AustraliaTravel and aviation2024 – ongoing

Flight Centre Travel Group

In-house natural language model for voice-of-customer analysis

Flight Centre's own language model found a process defect, and complaints about it fell 60% month on month

WorkingCompany's own claim

Problem

Flight Centre collected feedback across surveys, social media, review sites and call data. It struggled to see why customers were unhappy.

Approach

The company combined feedback from its channels into one layer and built its own natural language processing model to categorise comments. The model took about 150 hours to develop and works across millions of sentences. It surfaced a process defect that staff had not linked to a pattern of complaints.

Results

60 percent reduction month on monthfall in people discussing the defect after the fix, as described by Flight Centre's executiveCompany's own claim[1]

about 150 hourstime to develop the company's own model, as described by Flight Centre's executiveCompany's own claim[1]

Takeaway

Reading comments at scale let Flight Centre trace a recurring complaint to a specific process fault. The figures come from an executive's account.

Sources

Flight Centre finds the 'why' with natural language processing capabilities
iTnews1 Aug 2024News article

Source summaries

[1]iTnews1 Aug 2024

iTnews reports how Flight Centre merges customer feedback from many channels and categorises it with its own language model. A Flight Centre executive described a defect found through the data, resolved in May, and a 60 percent monthly fall in mentions. The model took about 150 hours to build.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026