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 YesLMS do?
- Which helpdesk does YesLMS use?
- How did YesLMS train their AI customer service agent?
- When did YesLMS decide to turn on 'direct replies' to customers?
- What was the biggest thing YesLMS did to improve their AI agent's resolution?
- How do YesLMS customize their AI agent setup to work for their business?
- Tone and communication style
- Escalation and handover
- Custom Answers for the recurring policy questions
- What impact is YesLMS's AI customer service agent having now?
- Where do YesLMS go from here?
Created time
May 26, 2026 05:23 PM
Title length (<60)
Author
Last optimised
Ecomm?
Tags
Image
Publish date
Jun 18, 2026
Video
Slug
my-askai-yeslms-case-study-2026
Featured
Featured
Type
Case study
Ready to Publish
Ready to Publish
YesLMS resolves ~228 of ~300 monthly Zendesk tickets with AI, holding 88% AI CSAT and saving ~17 hours every month.
The team behind YesLMS sits in a spot a lot of small support teams will recognize. A modest steady tail of "how do I" tickets every month: instructor invites, certificate re-issues, plan and seat queries, the usual long catalog.
They are a a small ops team and have customers whose own product promise is accessible, inclusive learning, so every reply has to feel both fast and on-brand.
That kind of tail is hard to fix with more humans (the volume doesn't justify another full-time agent) and hard to fix with a generic bot (the customers are sharp, and an off-tone reply lands badly). It's the gap that's tailor-made for an AI customer service agent, if you can get the trust and the training right.
Today, our AI inside Zendesk handles around 228 of those tickets every month at 88% CSAT, freeing the YesLMS team about 17 hours a month and covering both Zendesk Tickets and Zendesk Messaging side by side.
Here's how it came together.
What does YesLMS do?
YesLMS is a cloud-based learning management system built around one core idea: accessibility first. The platform pairs a traditional LMS with built-in conferencing, and it's built on Universal Design for Learning principles.
Organizations that prioritize accessibility, including the World Institute on Disability, rely on YesLMS because accessibility is built into the platform from the ground up. Unlike many learning management systems that add accessibility features after the fact, YesLMS was designed with accessibility as a core principle from day one.
YesLMS primarily serves nonprofits, businesses, and government agencies for employee onboarding, continuing education, professional development, and certificate management workflows. Its partners and clients include GWU, SDSU, Octopi Brewing, PolicyWorks, the National Disability Institute, CSAVR, and 45 state social service and employment programs, including vocational rehabilitation agencies.
Which helpdesk does YesLMS use?
YesLMS runs customer support out of Zendesk, using both Support Inbox (Tickets) for email-driven asks and Zendesk Messaging for in-product chat. The same My AskAI agent sits in both channels with the same training, the same Guidance, and the same Custom Answers.
So a learner who pings them in Messaging and a course admin who emails the inbox both end up talking to the same brain.
At this volume the two-channel coverage matters more than it would at consumer scale. Either channel on its own is modest (about 300 tickets a month between them), but the response-time bar is high.
Training admins are usually answering to their own internal stakeholders, so a slow reply on something like a seat-allocation question can stall a whole cohort's start date.
If you want to install our agent the same way YesLMS did, it's listed on the Zendesk Marketplace for Support and separately for Zendesk Messaging. Each install takes a few minutes.
How did YesLMS train their AI customer service agent?
Training started with the Zendesk help center, plugged in directly via the Zendesk Knowledge Base connector. For an accessibility-first LMS that help center is the biggest single piece of the puzzle: course-builder how-tos written for screen-reader users, instructor-invitation walkthroughs, certificate workflow docs, plan and seat policy pages.
The native connector means new and updated articles flow into the AI's knowledge on their own. No separate sync to babysit.
On top of the help center, we layered in Custom Answers: the boring-but-effective option for the dozen policy situations every LMS support team hits over and over. Custom Answers lock the exact wording for those situations.
That matters when the policy needs to be precise: rules around certificate re-issue, free-trial extensions, organization-admin role transfers, and gradebook export are easier on both sides when the AI says the same thing every time.
Then Self-Learning. Our Self-Learning watches AI-handled tickets, spots the recurring questions the existing help center doesn't answer well, and drafts new articles for those spots.
A human on the YesLMS side reviews and approves the drafts before they enter the knowledge base. On this account, those auto-drafted articles fed about 200 ticket responses in the last 30 days (a quiet but compounding effect on resolution).
One thing they haven't added to the training mix yet: the User Data API. There's no account-level customer data flowing into replies right now, and that's a deliberate hold while the help-center coverage matures.
When did YesLMS decide to turn on 'direct replies' to customers?
The team flipped direct replies on in both Zendesk Tickets and Zendesk Messaging from the live rollout. Same agent, same Guidance, same scope, just two channels at once.
Going direct in both straight away was simpler at this volume than running one in shadow mode for a few weeks: 300 tickets a month means the team can see every AI-resolved ticket and review the outcomes weekly without a backlog forming.
At this scale, we find small teams catch problems quickly because they're already reading every ticket. There's no need for a multi-week internal-notes phase to build comfort.
They need the AI to do the obvious wins from day one and the team to keep an eye on the edges.
What was the biggest thing YesLMS did to improve their AI agent's resolution?
If I had to point at one lever doing the heavy lifting on this account, it's Self-Learning.
In our setup, the Zendesk help center got the AI to a decent baseline. Custom Answers locked the recurring policy replies.
