How GE HealthCare Finds The System Failure Hiding Behind A Single Complaint

A complaint usually arrives as a single case to close. Refund this order, re-ship that box, apologize to this caller, and move on. What’s easy to miss is what that one complaint says about everyone who hit the same problem and never picked up the phone. That question gets harder when an AI virtual agent is answering the calls, and in medical equipment service, the caller on the line may have a patient waiting.

Burgoyne Hughes is Senior Director, Customer Service Centers at GE HealthCare, where he owns the customer service center, call centers, and repair depots across the US and Canada. CX Current named him to its 2026 Top 50 Leaders Transforming CX as an "Operating Architect," its category for people rebuilding how a service organization runs underneath what customers see. Hughes spent two decades in service operations at Staples, Teleflora, OfficeMax and the online costume retailer BuySeasons before joining GE HealthCare, and he draws on all of it when a complaint reaches his team. Under his leadership, the center has cut hold times by 60%, brought its error rate from 1.6% to 0.04%, and finished first in its size category in BenchmarkPortal's 2024 ranking.

The earliest sign that a process is breaking tends to surface on the front line and in a customer's own words, well before it reaches a report. "Let the people who are closest to the work make things work better. The person who is on the phone with the customer will understand more about our systems and customers than anybody else will. Instead of trying to go top-down on how to make our process better, we need to be learning from them and having them help guide us into how we fix the holes," says Hughes. The virtual agent is the newest place that principle gets applied, and it sits alongside the resolution-first model he has described for healthcare service.

The tip of the iceberg

Early at Staples, Hughes handled the complaint letters addressed to the president. Some of those complaints were settled with store credits, which printed at a distribution center and shipped like any other order. Then the company moved locations, and that printer stayed unplugged. "All of these things were in the system and they weren't printing out," Hughes recalls. "Literally thousands of these things had never gone out, but two customers took the time and were frustrated enough to write a letter to say this is happening."

Years later at BuySeasons, survey scores and emails kept flagging the same thing: the boxes were too big for what was inside them. Carriers bill by package size as well as weight, so an oversized box of light party supplies cost more to ship than it had to. The fix was a machine that measured each packed order, cut the box down to fit and resealed it. UK carriers had billed by size for years, but the equipment was in use nowhere else at the time. "We were wasting a ton of money and we were frustrating customers. They got this giant box with a little bit of stuff in it," Hughes notes.

In both cases, the wasted money kept adding up, because a few complaints looked like isolated incidents. Hughes came away from both jobs asking how many customers hadn't called. "When you hear complaints from customers, you have to know that it is the tip of the iceberg. Hearing that a couple of times tells you this is happening way more often than what we're hearing it, because most customers are just going to go on about their day and they're not going to take the time to tell you what the problem was," he explains.

Designing the handoff

GE HealthCare's virtual agent went through a multi-week user acceptance testing period, with the operations team and the IT quality assurance team working it at the same time. The point of that stretch was to find, before any customer did, the problems that would otherwise reach callers and mostly go unreported. Hughes credits the tally that came out of it with the launch holding up, and teams that skip the phase tend to meet the same problems in production, which is why ownership of an AI agent gets settled before deployment. "We found almost 600 issues that either had to be corrected or descoped as we went through that process, and any one of those could have potentially derailed the service for customers along the way," says Hughes.

Most of the team's design time went to the handoff from the virtual agent to a person. In the research they reviewed, the main reason customers give up is being transferred and then having to explain the problem again from the start. A handoff that strands the customer is where deflection goes wrong, and Hughes's team wrote a script for the first ten seconds after the transfer. "If we've got a customer who says they have a patient on the table in the AI, when it comes to a live agent, they're saying, 'I understand you have a medical emergency there, I'm going to get you help as quickly as possible,'" he adds.

The other half of the handoff happens in the software, and Hughes expects it to be where most rollouts fall down. A transfer code tells the agent which conversation they are picking up. "Not only is the information collected and sent over, but we'll create a skeleton service request in the background so that the agent doesn't even have to re-enter any of the data. It's already there, and so it really speeds up that transition," he says.

Feedback as the next input

Hughes has his team collecting customer feedback on the virtual agent three ways at once. Calls the AI completes on its own end in a survey. When a call is transferred, the agent logs what the customer said about the experience in the knowledge management system. A human agent in the US has also been phoning customers since launch to ask how it went. The first problems that came back were small and concrete: medical and GE HealthCare terms the AI misheard, and email addresses it couldn't capture. "We take all of those data points and we look at what's frustrating customers," Hughes notes.

Acting on what customers report takes longer in healthcare than in the industries Hughes worked in before. In e-commerce, order values were lower, the sector was unregulated, and a fix could go live in days. GxP quality rules mean any change touching a GE HealthCare customer runs through a documented approval process, a constraint other healthcare CX teams will recognize. "Our ability to iterate, even with this launch, is slower because we have to go through more quality assurance checks with more people," says Hughes. "There's such a tremendous risk if you get it wrong, and you shouldn't move quite as quickly."

Phase one of the rollout covered contracted customers opening a service request, the heaviest volume in the queue. Next come customers who pay by purchase order, then calls about using the equipment instead of repairing it, then service requests from biomed teams. Every phase ahead goes to the agents first, the way teams coaching an AI agent like a new hire find its weak spots. "They're going to recognize those gaps causing service failures before anybody else does, and they're going to be able to tell you," Hughes concludes. "Make sure that, as you go to a new system or a new process, you've got those people and they have an easy process to get that feedback to you."