Copyright © Emerce 2023

Large companies can staff their call centers with far fewer people because of a new AI technology from CM.com. The Breda-based tech company deploys LLMs who learn to communicate as if they were employees of a company.

The latest private tests are currently running at six companies, including Pathé CZ, Wijnvoordeel and Yoursurprise, with a generative AI that combines company-specific data with the best-known language models, aka LLMs. Depending on the company and industry, a decision tree determines which AI is deployed for which purpose. Ultimately, it is always about answering consumer questions to a customer service department.

CEO Jeroen van Glabbeek explains: "We are always working with customers to improve their question and answer systems," for example with chatbots on a site or customer contact by Whatsapp or text message. "For large companies that can matter because they have the scale to build customer service. Small companies don't have that. Until now."

Companies can train CM' s new Generative AI Engine with their own data. "We learn a company's language in a few minutes by reading the site, old customer conversations and chatbot interactions and numerous files - think JSON and PDF." "Then you can query your own company information. ChatGPT can't do that, because it's too generic. It doesn't learn from internal company data."

The learning process is divided into two steps. A company feeds its data to the CM cloud. This then strips out any potentially personally identifiable information, because of privacy rules, and only then puts it through to Google PaLM, Meta LLaMA, OpenAI GPT and Anthropic, for example. The heavy learning and computation is done there, and from there CM takes control on its own cloud. That's where the question and answer practices take place. That way the Brabanders don't have to keep sending overpriced requests to an American cloud.

The cleverness of this system's design, Van Glabbeek says, allows it to work efficiently. "We spent a lot of time on those routings. That leads to seventy to eighty percent lower costs" than if everything were processed at an exotic LLM.

"Customers have trouble finding budgets and staff for customer service. We say to them, just give us some of those HR budgets, and we'll provide the people work." In practical terms, that means fewer people can do more work. They become more productive because the AI machine listens in on customer conversations and suggests answers. "And so in multiple languages. So a Dutch speaker can speak to French customers of a French service provider or store ín Dutch." The AI does the translations.

CM.com is thinking of billing the AIs as if they were human employees, according to an hourly rate, in other words. The details of this have yet to be worked out.

In a few weeks, a large company will start using the new system. That will result in substantial savings in operating costs. Van Glabbeek outlines that during periods of action, for example, these types of organizations will not have to hire additional staff temporarily to handle a peak in demand. The AI takes on that task.

"You can also start using AI tools for sales purposes. Think of a scenario where someone visits a site and contacts the chatbot with a question. That could actually lead to a transaction, a purchase. Measurements show that chat has a conversion-enhancing effect. It looks like, that's about five percent with a human agent and up to 15 percent with an AI." Director: "We suspect AI bots stay friendly longer and give consistently better answers in terms of content. They are equally good at their work at the beginning and at the end of a workday. Humans, on the other hand, may be tired and therefore a bit short-tempered, for example."

© The Content Exchange, source News