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 Apartment List do?
- Which helpdesk does Apartment List use?
- How did Apartment List train their AI customer service agent?
- When did Apartment List decide to turn on 'direct replies' to customers?
- What was the biggest thing Apartment List did to improve their AI agent's resolution?
- How do Apartment List customize their AI agent setup to work for their business?
- Controlling their language and communication style
- Setting clear escalation triggers
- Using AI Tagging to control where AI replies
- What impact is Apartment List's AI customer service agent having now?
- Where do Apartment List go from here?
Created time
May 15, 2026 10:16 AM
Title length (<60)
Author
Last optimised
Ecomm?
Tags
Image
Publish date
May 15, 2026
Video
Slug
my-askai-apartment-list-case-study-2026
Featured
Featured
Type
Case study
Ready to Publish
Ready to Publish
Apartment List resolves 76% of ~1,582 monthly Zendesk tickets with AI, holding 97% AI CSAT and saving ~101 hours every month.
Apartment List's support team runs a high-volume Zendesk inbox: a B2C marketplace operation where every percentage point of AI deflection is real hours back for the team.
They layered us on top of their existing Zendesk setup, and today they resolve 76% of ~1,582 monthly tickets with AI, hold a 97% AI CSAT, and save roughly 101 hours of agent time every month.
Here's how it came together.
What does Apartment List do?
Apartment List is a US-wide rental marketplace. Renters can browse listings, get matched with places that fit their preferences, and find home using their platform. Listers can attract, convert, and keep renters through their AI-driven listing platform.
Apartment List has been helping renters find homes since 2011 and now cover most major US metros with millions of monthly visitors. At that scale, support volume isn't optional, and the patterns repeat week after week: payment confirmations, application status lookups, profile edits, listing questions, landlord-renter messaging. The same dozen-or-so questions, over and over.
That's the kind of support load AI is built for. And it's the load Apartment List set out to deflect.
Which helpdesk does Apartment List use?
Apartment List runs support inside Zendesk: specifically the Support Inbox / Tickets channel, which is the default for B2C marketplaces at their scale (and the helpdesk we see most often inside this category). Every renter or landlord question lands as a Zendesk ticket. Every reply, AI or human, goes back through that same ticket thread.
That's where we plug in. Our Zendesk Tickets integration sits inside the Zendesk inbox, reads incoming tickets, drafts (or sends) replies, applies tags, and hands off to humans on the cases that need them.
Apartment List didn't migrate off Zendesk to add AI; they didn't need to. Zendesk stays the system of record, and we're the layer doing the resolution work on top of it. (Generally, this is the boring-but-effective option: keep your helpdesk, add AI where the volume is.)
If you want to model what this would look like for a Zendesk support team of your own size, our Zendesk Tickets ROI calculator is a quick way to plug your own numbers in.
How did Apartment List train their AI customer service agent?
They pointed it at the knowledge they already had, in the order that made the biggest difference soonest:
- Zendesk help center. First connection in: our Zendesk Knowledge Base connector pulled every public help-center article into the AI agent's knowledge. This is the highest-density source of answers for any Zendesk-running team, and Apartment List had a substantial library to draw from.
- Public support website. Next, our unlimited website sync was switched on against their support pages, covering the answers that live outside the formal help center: FAQ pages, policy pages, landlord-facing instructions, renter onboarding guides.
- Supplemental CSV upload. Some structured knowledge didn't fit either the help center or the website: internal lookup tables, policy matrices, fee schedules. They uploaded a CSV directly as an extra knowledge source, filling in what the connectors couldn't see.
- Existing Zendesk macros, as Custom Answers. This was the single highest-leverage knowledge move (more on this in the next section).
Their human agents had been using macros for years: pre-written agent replies for the high-volume patterns of a rental marketplace, including rent payment confirmations, application status lookups, profile-edit guidance, listing-removal requests, and landlord-renter messaging questions.
Rebuilding those macros as Custom Answers gave the AI an instant library of brand-voice-correct, factually-tight scripted replies. The macros had already been trusted by agents, so turning them into AI-served answers needed almost no editorial work.
On top of that, Self-Learning was switched on to handle the long tail: every time a human agent answers a ticket the AI couldn't, Self-Learning drafts a candidate knowledge article from that reply. Approved articles become permanent knowledge, and the next time the same pattern appears, the AI resolves it.
When did Apartment List decide to turn on 'direct replies' to customers?
From go-live (yes, even on day one), but with deliberate scope constraints.
For tickets like data requests, scam reports, and other sensitive outreach, Apartment List used My AskAI’s AI tagging feature to tag, escalate, and block AI replies on those ticket types to keep humans in the loop when it mattered most.
