Trust is the product: two years of shipping AI to people who did not ask for it
Halden's AI assistant drafts replies for support agents at twelve thousand companies. Zara Haddad closes day one with what two years, two model changes and a public apology taught her about trust.
- Wednesday 9 June
- Day 1
- 16:30–17:20
- 50 minutes
- Main Stage
- Ground floor, hall A
- Keynote
- All levels
When Halden launched its AI reply assistant, support agents had not asked for it. Their managers had, and many agents were worried about what it meant for their jobs. In its first month the assistant was used on fewer than one reply in ten. A year later it was used on more than half, and agents were asking for more.
What changed was not the model. It was a series of product decisions about who stays in control: showing agents why the assistant suggested something, making it effortless to ignore, never sending anything a person had not seen, and telling customers plainly when a reply had been drafted with help. Along the way Halden also shipped a change that went badly wrong in public, and Zara talks about that too.
She closes the first day with an honest account of building AI features for people who did not choose them, and of what designing for trust looks like in practice.
What you will take away
- Product decisions that build trust in an AI feature among reluctant users
- How Halden measured trust, not just usage, and what each number told the team
- What to do in the first hour after your AI feature fails in public
