Heathrow cuts phone inquiries from 70% to 10% after launching WhatsApp AI agent built on Salesforce

Heathrow's WhatsApp AI agent cut phone inquiries from 70% to 10% of all customer contacts in its first year. The agent, Hallie, runs on 16 years of Salesforce infrastructure and a database of 500 internal articles.

Categorized in: AI News Customer Support
Published on: May 18, 2026
Heathrow cuts phone inquiries from 70% to 10% after launching WhatsApp AI agent built on Salesforce

Heathrow's AI Agent Cuts Phone Calls by 86% in First Year

Heathrow Airport deployed an AI customer service agent on WhatsApp in March 2025 that reduced phone inquiries from 70% of all customer contacts to just 10% within a year. The agent, called Hallie, answers traveler questions about flight status, security wait times, and terminal locations without human intervention.

The airport handled nearly 85 million passengers in 2025. Peter Burns, Heathrow's director of digital, e-commerce, and marketing, said the airport needed a technology solution to handle growing demand. Heathrow spent three years deepening its partnership with Salesforce - which began in 2009 - to build Hallie.

Building the Foundation

Heathrow's success with Hallie rested on years of unglamorous groundwork. The airport deployed Salesforce Service Cloud in 2009 to centralize customer service inquiries across multiple channels. By 2021, Heathrow had uploaded all customer and marketing data to Salesforce's real-time data platform, giving staff live views of passenger information.

This unified database became essential. Hallie pulls from a database of 500 internal articles outlining airport processes. Burns said some policies took several months to gather, update, and implement before the agent could use them reliably.

In July 2024, Heathrow added real-time flight-tracking data to its Salesforce platform. That addition - combined with airport maps, customer data, and Salesforce's generative AI capabilities - enabled Hallie's launch eight months later.

How the AI Works

Heathrow first tested Salesforce's generative AI tools in January 2023 on two internal tasks: drafting responses to customer inquiries and writing case summaries. These tests fed more information back into the database, helping refine response quality over time.

By late 2023, Salesforce's Agentforce - an autonomous AI platform - was deployed in Heathrow's contact center to assist human agents with those same tasks. Previously, agents typed every response and summary manually.

Hallie operates with strict guardrails. It scrapes data only from Heathrow's website and internal database, not the open web. Burns said external sources could contain inaccurate or outdated information. This limitation means Hallie cannot yet handle personalized questions like baggage pickup locations.

Why Customer Service Works as an AI Test Case

Customer service has become the most common entry point for generative AI across industries since ChatGPT's 2022 launch. Bern Elliot, a research vice president at Gartner, said contact centers work well as AI testing grounds because their success metrics are clear: time saved, call volume, response times, and response quality.

"You don't have to connect very many dots to get from what you deployed to what is the measurable improvement," Elliot said.

However, Elliot warned that deploying AI to customer service systems isn't straightforward. Large language models trained on outdated data produce incorrect responses. "Having AI-ready data is a huge issue," he said.

Next Steps

Heathrow plans to expand Hallie beyond WhatsApp this year, adding the agent to its website and app. The airport is also considering kiosk deployments throughout its terminals.

Burns said Heathrow plans to add more functionality to Hallie so passengers can handle additional tasks. The airport is working with Salesforce's team of UK-based engineers to identify new pain points where AI could assist travelers.

For professionals managing customer support operations, Heathrow's experience offers a practical lesson: AI for Customer Support requires months of data preparation before deployment. The generative AI and LLM capabilities matter less than having clean, current, unified data to feed them.


Get Daily AI News

Your membership also unlocks:

700+ AI Courses
700+ Certifications
Personalized AI Learning Plan
6500+ AI Tools (no Ads)
Daily AI News by job industry (no Ads)