About
A Senior UX Designer with 9+ years of experience designing complex digital products across enterprise, fintech, and data-driven platforms. I specialise in systems thinking, decision-centric UX, and scaling design practices that align user needs with business strategy. My work focuses on simplifying complexity, improving product adoption, and enabling cross-functional collaboration through research-backed design approaches. Alongside product design, I actively explore organisational change, AI-enabled workflows, and the evolving role of design leadership in modern teams.I am passionate about creating thoughtful, human centered experiences and contribute to conversations that push UX beyond interfaces into strategy, culture, and meaningful impact.
Talk details
The AI Was Wrong; Now What? Designing Recovery Into High Stakes AI
About this talk
Three weeks after we launched an AI-powered banking assistant, a support ticket landed in our internal Slack. A user had written: 'This app flagged my rent payment as suspicious. Now I do not trust anything it tells me anymore.' Our accuracy metrics were green. Leadership was happy. And somewhere out there, a real person had quietly stopped trusting us, not because the model broke, but because we had given her nowhere to go when it got something wrong.That ticket broke something open for me. We had spent months designing the AI. We had never once designed the moment after it failed. This talk is the story of rebuilding that product around four recovery moments. What I would do differently: start these conversations in discovery, not after launch. The recovery layer is not a patch. It is a design decision that should shape the whole product. I learned that the hard way and this talk is honest about that.
Key takeaway
- A four part Recovery Layer Framework; Explanation, Correction, Escalation, Confidence with concrete design decisions behind each one, not just labels on a slide.
- A reframe that genuinely changes how you build: users do not need AI that is always right. They need AI that is honest when it is not and gives them a clear path forward.
- A practical way to design AI explanations not tooltips or legal disclaimers, but reasoning that actually reduces user anxiety at the moment they are most frustrated.
- Language for the internal arguments you will have specifically how to make the case for showing uncertainty when engineering wants confidence and compliance wants simplicity.
- Four questions to pressure test any AI feature before it ships: Can users understand why? Can they correct it? Can they reach a human? Can they tell how confident the system actually is?