Home Projects Blog About Schedule a call DE

Localization of an Airport AI-Assistant

Localizing an intelligent airport assistant for deployment in Sweden and Japan — ensuring the AI's tone, initiative, and error recovery behaviour aligned with local cultural norms.

Role
UX Researcher / AI Design Lead
Duration
5 months
Stack
LLM · Prompt Engineering · Wizard-of-Oz Testing · UX Research
Year
2024
Localization of an Airport AI-Assistant

The Challenge

An airport technology company had built an AI wayfinding and information assistant. European pilots showed strong usability scores, but early feedback from Japanese airport partners flagged discomfort with the assistant’s direct communication style, proactive suggestions, and recovery phrases after misunderstanding.

Approach and Execution

I ran a comparative UX study across Swedish and Japanese airport staff and frequent flyers using Wizard-of-Oz prototypes. The study mapped cultural dimensions (Hofstede’s uncertainty avoidance, power distance, individualism) to specific conversational patterns — hedging language, initiative level, formality register, and error recovery scripts.

Results

91%
User comfort rating in JP market post-redesign
4 dimensions
Cultural variables mapped to AI behaviour
2 markets
Localized and deployed

Key Insight

The most impactful single change was re-framing the assistant’s error recovery. In the Swedish version it was direct: ‘I didn’t understand that — could you rephrase?’ In the Japanese version it became: ‘I want to make sure I help you correctly — could we try a different way?’ Same function, very different reception.

Outcome

The redesigned assistant was deployed at two Japanese airports and received 91% positive comfort ratings in post-deployment surveys. The cultural localization framework was documented as an internal design standard for future markets.

Tech Stack

LLM · Prompt Engineering · Wizard-of-Oz Testing · UX Research

LLMPrompt EngineeringUX ResearchLocalizationAI DesignAviation

Got a similar challenge?

30-minute call · No prep needed · Clear gaps identified

Start the conversation →