How Medtelligent achieved a 60% resolution rate with Duckie, turning five years of support knowledge into an AI agent in two hours

Sam Wacker has managed Medtelligent's customer success team for five years. His team handles a wide spectrum of tickets: complex pharmacy integration issues, assessment configurations, financial reporting questions, and the deep clinical workflows that keep assisted living communities running safely.

Those are the tickets his team loves working on. But a large portion of the volume was something else entirely.

"A lot of the issues are just login issues," Sam says. "You want to spend as much time as you can on the complex things, and you get taken away from those by run-of-the-mill everyday questions that just require one email response. Point somebody in the right direction, give somebody a piece of information."

The real cost was not the time per ticket. It was the context switching. An engineer deep in investigation of a pharmacy integration problem would get pulled out to solve a login reset. Then they had to find their way back into the original issue. Multiply that across a growing client base of 500+ communities, and the pattern becomes unsustainable.

"It deters from the overall ability to stay focused, do deep work on a specific subject," Sam says. "Constantly getting taken away to solve first-surface-level issues."

Medtelligent uses Zendesk as their helpdesk, and when the market started talking about AI support, they reached for the tool already in their stack. The experience was the opposite of what Sam expected.

"We had been trying to implement the Zendesk AI solution and it was very manual, required a lot of work. You have to build out a bunch of trees and workflows. It wasn't what I thought AI was supposed to be."

The implementation meant sitting in a configuration page with a flowchart map, picking one topic at a time, and filling out decision trees branch by branch. If the customer says this, then do that. If they say that, then say this. One resolution tree took four to five hours of work.

And when something went wrong? "If I have a problem with my Zendesk account, I get sent some help desk articles and a good luck."

Sam needed something that worked the way AI was supposed to work: you talk to it, it helps you, and it gets smarter over time.

The difference was obvious from the start. Where Zendesk required Sam to manually map every decision branch, Duckie let him describe what he needed in plain language, and the system built the workflows for him.

"I would go in and say, my customers are reaching out about these issues and this is how I would want Duckie to respond. And the assistant would tell me what I should do. It would tell me what I could do manually, but it would also tell me that it could do it for me."

Duckie still gave Sam the manual control he wanted for sensitive workflows. But for most things, the AI implementation engineer proposed the workflow, Sam clicked approve, and it was done.

"Whereas Zendesk took me four to five hours to build out a resolution tree, with Duckie it's like 30 minutes at most if I really need to get my hands into something."

The other difference was responsiveness. Medtelligent's setup is unusual: each client community has its own individual website. Getting the Duckie messenger widget deployed across every site required a platform-level change.

"There was a change that Duckie needed to make and they made it same day. Duckie has a team of people there to help me, care about my issue, understand my issue. The team is not being pulled in a bunch of different directions. The focus is this AI agent and AI support."

The moment that convinced Sam was onboarding. Medtelligent had 12,000+ Zendesk tickets, five years of back-and-forth conversations containing the institutional knowledge of how the support team actually solves problems. Not help desk articles, but the real, messy, context-specific ways an experienced rep unblocks a frustrated nurse at 2 a.m.

Duckie ingested all of it in two hours.

"We met with the Duckie team on a Thursday afternoon, maybe 3:00 p.m. Central. We set up the Zendesk integration. That run to learn all of the Zendesk history probably took two hours. When I came in on Friday morning, it was ready to go."

Sam started testing in the playground immediately. The agent already knew how Medtelligent talked, the tone, the level of detail, the way they explained medication workflows. He felt comfortable with it responding to clients almost right away.

"It would take a human being forever to go through every single one of those 12,000 tickets and pull out the nuggets of good information. Duckie did that in two hours. Those interactions were valuable and meaningful at the time, but Duckie has highlighted that and taken all of that information and knowledge and condensed it into itself."

