Most companies that have poured money into generative AI have little to show for it. An MIT study last year found that about 95 per cent of enterprise pilots produced no measurable effect on profit and loss.
Projects stall at proof of concept, staff work around the tools they are given, and the data the models need sits in systems nobody has cleaned up.
In Asia-Pacific, a handful of organisations have broken out of that pattern. Bumrungrad International Hospital in Bangkok and The Grout Guy, an Australian tiling and re-grouting business, have little in common on the surface. One delivers clinical care under heavy regulatory scrutiny; the other sends tradespeople into suburban bathrooms.
Both went about AI the same way. They started with specific bottlenecks, sorted out their data and guardrails before letting agents loose, and judged the results by revenue rather than by the number of pilots launched.
One of the largest medical facilities in Southeast Asia, Bumrungrad International Hospital is not new to AI.
It used the technology in radiology for more than four years. Its current push is into hospital operations and the paperwork that eats into doctors’ time, said James McLeary, the hospital’s chief innovation officer and chief information security officer.
In consultation rooms, the hospital moved its electronic health records onto an AI-enabled platform, IntelliCare, which listens to the conversation between doctor and patient and turns it into structured notes, he told Techgoondu on the sidelines of Dreamforce, a Salesforce event, in September.
The hospital holds one firm line, stressed McLeary. No AI output, whether a consultation note or a radiology assessment, goes into the record until a physician signs it off. “We don’t allow any AI to make the clinical decision.”
Another early move was to deal with doctors who were already using AI without permission. Frustrated by administrative bottlenecks, some physicians had installed unsanctioned tools such as OpenClaw on their own devices.
A ban would only have pushed the practice out of sight. “If you don’t allow it, it will happen anyway,” McLeary said.
Instead, the IT team 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,” said McLeary.
This arrangement turned the hospital’s most impatient users into a source of ideas, and gave IT a steady supply of use cases that clinicians actually wanted.
The financial return also shows up in the contact centre. Patient e-mails are summarised, sorted into three tiers and routed immediately. Agents on WhatsApp and Line answer routine questions about prices and promotions and pass complicated cases to staff.
Multilingual voice agents able to handle other languages are now in the works, said McLeary. The hospital is also going back to its list of dropped calls, turning them into confirmed bookings, thus capturing demand the hospital used to lose.
The Australian grouter
Australian company The Grout Guy took a rougher road. Based in Perth, the business has grown from 30 staff in Western Australia to nearly 200 across six states, said chief technology officer Anthony Messina.
Its first attempt at a systems overhaul failed because the outside partner it hired did not understand how trade work runs. The Grout Guy mainly serves customers living outside urban areas.
Messina explained that customer journeys built for digital businesses do not fit a job that involves a van, a tradesperson and a wet bathroom floor.
Before setting out, technicians call ahead to make sure the customer will be home. A wasted trip to the outer suburbs costs time and opportunity.
“Australia is big, so travel time is a factor,” said Messina. “Unlike other enterprise services that customers can sign up for on a, mobile phone or laptop, grouting and tiling needs a technician on site, so the customer has to be there too.”
The company started again, launching a mobile app built around a single source of data before adding AI on top.
Homeowners now upload photos of damaged tiling to the app, and computer vision assesses the condition of the job. That lets the company prioritise the work according to level of importance and revenue potential.
Technicians snap photos of the site before and after work to record what they have done. Once the job is completed, a service report and invoice is generated, the customer pays.
Field technicians and sales staff work from one mobile app with AI agents built in. One, named Walter, processes PDF work orders around the clock. Another, Penny, chases unpaid invoices. A third, Mason, handles cancellation and rescheduling emails that arrive after hours.
The work is recorded in the cloud. This strategy means that the company does not have to increase administrative staff its office as it expands throughout the country.
The gains came from speed: how quickly the company responds to a lead and how quickly it sends a quote. Quote conversion rose from 30 per cent on the old software to 50 per cent with the agents. The company credits that with helping drive average revenue growth of 40 per cent a year over five years.
The management team has since spun its technology arm out as a separate company, Servly, which sells 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.
The harder part comes next
Success brings its own problem. As agents spread through a business, keeping track of them gets difficult.
Gavin Barfield, field CTO at Salesforce Asean, expects the number of agents in a typical enterprise to “explode”, and says companies will need a control plane that can find every one of them.
“You don’t want rogue agents that no one has visibility of,” he said. As the number of AI agents grow, “all those things that we do for humans have to apply to agents, such as monitor what they’re spending, the guardrails they are using, the data they have access to and their permission levels”.
In practice, that means an agent should carry exactly the permissions of the person it acts for and nothing more. It also means companies need not wait years for a perfect central database.
The Grout Guy’s experience suggests that cleaning up the data for a few high-value tasks, and expanding from there, gets results sooner.
Neither Bumrungrad Hospital or The Grout Guy organisation began with a grand transformation programme. Each picked a problem that was costing money, put guardrails around it, gave users somewhere safe to experiment and measured the outcome in bookings and quotes, said Barfield.
This strategy appears to be what works.
