Mike HeapMike is an experienced Product Manager who focuses on all the “non-development” areas of My AskAI, from finance and customer success to product design, copywriting, testing and more.
Table of Contents
- What does the provider do?
- Which helpdesk does the provider use?
- How did the provider train their AI customer service agent?
- How did the provider turn on 'direct replies' to customers?
- What was the biggest thing the provider did to improve their AI agent's resolution?
- How does the provider customize their AI agent setup to work for their business?
- Setting the tone and the escalation rules with Guidance
- Giving agents an AI Copilot inside their CRM
- Resolving order questions end-to-end
- What impact is the provider's AI customer service agent having now?
- Where does the provider go from here?
Created time
Jun 3, 2026 09:14 PM
Title length (<60)
Author
Ecomm?
Tags
Image
Publish date
Jun 3, 2026
Slug
telehealth-ai-customer-support-case-study
Featured
Type
Case study
Ready to Publish
Ready to Publish
A telehealth provider resolves around 72% of its monthly Gorgias support tickets with AI, holding a 75% AI CSAT score and saving roughly 142 hours every month.
Shipping prescription treatments to people's doors generates a very specific kind of support queue. The questions are exactly what you'd expect (where's my order, when does it ship, am I eligible, what happens next), they arrive on both chat and email, and they land in a small team that was growing faster than it could hire.
Today our AI agent handles the bulk of that queue directly inside Gorgias. It resolves around 72% of those tickets without a human, holds a 75% AI CSAT score, and hands the team back about 142 hours every month, off the back of a help center the AI largely wrote itself.
Here's how it came together.
What does the provider do?
The provider is a direct-to-consumer telehealth and online-pharmacy business. Customers get an online consultation, and prescription treatments are shipped to their door, across several consumer-health categories.
It's a Series A company, and like a lot of fast-growing healthcare brands, its support volume climbed faster than headcount could keep up with. Half the queue is ecommerce questions (where's my order, can I change my address, when's my next delivery), and the other half needs the careful, sometimes-human handling that anything health-related deserves.
Which helpdesk does the provider use?
They run support on Gorgias, across both chat and email. Our Gorgias AI agent plugs into that same inbox, so there was nothing to rip out and no new tool for the team to learn.
The thing that tends to matter most on a queue like this is how the AI is priced. When most of your tickets are repetitive order and FAQ questions, you resolve a lot of them, and the gap between paying per ticket and paying per resolution gets wide fast.
We charge per ticket (roughly ten cents), rather than per resolution, which keeps the economics sane when the AI is doing most of the answering.
How did the provider train their AI customer service agent?
Here's the part I like most about this one. They didn't have a mature help center to point the AI at, so the AI built one for them.
We used Train on Historic Tickets to auto-draft help articles from their past tickets, learning from how real questions had actually been answered. The team reviewed those drafts before any went live, so the agent started from their own answers rather than a thin set of generic articles.
From there, Self-Learning kept the knowledge growing. It drafts new articles by comparing the AI's reply to the human agent's actual reply on handed-over tickets, so the agent quietly widens what it can answer over time (you can read more on how self-learning works). To date, those auto-drafted articles have answered more than 3,000 tickets between them.
How did the provider turn on 'direct replies' to customers?
They run the AI as a direct responder on both channels in Gorgias. A self-built help center plus clear rules on when to escalate gave them the confidence to let it reply to customers directly, rather than sitting in a notes-only mode forever.
Chat answers can be shorter and faster, email answers a little more thorough, and it's the same agent doing both.
What was the biggest thing the provider did to improve their AI agent's resolution?
For an online pharmacy shipping physical product, one question dwarfs all the others: where's my order. So the single highest-leverage thing they built was a Task that checks order status automatically.
A Task is an agentic workflow, and here's the thing: instead of a human looking up an order and pasting the status back, the AI recognizes the question, calls the provider's order system itself, and answers with the live status.
That turns the most common ticket in the queue from a manual lookup into something the AI resolves end-to-end. Underneath it, the boring-but-effective Self-Learning keeps widening the range of things the agent can handle, and together they hold a 72% resolution rate steady.
For context on that number, 72% is solid. Across a large field of AI support deployments the median resolution rate sits around 70%, and this provider is already a little above it before they've even connected live customer data (more on that in a moment).
How does the provider customize their AI agent setup to work for their business?
A healthcare brand can't run a generic bot, so they tuned the agent around a few things in particular (the small settings that, in our experience, end up doing most of the work).
Setting the tone and the escalation rules with Guidance
They configured two kinds of Guidance. Communication guidance sets the brand voice and sign-offs, so the AI sounds like them.
Handover and escalation guidance sets the rules for when the AI should stop and pass a ticket to a person, which for a healthcare brand is the part you really cannot get wrong.
Giving agents an AI Copilot inside their CRM
Not all of their work lives in Gorgias. So our team built them a custom Copilot inside their CRM, so the AI keeps agents efficient even on the work that happens outside the helpdesk.
Resolving order questions end-to-end
The order-status Task we built is the other piece of ongoing tuning. As their systems and policies change, that workflow keeps the most common question in the queue resolving on its own.
What impact is the provider's AI customer service agent having now?
The numbers from the last 30 days:
- 72% AI resolution rate
- ~2,600 tickets handled per month (of which roughly 1,870 are resolved by the AI with no human involved)
- ~142 hours saved per month (at roughly five minutes of agent time per resolved ticket)
- 75% AI CSAT across the tickets the AI handled
- 3,000+ tickets answered to date by the help articles the AI drafted for itself
The hours-saved figure is the one I always point to. 142 hours a month is the best part of a full-time agent the team never had to hire, on a queue that would otherwise keep growing with order volume.
Where does the provider go from here?
More Tasks and Tools would let the agent move from answering to doing: changing a delivery address, arranging a re-ship, or updating a subscription, through their own systems with the right approvals (and the clinical guardrails kept firmly in place).
I'll be straight about what 'fully automated' looks like for a healthcare brand, because it's easy to oversell. It means the AI owns the big repetitive core (the order questions, the eligibility basics, the same FAQs arriving all day), while the team keep the clinical judgment and anything sensitive.
That's the split they've built, and it's why a 72% resolution rate comes with a 75% satisfaction score rather than at the expense of one. If you'd like more stories like this one, browse all our case studies, see how Edel Optics and YouGarden run the same playbook, or check our pricing to model what this would cost at your own ticket volume.