Medtelligent is not a typical SaaS company. Medical data flows through their system. Financial information flows through their system. The people asking for help are caregivers responsible for vulnerable populations. The margin for error in how an AI handles this information is zero.

Sam configured Duckie to use only first initial and last name in responses. If a client includes personal health information in their message, Duckie redacts it from the reply. These are not edge cases for a healthcare platform. They are the baseline.

"I asked it to handle sensitive things, like don't relay specific information about a resident or personal health information. It added that into the workflow and I've not seen any issues with that."

But beyond data handling, the quality that matters most to Sam is tone. Medtelligent's clients are people who take care of the elderly. The support team tries to match that energy with kindness and patience in every interaction. Duckie reflects that.

"The thing that has impressed me is the clients we work with are some of the most kind human beings. They take care of our elderly population. And Duckie has reflected that. It gives honest answers. If it doesn't know something, it tells them. It's not trying to force-feed answers. It knows its own limitations."

When a client writes in frustrated or angry, Duckie does not match the emotion. It takes in the issue, focuses on resolving it, and responds with care.

"I do feel it does a good job of being an extension of what my team is all about. I feel comfortable having it take in this sensitive information and handle it with care."

Sam launched Duckie expecting to hit maybe 20-30% resolution. The actual number: 60%.

"It has far surpassed that," Sam says. "And the other thing is, you think about the benefit to your team, but it's a benefit to our clients too. Clients don't want to be waiting to get their login issue fixed. Now I have clients who I think prefer to speak to Duckie than to get to an agent."

When Duckie does escalate, the handoff is clean. The full conversation history lands in a Zendesk ticket, formatted the same way the team already receives messages. No new workflow to learn. And the ticket arrives pre-investigated: who is the client, what community do they work with, what is the issue, and what context has already been gathered. The back-and-forth that used to eat up the first three replies of every human interaction is already done.

"Not only are we saving that frustrating back and forth for the client, we're getting that information and able to grab that ticket and run with it right away."

One of the unexpected benefits came when a member of Sam's team was working on implementation. During the implementation process, she used Duckie's Playground and batch testing feature to ask questions the way a real customer would. Next, she asked the assistant to surface areas where the Duckie agent was struggling to handle customer-facing issues. After the AI assistant surfaced its findings, she was able to tell the assistant how she wanted to address the gaps in documentation and across the entire Duckie system.

Now the process is ongoing. When Duckie cannot answer a specific issue more than a handful of times, the team knows they need to create supporting documentation, both so Duckie can handle it next time and so clients have the resource independently.

Zendesk's reporting gave raw data. Duckie gives direction.

"With Zendesk, it's all on you. You figure out what you need to do. Whereas with the Duckie assistant, I can literally ask it where the highest volume of escalations is coming from and it will tell me and present that information to me on a platter."

Something Sam did not expect: the AI implementation engineer became a daily tool for his own work, not just for managing the agent.

"I realized I can just use it like any other AI platform. I ask it how to help me pull metrics not only from Duckie but from Zendesk. I can have it do Duckie things, but I can also have it help me soundboard and work through issues. It's like having one platform where I can have this AI agent help my customers but also help me."

Medtelligent's client base is growing. Sam's goal has never been to scale support by adding headcount. It's been to build a team that has the time to go deep.

"My team gets to spend more time on the nuanced and complex stuff. I'm building a team that is more knowledgeable, more experienced, has more time and capability to focus on higher priority things. That builds, you know, my goal for my team is that they have time to work on things they enjoy, that is fulfilling, that is servicing our clients in a deeper way."

As Medtelligent grows, Sam is focused on something harder than automation: maintaining a support experience that feels human in a world that is becoming more digital.

"I'm really trying to marry that concept with AI. Creating a support experience that feels like you're cared about, listened to, and that there's a human on the other end that can help you out."

For Medtelligent, that balance is working. Duckie handles the volume. The team handles the complexity. And the clients who care for the country's elderly population get the responsive, knowledgeable support they deserve.