Setting the scene. Mid-March, Honolulu. Seventy people are in a room - engineers, product managers, designers, the sales team. At the front, three customers. Not on a stage at a user conference with rehearsed talking points. Just sitting in chairs, with a moderator, telling a room full of the people who build the product what it's actually like to use it every day.

Joel runs customer support at Mosh, a digital health company in Australia. Jess leads CX at Arbor, an energy company navigating a deregulated market where most consumers don't even know how to read a utility bill. The third panelist leads CX at a US telehealth company serving millions of patients. They've been using Lorikeet for different lengths of time, in different industries, with different problems. They did not coordinate their answers.

Thank you, Moe

Joel's "wow moment" was specific. A patient - in healthcare, where people are dealing with sensitive medical issues and are often at their worst - thanked the AI agent for the service it provided. Not a five-star CSAT survey. An actual thank you, unprompted, mid-conversation.

"Especially in the healthcare space, people can be at their worst and not very nice to deal with," Joel said. "So thanking an AI bot - or not even realizing it's an AI bot - for the service provided, that's pretty huge."

The Mosh team had the data to back it up. In the first two weeks of deployment, Lorikeet's customer effort scores were higher than the human agents' scores. "Which upset our humans." The scores have since equalized - but with a caveat worth noting: the AI handles the straightforward cases while the humans handle the high-tension ones. Equal scores under those conditions is not parity. It's outperformance.

"We were just so bored"

Jess started evaluating AI support vendors at the end of 2024. She had calls with Sierra, Ada, and several others.

"We went in so excited - we really firmly believed in the power of AI support, that it could solve hard problems and take action. And we were just so bored in those calls. They were cagey, and what they thought was flashy was just not impressive."

Arbor is not a simple SaaS product. Electricity is personal - costs vary, data comes from multiple vendors operating in the dark ages, and most of the customer base has never been educated on how their billing works. "We need more love," Jess said.

When she reached out to Lorikeet, the first call was different. "My coworker and I were Slacking behind the scenes... we were just so inspired. He was asking us really great questions about our support philosophy and what we were trying to get out of it. We just felt his investment from the first minute."

The Mosh team had a similar experience from a different angle. They were choosing between Lorikeet and another vendor whose solution was out-of-the-box templates with an attitude of "here's what you get, good luck, make it work for you," whereas Lorikeet was wanting to build with us, take our feedback and build a tailored solution.

The pattern across all three companies was the same. The buying decision was not a feature comparison. It was a bet on whether the vendor would co-build with you or ship you a template.

Two features in a month

Joel had been a Zendesk customer. The contrast in velocity was immediate.

"Since we brought Lorikeet in, we've already had two separate features go live. Those have been really valuable right off the bat, communicated clearly, with us being part of test phases and our input taken into consideration. Whereas it takes me weeks to get anyone from Zendesk to actually look at something."

For Jess, the velocity showed up in what it let her team become. "It means rising above the operational minutiae and really being able to focus on strategy," she said. "Making the work exciting and worthwhile, with people really invested in Arbor." She'd seen the opposite play out at a previous startup - rapid growth led to contractors who weren't invested, degraded quality, constant turnover. AI done well is the inverse of that pattern. The team stays small, engaged, and close to the product.

The Mosh team described gamifying their workflow around Lorikeet's metrics. "The numbers I see - the percentages, the quality score, the independently resolved - that's sort of my main job, making sure that number goes up. Greener and greener." Their product team had been relying on a separate analytics tool to find friction points in the customer journey. It was being deprecated. Lorikeet already had that data.

This is the part of AI adoption that doesn't show up in vendor pitch decks. It changes the shape of the CX leader's job. Less managing headcount and attrition. More operating the system that serves your customers.

The loading screen problem

The telehealth CX leader made an observation that stopped the room. Email responses from the AI were arriving too fast.

"When you send an email, you're not expecting an instant response. I wonder how many of our clients are missing the response - they're not expecting it that quickly, so they miss the notification." Some complaints: "There's no way you looked at my account."

She referenced TurboTax's fake loading screen - it doesn't need the processing time, but the psychological impact of feeling like something is taking its time to look into your problem personally is better than an instant response.

This is the kind of problem that never surfaces in a product demo. It only appears in production, in the gap between "technically correct" and "actually trusted." And it's the kind of insight that only reaches the engineering team when the person experiencing it is sitting three meters away from them.

What they want next

All three panelists converged on the same answer: proactive support.

Jess described a cancellation workflow where the AI needs to follow up if a customer doesn't reply. "There are so many cases beyond that. If an unlinked customer reaches out and Olive informs them, and a week later it notices they still haven't completed the link - being able to proactively reach back out without the customer needing to respond first."

Joel wanted the same. "In an ideal world, Moe could reach out three months later and say 'hey, I spoke to you three months ago - just checking in.' You're generating the ticket. It means providing a service that no one else is."

The telehealth leader wanted to go further - AI that educates customers on how to talk to their insurance company, how to ask the right questions, how to navigate a system designed to deflect them. "Insurance companies are trained to not give you the answer you're looking for unless you know the exact right question to ask."

Lorikeet is a distributed company. London, New York, San Francisco, Sydney. The product ships around the clock and the agents never sleep. That's the operating model, and it works. But last week in Honolulu, three customers told seventy engineers what they needed next - and the roadmap shifted before anyone left the room. The offsite was over. The work had already started.