Jet2 Analyzes Customer Sentiment Across Thousands of Daily Service Calls

Over the last ten years, Jet2 has been busy. Rapidly growing its business and footprint across the UK travel and tourism market, it’s now the country's largest tour operator and third-largest airline. Customers appreciate Jet2’s affordable package holidays to over 75 sun, city and ski destinations across Europe. They also appreciate Jet2’s relentless focus on the customer, proven by the tour operator’s top-ten ranking on the UK Customer Satisfaction Index and its consistent awards from Which?, Globe Travel Awards and other industry bodies.

But to stay customer-focused, Jet2 needs to know who its customers are and what they want from their travel experiences and tour providers. When you receive thousands of customer communications a day, that’s no small feat.

Jet2 has been using Snowflake’s AI Data Cloud to scale both its data and its understanding of customer intent. And now, using Cortex AI, Jet2 has been making the most of large language models (LLMs) to analyze customer intent and sentiment across thousands of call transcripts — and make improvements across its services.

Jet2 is on a journey to become increasingly data-driven.  “Putting data in the hands of our analysts is vital,” says Mark Atkinson, Head of Data Science at Jet2. “But that’s been a challenge in the past due to infrastructure limitations, inaccessibility and silos.”

And as Jet2 scaled its business, manual processes and limitations across its legacy data analytics platforms began to get in the way of business efficiency and innovation.

One area where scaling issues were particularly apparent was in the customer contact center. With thousands of calls a day, some of which could stretch to 40 minutes, there was a wealth of data for Jet2’s leadership about customer intent, preferences and pain points. But there was no way to easily summarize their content.

After assessing several other cloud data platform providers, Jet2 chose Snowflake’s AI Data Cloud on Amazon Web Services (AWS) for its scalability, feature set and rigorous cost controls. But it got even more benefits in the form of support for new AI features that could solve its customer intent analysis challenges for good.

The team had explored traditional analytics approaches and high-level machine learning, but wanted a solution they could deploy quickly. To achieve this, they decided to use LLMs to automatically analyze the content of call transcripts. But there was just one problem: Prior to this project, Jet2’s data science team had never used generative AI or LLMs before. 

Thankfully, the team could get up and running quickly with Snowflake’s Cortex AI functionality, using familiar SQL code to deploy a fine-tuned Mistral model that achieves the right balance between cost, complexity and accuracy.

“We’d tried prompts in general-purpose LLMs, but as we got more descriptive with our prompts, we saw the price kept increasing,” explains Niraj Kulkarni, Technical Delivery Manager, Data Science & Machine Learning Engineer at Jet2. To keep costs down, Jet2 uses a mixture of manual classifications and Mistral LLM that’s been fine-tuned around its sentiment analysis use case. Jet2 also uses Cortex to maintain an LLM ops framework that measures the accuracy and classification performance of its model over time — delivering a model that maintains the accuracy of a large LLM but at a much lower cost.

Having found a way forward with Cortex AI, Jet2’s data science team was able to get the use case up and running in just three months. Now, it processes 25 percent of the company's pre-travel services line calls, which still means around 2,000 calls per day. And even processing this small subset of Jet2’s customer service calls is yielding significant benefits.

Having delivered a successful initial LLM use case, Jet2 now has a robust, repeatable process to better understand customer needs and intent — with that kind of transparency into what customers are looking for, Jet2 can give them more of what they truly want. Even across the small proportion of pre-travel service calls it analyzes today, Jet2 can see trends where some customers are calling for basic hotel and tour information which could be provided through a self-serve online portal. 

The implications are significant for Jet2 and its customers alike. With better access to information and a smoother customer experience based on real insights into common issues and queries, Jet2’s customers get a superior experience. And for Jet2, this only adds to the many reasons customers keep coming back to the tour operator and airline. Plus, opening up self-serve options can help reduce the cost to serve and keep phone agents free for more complex queries.

Snowflake empowers Jet2 to better understand its customers’ questions before they embark on their trips — but that’s only the beginning. Next, the team intends to apply the same principle to operations, sales call lines  and beyond. “The general framework helps us better interpret any kind of communications from customers,” explains Atkinson. “So, we’re now looking at how we can use it for other use cases across our messaging channels.”

The plan is to also extend the benefits of Snowflake on AWS more broadly to all teams within Jet2, from data scientists and data engineers to analysts — no matter how data savvy they are. “We want to use Snowflake to enable self-service insights for everyone,” explains David Herd, Head of Data Engineering and Analytics at Jet2. “In this way, it supports our whole data community, and underpins everything we’re doing with data.”