Apartment List also deployed in Zendesk Tickets in "reply only to the first message" mode. In practice: a new ticket comes in, the AI agent reads it and either replies directly to the customer or hands off to a human if the case warrants it. As soon as the customer sends a follow-up message, the ticket is owned by a human for the rest of the conversation.
The AI handles the first touch; humans own every second touch onward.
I've been on enough Zendesk customer calls to know why this is a sensible setup: keeping AI direct replies tight to the first touch builds trust faster than copilot-mode-first deployments, because the customer experience of speaking directly to AI is real from day one, while the support team retains full control of any conversation that needs nuance.
What was the biggest thing Apartment List did to improve their AI agent's resolution?
If I had to pick one move that turned Apartment List from "AI replying" to "AI resolving 76% of tickets", it's rebuilding their existing Zendesk macros as Custom Answers.
Their human agents had been using macros for years, and every high-volume rental marketplace pattern had a macro behind it: application status, payment confirmations, profile edits, listing edits, landlord verification, account closures. The macros worked (we've seen this in a few Zendesk shops): they were brand-on, factually tight, and the team trusted them.
Loading them straight in as Custom Answers meant the AI agent inherited that library wholesale. The AI didn't have to invent answers; it had a curated reply for every pattern an agent had thought worth scripting.
That's the shape AI agents handle best. When the question is "where do I find my application status?", you want the AI to serve the exact answer the team would write. Custom Answers do exactly that: deterministic, fact-checked replies for the patterns where determinism matters.
The compounding layer on top is Self-Learning. Custom Answers cover the patterns Apartment List knew to script; Self-Learning surfaces the patterns they didn't. When a human agent replies to a ticket the AI couldn't resolve, Self-Learning drafts a new knowledge article from that reply and queues it for review.
Together, the two compound: Custom Answers handle the bulk, Self-Learning handles the long tail, and the resolution rate climbs as the library grows. The one-line version: macros got them the floor, Self-Learning is getting them the ceiling.
How do Apartment List customize their AI agent setup to work for their business?
Three customisation moves make the AI feel like Apartment List's, not a generic chatbot:
Controlling their language and communication style
Communication Guidance sets the tone, format, and length of every AI reply.
For Apartment List that means a brand-on voice: warm but professional, matching the voice their support team already uses. Short answers when the question is direct; longer walkthroughs when the renter is troubleshooting; clarifying questions before providing answers.
Setting clear escalation triggers
Handover & Escalation Guidance defines the cases that should always reach a human.
For a rental marketplace, those are the cases where the stakes are highest and the context is messiest:
- Listing fraud reports
- Account closures and data-deletion requests
- Questions from property managers that need to be directed to Client Services
When any of these patterns is detected, the AI hands it off with full conversation context attached, so the human agent picks up the ticket already knowing what the renter has said and what's already been established.
Using AI Tagging to control where AI replies
AI Tagging for Zendesk does two jobs for them.
First, it auto-classifies every incoming ticket into categories: payment, application, profile, listing, fraud, account, other. That classification is what lets the escalation triggers above fire on the right category every time.
Second, the tags are useful by themselves: Apartment List's support leads use them to audit AI-handled tickets at scale, spotting categories where resolution is dipping or sentiment is sliding before it shows up in the headline metrics. (Sentiment auditing was a nice side-effect; they didn't set up AI Tagging for that, but it's where a lot of the operational value lives now.)
What impact is Apartment List's AI customer service agent having now?
Last 30 days, on Zendesk Tickets:
- 76% AI resolution rate
- ~1,582 tickets handled per month (~1,202 resolved by AI)
- ~101 hours saved per month (at 5 min per ticket)
- 97% AI CSAT score
The combination is what makes this work: a mid-70s resolution rate on real customer-facing replies (not copilot-mode notes), and a 97% CSAT on top of it. Apartment List isn't getting volume by trading off quality; they're holding both at once.
Where do Apartment List go from here?
The most obvious next move is our User Data API. Today's setup answers questions from knowledge sources (articles, macros, website pages, the CSV), but the User Data API would let the AI answer questions about the renter's own account state (application status, payment status, listing favourites, saved searches) without needing a human to look it up.
For other Zendesk customers we've worked with, that single step has taken resolution rates from the mid-20s into the high-70s; for Apartment List, it would close the gap on the cases where the AI today says "let me hand you to a teammate who can check that."
Tools and Tasks are the next layer beyond that: agentic multi-step workflows for actions the AI takes on the renter's behalf, such as refund issuance, profile updates, application withdrawal, listing favourites management, and application document re-uploads. Each one removes another reason to involve a human, once trust in direct replies is fully established.
The plan is to keep raising the resolution ceiling without dropping the CSAT floor. And so far, the numbers say they're succeeding.
If you'd like to see more customer stories like Apartment List's, browse all our case studies here. And if you'd like to see what those numbers would look like for your own support team, our Zendesk Tickets ROI calculator and pricing page are the two starting points.
