top of page
  • LinkedIn
  • LinkedIn
Arrival experience

AI-powered insight generation: Linking VOC to booking behavior

Hospitality | Guest experience

citizenM had rich guest data—surveys, feedback, and booking behavior—but no structured way to connect guest sentiment to outcomes. This case shows how we applied NLP and clustering to bridge that gap and drive a 10% lift in NPS and conversion.

I led the initiative as Head of UX at citizenM, in collaboration with UXR, internal data teams and an external ML partner. This became one of the company’s first AI-driven insight pilots—and a foundational test case for future scaled use.

CLIENT

citizenM hotels

01. Situation

Feedback was fragmented across tools and teams. Insights were often anecdotal (shared via Slack), and there was no connection between what guests said and how they behaved (e.g., repeat bookings, conversions).

02. Task

Design and lead a cross-functional proof of concept that:

  • Consolidated qualitative feedback with behavioral data

  • Used AI to surface guest themes and sentiment

  • Connected insights to real-world business levers

  • Built stakeholder confidence in AI-driven decision making

03. Action

  • Formed a cross-functional team with UX, Tech, Marketing, and an external ML partner

  • Consolidated 12 months of guest feedback + booking data

  • Applied NLP techniques:

    • Topic modeling for theme identification

    • Sentiment analysis to map tone

    • Clustering to reveal guest behavior patterns

  • Co-developed a feedback taxonomy aligned to journey stages (check-in, sleep, staff, etc.)

  • Audited classification logic to improve accuracy (e.g., resolving misclassified themes)

  • Delivered results through a dashboard and a series of cross-functional workshops

04. Result

  • The analysis uncovered specific correlations between guest sentiment and behavior, leading to clear strategic takeaways. For instance, sleep quality emerged as a top driver of repeat bookings—confirming its importance in brand loyalty. Conversely, lower sentiment around room experience and ambassador friendliness signaled potential deterrents to conversion, guiding targeted service improvements.

  • Surprisingly, room controls, despite internal investment and dedicated staff attention, showed little influence on guest behavior—enabling the team to redirect resources more effectively.

  • This AI-driven approach led to a 10% lift in both NPS and conversion. More importantly, it introduced a new standard for triangulating insight across UX, Data, and Strategy, helping citizenM operationalize AI beyond experimentation.

Illustrating the process

This section outlines the architecture of our insight generation workflow. We applied NLP techniques—topic modeling, sentiment analysis, and clustering—to 12 months of guest feedback, mapped results to a custom journey-based taxonomy, and linked them to behavioral metrics such as repeat bookings and conversion. The resulting pipeline enabled scalable, interpretable insight delivery across UX, data, and strategy functions.

Project kickoff session

Aligning on goals, taxonomy, and collaboration structure

citizenM.png

This early-stage workshop brought together UX, data, and marketing stakeholders to define project goals and co-develop the initial feedback taxonomy. It set the foundation for shared language and cross-functional alignment throughout the AI insight pilot.

VOC to action: Insight pipeline

Mapping unstructured feedback to behavioral and business outcomes

flow diagram

A visual flow diagram showing the pipeline from raw guest feedback to executive insights. Steps include data consolidation, NLP processing (topic modeling, sentiment analysis, clustering), taxonomy alignment to guest journey stages, and behavioral correlation—culminating in an executive dashboard.

Feedback taxonomy

Structuring guest feedback by journey stage

Taxonomy

The guest experience taxonomy was developed in collaboration with internal teams. Aligned to key journey stages—'Moments of truth'—it allowed for consistent classification of qualitative input.

Sentiment & behavior dashboard

Identifying high-impact themes through AI-generated insight

Relation_edited.jpg

This outcome shows how guest sentiment across feedback themes aligned with booking behavior. By surfacing key patterns—such as the strength of sleep as a loyalty driver and the limited impact of room controls—it helped teams prioritize improvements where they mattered most. The visualization translated abstract AI outputs into actionable insights for business and service design.

Key takeaway

This project reinforced the importance of building systems that translate across disciplines. The real work wasn’t just in surfacing insights—it was in making them meaningful to UX teams, marketing, data scientists, and executives alike. That experience continues to shape how I approach AI collaboration today.

Navigator methods & frameworks used

 

  • Feedback taxonomy creation

  • NLP + Behavioral Correlation Model

  • Human-in-the-loop model review

  • AI literacy workshop

I was skeptical at first—we were already looking at guest feedback on a day-to-day basis. But seeing longer-term patterns emerge from the data was a real win. It helped us understand what truly matters to guests.

– PM, citizenM hotels

bottom of page