Singapore Airlines doubles Kris CSAT with OpenAI

Singapore Airlines has adopted OpenAI models and an application programming interface to improve customer support, travel discovery and employee productivity, according to OpenAI News. The airline also reported that its Kris chatbot doubled customer satisfaction scores, making the case study a notable example of AI moving from experimentation into day-to-day travel operations.

What did Singapore Airlines announce?

According to OpenAI News, Singapore Airlines is using OpenAI models and the OpenAI API across three areas: customer support, travel discovery and employee productivity. The report highlights the airline's Kris chatbot as a visible example, saying that its customer satisfaction score doubled.

The announcement is brief on technical specifics. It does not provide a full architecture diagram, model-by-model breakdown, pricing details or a timeline for every deployment. What it does provide is a clear signal: a major airline is treating generative AI as part of its service and productivity stack, not as a standalone experiment.

For readers who follow enterprise AI, the interesting part is the combination of use cases. Customer support is often the first place companies test language models because the interactions are frequent and measurable. Travel discovery is more complex because it can involve preferences, constraints and changing availability. Employee productivity points to internal tools, where staff may use AI to draft, summarize or retrieve information.

Why does this matter for airlines and travel platforms?

Airlines operate in a high-pressure service environment. Customers ask about bookings, baggage, changes, loyalty benefits, destinations and disruptions, often across multiple channels and time zones. A support assistant that can understand intent and respond quickly can reduce friction, especially when travelers need help outside normal call-center hours.

The travel discovery angle is also significant. Instead of only searching flights or hotels by rigid filters, travelers increasingly expect conversational guidance, such as requests for family-friendly routes, short layovers or quieter cabins on long-haul trips. AI systems can help translate those requests into options, but they must be checked against live inventory, fare rules and airline policy.

Employee productivity matters because airline operations involve many internal workflows. Staff may need to retrieve policy details, summarize passenger issues or prepare responses. If AI tools are deployed carefully, they can shorten routine tasks and let human agents focus on exceptions that require judgment.

What does doubling CSAT mean for the Kris chatbot?

CSAT, or customer satisfaction, is a common support metric. It usually reflects how customers rate an interaction after receiving help. According to OpenAI News, Singapore Airlines said the Kris chatbot doubled its CSAT, which suggests the assistant became more useful or more satisfying to customers in the reported deployment.

The claim is meaningful, but it needs context. A doubled score can look dramatic, yet the baseline matters. If the chatbot started with a very low satisfaction score, doubling it may still leave room for improvement. If the baseline was already high, the result would be even more notable. The source does not provide the underlying numbers, sample size or measurement period, so readers should treat the figure as a reported outcome rather than a fully detailed benchmark.

For support teams, the practical lesson is not simply deploy a chatbot. It is to measure the right things: resolution quality, escalation rate, response time, customer effort, containment and post-interaction satisfaction. A chatbot can improve CSAT by answering faster, but it can also hurt trust if it gives confident but incorrect answers. The Singapore Airlines case is best read as evidence that conversational AI can improve a customer-facing metric when it is integrated into a real service environment.

How could other teams explore a similar approach?

Teams outside aviation can borrow the pattern without copying the exact deployment. The first step is to identify a narrow support flow with clear success criteria, such as answering policy questions, guiding users through a booking change or helping customers find relevant information. The second step is to test the assistant against real examples, including messy language, multilingual requests and edge cases.

A useful exercise is to compare several models on the same support prompts. Different models may vary in tone, accuracy, latency and cost. Grafy Chat can be used for this kind of side-by-side comparison, letting a team put multiple AI models in one workspace and evaluate answers before choosing a direction. For teams that want to prototype an embeddable support experience, Grafy Bots is designed for building embeddable AI support bots. That makes it a practical place to test a help-desk assistant, refine its responses and see how it behaves in a support-style workflow.

The key is governance. Airlines and travel companies handle personal data, payment details, loyalty information and operational constraints. Any AI assistant should be scoped to approved content, include escalation paths to human agents and be monitored for incorrect or unsafe answers. The goal is not to remove humans from support; it is to make the first response faster and more consistent while preserving human judgment for complex cases.

What should developers and researchers watch next?

This story is a reminder that enterprise AI value often comes from integration, not just model capability. The airline did not simply add a chatbot. According to OpenAI News, it connected OpenAI technology to customer support, travel discovery and employee productivity, which implies a broader operational rollout.

Researchers and developers should watch for more detail on evaluation. Useful questions include: How was CSAT measured? Were answers grounded in approved airline content? How did the system handle multilingual travelers? What percentage of conversations required human escalation? How were privacy and security controls applied? These details matter because they separate a polished announcement from a reproducible engineering practice.

For the wider market, the case suggests that travel companies may increasingly compete on service intelligence: how quickly they can answer, how well they understand context and how safely they can act. As AI agents become more capable, the next step may be systems that can not only answer questions but also complete tasks, such as checking options, preparing summaries or routing requests to the right team. The Singapore Airlines report is an early signal of that direction.

Frequently asked questions

Did Singapore Airlines say the Kris chatbot doubled customer satisfaction?

Yes. According to OpenAI News, Singapore Airlines reported that its Kris chatbot doubled CSAT in connection with its use of OpenAI models and API. The source does not provide baseline scores, measurement period or sample size, so the figure should be read as a reported outcome rather than a fully detailed benchmark.

What did Singapore Airlines use from OpenAI?

According to OpenAI News, Singapore Airlines used OpenAI models and the OpenAI API to enhance customer support, travel discovery and employee productivity. The announcement focuses on those three areas and the Kris chatbot result, rather than naming every model, integration detail or internal workflow.

Can Grafy be used to build a similar airline support bot?

Grafy includes Bots, an app for building embeddable AI support bots. A team could use it to prototype an airline-style assistant, test responses and refine a support workflow before wider deployment. Grafy Chat can also help compare multiple AI models side by side during evaluation.

Source: OpenAI News

---

Source: OpenAI News — Singapore Airlines elevates travel & work with OpenAI technology