Tony North has been in this industry since 1994. In his lightning talk at HDI’s Service & Support World, he covered a lot of ground in what he’s seen in his career: real-world disasters, governance frameworks, ethics training and the most underutilized asset in any AI deployment. Spoiler: it’s the person answering your support queue.
“Trust is the foundation of AI,” North says. “If people don’t trust it, they won’t use it.”
That sounds obvious, but most organizations treat “trust” as something to do at the end of the AI rollout. North says it has to be the first design principle, baked into how the system is built, how it’s explained and how it’s governed.
He shared three examples of what happens when you skip the trust step:
- Amazon: In 2018, the company rolled out an AI-powered hiring tool built on a decade of historical data, but the data was overwhelmingly male. The system penalized resumes that included the word “women.” Even though Amazon was actively trying to build diverse teams, the tool worked against them.
- The city of San Francisco: In 2023, the city deployed an AI system to manage employee benefits. It incorrectly denied benefits to 14% of employees. When they dug into why, they found the system had been trained on English-dominated historical data. Anyone who submitted a request in another language got flagged.
- Delta Airlines: In 2024, the CrowdStrike outage hit Delta’s AI-dependent scheduling system. It collapsed because there was no manual backup process. The company lost $5 million in revenue and 7,000 flights were canceled.
The Delta example illustrates North’s main point. While executives were calculating losses, it was the support teams who staffed the airports, rebooked passengers, reviewed logs, removed corrupted files, rebooted systems and managed the chaos.
“Support teams are the first to see and solve AI issues,” North says. “They’re the first ones to talk with customers when AI issues occur. They’re the first ones to detect issues. They are the AI frontline.”
The San Francisco and Amazon examples share a common failure: governance came after the fact, if it came at all. North encouraged attendees to build a classification system before the AI tool or software goes live. This means classifying every type of intake your system will handle: by risk, by automation, by prediction, by decision support, by generative content. Every category gets its own rules.
“You’ll be able to explain everything that comes through your AI system,” North says. “And most important, you'll be able to solve problems much quicker.”
North’s favorite example of AI done well is from the city of Amsterdam. It’s the first government in the world to publicly publish its AI algorithms. It’s a complete catalog of every AI system the city uses, what decisions it influences and how residents can challenge it.
Finally, North encouraged attendees to train every staff member on AI ethics. He wants to see more conversations on fairness, accountability, transparency, privacy and safety.
“We want employees to make the right decision for the right reason, and not be told what to do,” North says. “It’s a little similar when we talk about AI systems. We want a culture of trust.”