KRO-NCRV
Led the user research that built the Netherlands' first voice-controlled cooking assistant on Google Assistant — bringing BinnensteBuiten's chefs into the kitchen, hands-free, and earning two golden awards.
Voice UX in 2017 had no playbook. Voice UX in the kitchen had even less.
KRO-NCRV wanted to bring the chefs of BinnensteBuiten — a beloved Dutch public-broadcaster show about sustainable cooking — into the kitchen with viewers. Not on a screen. In their own voice. Hands-free, on Google Assistant, while users actually cooked the recipes.
The brief was bold: someone with flour on their hands, a pan on the stove, and a glass of wine within reach should be able to follow a chef-led recipe from start to finish without ever touching a device. The problem: voice UX was a frontier discipline at the time. Almost nobody had shipped a non-trivial conversational interface in Dutch. The kitchen made it harder — recipes are non-linear, full of "wait, what was step three again?", and users multitask in ways no UX lab can simulate. There was no playbook to copy. The research had to invent it.
"If we get the words wrong, the user gives up on dinner. There's no second screen to fall back on."
Weekly sprint research. Designed for a discipline nobody had standardised yet.
No legacy methodology applied cleanly to voice. So I built a sprint-based research practice from scratch — fast enough to keep up with weekly design iterations, rigorous enough to catch the things that break a conversation.
A custom weekly research rhythm matched to the design team's sprint cadence — recruit, run, synthesise, deliver in five days. Voice design moves fast, and research had to keep up without compromising rigor.
Tested in real kitchens — sticky fingers, sizzling pans, kids in the background. Lab tests would never surface the moments where the user's hands were full and the assistant said the wrong thing.
Tested multiple variations of every key moment — confirmations, error recovery, recipe transitions. Worked closely with the BinnensteBuiten chefs so the assistant kept their voice and personality, not generic TTS politeness.
Closed the gap between what users naturally said and what the system could understand. Mapped the actual phrases users tried — "next bit", "what was that again", "skip ahead" — and fed them back into the model.
Ten months. From frontier to first-place.
Defining what "good" sounds like
Started by mapping the cooking scenarios voice had to handle — and where the failure modes lived. Without prior playbooks, every framing decision had to be made from first principles.
Weekly sprints across recipe scenarios
Ran weekly sprint research alongside the design team — testing scripts, tone, recipe variations, and voice commands. Every sprint surfaced new failure modes the team would never have caught at the desk.
Tone, error handling, edge cases
Tightened the chefs' voice persona, hardened error recovery, and stress-tested the edge cases — interrupted users, unclear pronunciations, broken kitchen audio. The unsexy work that decided whether the product shipped or shipped well.
Chef-kok BinnensteBuiten ships
Launched as the Netherlands' first voice-controlled cooking assistant on Google Assistant — bringing the show's chefs into real kitchens, recipe by recipe, hands-free.
Two golden awards
Gouden Spinaward 2019 in Artificial Intelligence & Voice. Gouden Dutch Interactive Award 2019 in New Interfaces. Recognition as a category-defining piece of Dutch voice UX work.
Voice UX isn't designed on a screen. It's designed in words, pauses, and tone — and getting them wrong means the user gives up on dinner.
A research practice for a discipline that didn't exist yet.
Every artefact had to be invented from the ground up — there were no existing voice-UX research playbooks to copy from in 2017.