But the gap between "answers the FAQs" and 76% resolution lives in the long tail: questions where the help center has a page that's nearly right but not quite, or where the answer needs combining two articles, or where the situation is common enough to be worth a reply but rare enough that nobody has written it down yet. That's the gap our Self-Learning closes.
The mechanics here are simple. As the AI handles tickets in Zendesk, Self-Learning watches for the patterns: questions the AI couldn't answer well from the existing help center, questions where the AI gave a partial answer and a human had to fill in the rest, questions where the customer asked the same thing twice in slightly different ways.
It drafts new help center articles for those patterns and surfaces them for a human to approve. Approved drafts then feed back into our knowledge layer and improve replies on the same topic going forward.
Over the last 30 days on this account, we've seen those auto-drafted articles feed about 200 ticket responses. That's roughly two-thirds of the AI-handled tickets benefiting from Self-Learning-sourced knowledge.
Not from articles the team had to manually spot and write, but from articles our system suggested based on what learners and admins were actually asking. The cumulative effect is the climb into the 76% range the team holds today.
Self-Learning works alongside Guidance, with each doing a different job. Communication Guidance shapes how the AI replies; Context & Clarification Guidance tells it when to ask a follow-up question instead of guessing; Handover & Escalation Guidance tells it when to stop and route to a human.
All three are configured on this account. Self-Learning closes the gap between what the help center covers and what learners actually ask; Guidance shapes how our AI behaves until that gap closes.
How do YesLMS customize their AI agent setup to work for their business?
Beyond the core training, we've helped the team tune the agent in three specific ways.
Tone and communication style
YesLMS's whole brand is built on accessible, inclusive communication. Communication Guidance lets the team set the rules for how the AI replies: sentence structure, sign-off, when to use lists, when to use links instead of long paragraphs.
For a workforce-training LMS where the people on the other end are themselves teachers, trainers, and accessibility advocates, the bar for "sounds like us" is higher than it would be at a generic SaaS company. We treat Communication Guidance as the lever that enforces it.
Escalation and handover
Some ticket types should never be auto-resolved, no matter how confident the AI sounds. Handover & Escalation Guidance is where YesLMS draws those lines.
Requests that touch certificate verification, learner data export, instructor account ownership transfer, or organization-admin role changes route to a human with the AI's context attached. I think of this as the AI's "stop and pass it on" switch.
The AI prepares the ticket (pulls the relevant background, flags the customer's plan and the specific request) and hands off cleanly. The human picks up faster because the prep work is already done. And the customer doesn't get a wrong answer because the AI tried to handle something it shouldn't have.
Custom Answers for the recurring policy questions
LMS support has a fat middle of repeating questions: how do I extend a free trial, can I transfer a course between admins, is SSO available on my plan, how do I export a gradebook, can I re-issue a certificate. Custom Answers locks the exact policy reply for each of those.
When the question comes in (phrased one of fifteen different ways), the AI matches it to the right Custom Answer and replies with the canonical policy wording. The win here, in our experience, is consistency: the same question gets the same answer, no matter which day of the week the ticket lands.
What impact is YesLMS's AI customer service agent having now?
Last 30 days, here are the scores on the doors:
- 76% AI resolution rate, across Zendesk Tickets and Zendesk Messaging combined
- 88% AI CSAT across AI-handled tickets
- ~300 tickets/month handled by the AI
- ~17 hours/month saved
In our experience, the 88% CSAT is the number that travels furthest. Resolution rate tells you the AI is doing work; CSAT tells you the work is good.
For an LMS team whose customers expect inclusive, careful replies, 88% is a strong floor. It means the AI's wins are clean wins, where the customer ends the conversation happy. (Another Zendesk team we work with, Apartment List, holds the same 76% resolution rate at higher volume with a 97% CSAT, which is a useful upper-bound for what's reachable on Zendesk once volume gives Self-Learning more to chew on.)
The ~17 hours back per month free the human team to do the work the AI can't: the multi-step instructor escalations, the accessibility-feature deep dives, the conversations that turn a confused first-time admin into a confident one. That's the work that earns renewals (and the word-of-mouth that sells more workforce-training software). At our per-ticket pricing, the maths on freeing those hours tends to make itself.
Where do YesLMS go from here?
The most obvious next lever we'd point at is the User Data API. Right now the AI replies from help center content, Custom Answers, and Self-Learning-sourced articles.
But it doesn't yet know which customer it's talking to. Plugging in the User Data API would let it pull the requesting customer's plan tier, seat count, billing status, and organization type into every reply.
That changes what the AI can answer accurately: "Is feature X on my plan?" becomes a real answer rather than a deflection to the pricing page. "How many seats do I have left?" becomes a specific number rather than a link to the admin dashboard. (Edel Optics jumped from 25% to 79% AI resolution the week they plugged in the same User Data API on a different account.)
The natural step after that, in our view, is Tools and Tasks: agentic actions the AI can take inside a reply. For a workforce-training LMS, the obvious candidates are certificate re-issue, course transfer between admins, seat top-ups, instructor invitations, and organization-role changes.
Each of these is a frequent ticket today that ends in a human clicking a button in the admin UI. Turning the button-click into an AI-driven action is how the time saved compounds.
We're not chasing 100% automation here. For a platform where instructor-level questions can be genuinely complex, humans stay in the loop for the deep stuff.
Our job is to free them up for the work that drives renewals and word-of-mouth: the conversations that sell workforce-training software. The 76% resolution the team holds today, with Self-Learning still closing fresh gaps every month, suggests that lane is already wide open.
If you'd like to see more customer stories like YesLMS's, browse all our case studies here.
