Bangkok hospital and Australian tiler show how AI returns real revenue

About 95 per cent of enterprise generative AI pilots produce no measurable effect on profit and loss, according to an MIT study published last year. Two Asia-Pacific organisations have broken that pattern, and both did it the same way: they targeted specific bottlenecks, cleaned their data before deploying agents, and judged results in revenue rather than pilot counts.

Bumrungrad International Hospital in Bangkok, one of the largest medical facilities in Southeast Asia, and The Grout Guy, an Australian tiling and re-grouting firm, share little on the surface. One operates under heavy clinical regulation; the other sends tradespeople into suburban bathrooms.

What did the hospital deploy?

Bumrungrad has used AI in radiology for more than four years. Its current push targets hospital operations and administrative paperwork, James McLeary, chief innovation officer and chief information security officer, told Techgoondu at Dreamforce, a Salesforce event, in September.

In consultation rooms, the hospital moved its electronic health records onto an AI-enabled platform called IntelliCare. It listens to doctor-patient conversations and turns them into structured notes. One rule stays firm: no AI output, whether a consultation note or a radiology assessment, enters the record until a physician signs it off. "We don't allow any AI to make the clinical decision," McLeary said.

The hospital also confronted unsanctioned AI use. Frustrated physicians had installed tools such as OpenClaw on their own devices. A ban would only push the practice out of sight. "If you don't allow it, it will happen anyway," McLeary said. Instead, IT built a sandbox with a capped budget where doctors can try frontier models under enterprise controls. When a tool proves useful, IT rebuilds it properly and puts it into production. "We wanted to provide them with a safe environment to apply their creative ideas," McLeary said.

The contact centre shows the financial return. Patient e-mails are summarised, sorted into three tiers and routed immediately. Agents on WhatsApp and Line answer routine price and promotion questions and pass complex cases to staff. Multilingual voice agents are in development. The hospital is also working through its list of dropped calls and turning them into confirmed bookings, capturing demand it previously lost.

How did the tiling firm get its return?

The Grout Guy took a rougher road. The Perth-based business grew from 30 staff in Western Australia to nearly 200 across six states, chief technology officer Anthony Messina said. Its first systems overhaul failed because the outside partner did not understand trade work, which serves mainly customers outside urban areas. Customer journeys built for digital businesses do not fit a job involving a van, a tradesperson and a wet bathroom floor, Messina explained. "Australia is big, so travel time is a factor," he said.

The company restarted with a mobile app built on a single data source before adding AI. Homeowners upload photos of damaged tiling; computer vision (AI-based image analysis) assesses the condition and lets the company prioritise work by importance and revenue potential. Technicians photograph sites before and after work, and the app generates the service report and invoice.

The app carries three named AI agents:

• Walter processes PDF work orders around the clock.
• Penny chases unpaid invoices.
• Mason handles cancellation and rescheduling e-mails arriving after hours.

Work is recorded in the cloud, so the company does not need to grow administrative staff as it expands nationally. Quote conversion rose from 30 per cent on the old software to 50 per cent with the agents, which the company credits with helping drive average revenue growth of 40 per cent a year over five years. Management has spun the technology arm out as Servly, selling the same trade-specific agents to other trade businesses. "IT needs to be seen not as a cost in the business, but as the engine room," Messina said.

What comes next?

Success brings its own problem: tracking agents as they spread. Gavin Barfield, field CTO at Salesforce Asean, expects agent numbers in a typical enterprise to "explode". "You don't want rogue agents that no one has visibility of," he said. Companies will need a control plane that monitors what agents spend, their guardrails, data access and permission levels. In practice, an agent should carry exactly the permissions of the person it acts for and nothing more.

Neither firm began with a grand transformation programme, Barfield noted. Each picked a costly problem, put guardrails around it, gave users a safe place to experiment and measured outcomes in bookings and quotes. That strategy, he said, is what works.