Migrating from Zendesk Chat: The Step-by-Step Guide

Why teams are leaving Zendesk Chat now

Zendesk Chat traces back to the 2014 Zopim acquisition, and Zendesk has been winding the legacy product down for years, pushing accounts toward Suite messaging and the Agent Workspace. That transition is not free: as of mid-2026, Zendesk Suite plans run roughly $55-$115 per agent per month on annual billing (check current pricing), so a five-agent team is looking at approximately $3,300-$6,900 per year just to keep a chat channel they already had.

The bigger issue is architectural. Legacy Chat is built around rule-based triggers and canned shortcuts, a model that typically deflects only 10-20% of conversations, because a trigger can start a conversation but cannot answer one. A modern website AI chatbot trained on your help center routinely resolves 50-70% of inbound questions with no human involved. If a forced migration is coming either way, it is worth comparing the whole field. Our roundup of the best AI chatbot platforms in 2026 is a good starting point, rather than defaulting to the in-family upgrade path.

Step 1: Export everything before you cancel

Access disappears when the subscription does, so archiving comes first. Budget 2-3 hours and capture all of the following:

• Chat transcripts. Smaller accounts: request a history export from the dashboard (emailed CSV). High-volume accounts: use the Chat API incremental export endpoints and archive the JSON.
• Shortcuts. Export your canned replies. They are a goldmine for seeding AI answer content.
• Trigger definitions. There is no clean export; copy each trigger's conditions and message text into a spreadsheet.
• Departments and agent rosters. You will remap these to routing rules.
• Banned visitor list and chat ratings history. Small files, easy to forget, annoying to lose.

Step 2: Baseline your KPIs so you can prove the switch worked

Before anything changes, pull 30 days of reporting: weekly chat volume, median first response time (human-staffed chat typically lands at 1-3 minutes), missed-chat percentage, resolution rate and CSAT. This baseline is what turns the migration from "we switched tools" into "we cut first response time by 95% and deflected half our volume." If you have never formally measured these, our guides to first response time and deflection rate define each metric and give industry benchmarks to compare against.

Step 3: Remap triggers and shortcuts to AI-first flows

Almost everything you built in Zendesk Chat has a direct, usually better, equivalent. The mental shift: triggers and shortcuts were workarounds for a widget that could not think. In an AI-first setup, the proactive rule still exists, but what it opens into is a bot that actually answers.

Step 4: Retrain your answers on the help center

This is where the migration pays for itself. Point EzyConn at your help center and website URL and it crawls, chunks and indexes the content in about 15-30 minutes. The full walkthrough is in our guide to training an AI chatbot on your website. A typical SMB help center of 50-200 articles gives the AI enough coverage to answer the majority of routine questions on day one.

Before you flip anything live, run a gap check: take the top 20 question intents from your exported transcripts and ask the trained bot each one. In most migrations, 5-10 intents surface with weak or missing answers, usually pricing edge cases, cancellation policy and shipping exceptions. Write or fix those articles first; each closed gap typically adds 2-4 percentage points of deflection.

Step 5: Rebuild routing and human handoff

Recreate departments as routing rules (by topic, page or language), set staffed hours, and define when the AI escalates: an explicit "talk to a human" request, negative sentiment, high-value keywords like "refund" or "cancel," or two consecutive failed answers. Make sure the full conversation transcript and captured context travel with the escalation. Agents re-asking questions the bot already asked is the fastest way to burn the goodwill the AI earned. Our human handoff guide covers escalation design in depth, including how to keep handoff rate in the healthy 20-35% band while quality stabilizes.

Step 6: Parallel run, cutover and the first 30 days

Never hard-cut a live chat channel. Run the new widget on a subset of pages (or a percentage of traffic) for 1-2 weeks while legacy Chat keeps serving the rest, and compare weekly numbers against your baseline. When the new channel matches or beats CSAT and beats FRT, which it will, usually by week two, remove the old snippet everywhere and redirect any offline-form links. The complete timeline:

For the first 30 days post-cutover, review AI conversations weekly and patch knowledge gaps as they appear. Reasonable targets: AI first response under 5 seconds, deflection at 40-50% by day 30 and climbing toward 60%+ by day 60, CSAT at or above your Zendesk Chat baseline. Pricing-wise, you are also trading per-agent seat math for platform pricing. See our pricing page for what that looks like at your volume.

Frequently Asked Questions

Last updated July 2, 2026. Export options and Zendesk pricing reflect publicly available information as of mid-2026. Verify specifics for your account with Zendesk's documentation. View more guides.