At Rise of AI 2018 in Berlin, Dr. Jan Hofmann, Vice President at Deutsche Telekom, describes how the company built internal and external trust while rolling out AI-driven chatbots. He frames the opportunity through scale, roughly 160 million mobile customers and large volumes of repetitive processes and data, set against a mix of enthusiasm and fear inside the organization about jobs being automated away. Rather than a hidden skunkworks project, a scattered grassroots effort, or a CEO-mandated push across every department, Deutsche Telekom concentrated resources on one focus area and built trust through it, a strategy he calls the trust game. He identifies rigorous focus, fast delivery of small working products, and close day-one collaboration between R&D and business units as the three drivers that made this work, alongside publishing public AI guidelines to build trust with customers and society. He then walks through the resulting rule-based and AI customer service bots in Austria and Germany, and describes the company's next steps into AI-driven process automation and predictive analytics.
Key takeaways
- Deutsche Telekom rejected a hidden skunkworks project, a grassroots approach and a CEO-mandated push, choosing to concentrate resources instead.
- The team spent over a year iterating a rule-based bot in front of customers before introducing AI, favoring small wins over moonshots.
- Deutsche Telekom fully switched its Austrian customer service bot from rule-based to AI about four weeks before the talk.
- Close, day-one collaboration between R&D and the business side, starting with an Austrian e-business team, was a key trust driver.
- Deutsche Telekom published public AI guidelines shortly before the talk to build trust with customers and society.
Detailed notes
Why AI matters at Deutsche Telekom's scale
Hofmann sets up the case for AI at Deutsche Telekom with its roughly 160 million mobile phone customers worldwide, which generate large volumes of repetitive processes and data, even though the company does not hold data at the scale of Google or Facebook. He notes a mixture of enthusiasm and fear about AI both across the country and inside the organization, driven partly by concern that jobs are being automated away.
Choosing a strategy: the trust game
Hofmann outlines three ways an organization could approach AI: a hidden 'big surprise' skunkworks project revealed once finished; a grassroots approach that lets AI happen organically with scarce existing resources; or a CEO-mandated push telling every department to use AI, which he credits to companies like Amazon giving every system an API. He argues none suited Deutsche Telekom, and that the company instead concentrated resources in one area and used it to build trust across the organization, which he calls the trust game.
Three drivers of internal trust
The first driver, rigorous focus, meant that after starting more than two and a half years earlier, the team worked almost exclusively on chatbots and resisted pressure to expand into other problem areas colleagues assumed a general 'AI brain' could solve. The second, fast delivery, meant spending over a year building and iterating a rule-based bot in front of real customers before introducing AI, favoring small, fast wins over big moonshots. The third was working as one team with the business side from day one rather than running an R&D project in isolation, starting with an Austrian subsidiary's e-business team that wanted the technology. Hofmann adds that Deutsche Telekom published AI guidelines about a week or two before the talk to build trust with customers and society more broadly.
From rule-based bot to AI bot
The rule-based customer service bot, called Tinker, launched first in Austria and was extended with new media formats and a handover to a live agent when it reached the limits of its knowledge. Rather than waiting for feature parity, the team launched a new AI-based bot early, inviting users of the rule-based bot to try it, and gathered feedback for more than a year before switching completely to the AI bot about four weeks before the talk, first in Austria and similarly for Deutsche Telekom Germany. All of these bots are focused purely on customer service.
Scaling beyond bots
Building on that internal trust, Hofmann says the team is now also building voice bots and agent-facing systems that give service agents faster access to information. Beyond chat, the company is moving into two new fields: automating the manual handoffs between IT systems that go beyond classic rule-based robotic process automation, and 'smart decisioning', using machine learning to analyze data within a system and generate action recommendations, with a long backlog of use cases identified especially in finance.
Quotes
We just do things and don't create moonshots.
Jan Hofmann
We don't do it, we just keep that focus.
Jan Hofmann
Questions answered in this talk
What strategies did Deutsche Telekom consider for rolling out AI across the organization?
According to Jan Hofmann at Rise of AI 2018, the options were a hidden skunkworks project, a grassroots approach, or a CEO-mandated push across every department; Deutsche Telekom instead concentrated resources in one area and used it to build organization-wide trust.
How did Deutsche Telekom build internal trust for its AI chatbot project?
According to Jan Hofmann at Rise of AI 2018, the drivers were rigorous focus on bots only, fast delivery of small working products instead of moonshots, and close day-one collaboration between R&D and the business side.
When did Deutsche Telekom switch its customer service bot from rule-based to AI?
According to Jan Hofmann at Rise of AI 2018, the company launched a new AI bot early without full feature parity, gathered feedback for over a year, and switched completely about four weeks before the talk, first in Austria and then similarly in Germany.