,” says AI expert Staub. “Instead, a department should start with a few selected employees. The test phase starts with them and their feedback then forms the basis for adjustments to the smart assistant. A project like this can be up and running within two weeks.”

The future of smart assistants

But then there are few limits to smart assistants. “Small individual projects could soon become multi-agent tools. This involves several agents, i.e. AI systems with a fixed task, working together,” adds Max Pillong, Director AI & Data Analytics at LHIND. One assistant finds information on the internet, a second uses the information from the company’s own CRM system, a third has an overview of the key figures from the data warehouse and a fourth solves mathematical tasks. Together, they will be able to solve even more complex challenges in the future.