Zendesk AI Resolution Rate Report: Track and Optimize Bot Performance
AI agents handle a growing share of support conversations. But “handled” doesn’t mean “resolved.” If your bot is closing conversations that customers later reopen through other channels, your resolution numbers are inflated and your team is doing invisible rework. A proper AI resolution rate report separates genuine automated resolutions from abandoned conversations — and gives you the data to decide where bots help and where they hurt.
What AI resolution rate measures
Bot resolution rate is the percentage of customer conversations fully resolved by your AI agent without human intervention. Zendesk uses LLM verification to confirm that the customer’s issue was actually addressed — not just that the conversation ended.
The formula:
AI Resolution Rate = Conversations Resolved by AI ÷ Total Conversations Handled by AI × 100
A conversation counts as “resolved” when the customer’s issue is confirmed solved within an evaluation window: 2 hours for messaging channels (configurable up to 72 hours) and 72 hours for email. Conversations that are escalated, reopened, or result in the customer contacting support again through another channel are excluded.
This matters because the gap between “bot responded” and “bot resolved” is often 30–50 percentage points.
Where AI resolution rate fits in your metrics stack
AI resolution rate doesn’t exist in a vacuum. It connects to other operational metrics that tell the full story:
- Ticket deflection — Measures total volume diverted from agents. AI resolution is the quality check on that deflection.
- First contact resolution — For conversations that reach agents, FCR tells you whether the handoff created a worse experience.
- CSAT — The ultimate check. High resolution rate with falling CSAT means the bot is closing conversations, not solving problems.
- Escalation rate — The inverse signal. If escalation rate is rising alongside AI resolution rate, the bot may be creating more complex problems downstream.
- Cost per ticket — AI resolutions typically cost $1–$2 vs $15–$30 for agent-handled tickets, making the ROI clear when resolution quality holds.
How to report AI resolution rate in Zendesk Explore
Zendesk’s AI agent reporting dashboard provides hourly-updated metrics. Here’s how to build a useful report:
Step 1: Open the AI agent dashboard
Navigate to Explore → Dashboards → Zendesk AI agents. This prebuilt dashboard shows automated resolution volume, escalation rate, and conversation outcomes.
Step 2: Build a custom resolution rate report
For deeper analysis, create a custom report:
- Go to Explore → Reports → New Report
- Select the Support: AI agents dataset
- Add metrics:
- Automated resolutions (count)
- Total AI-handled conversations (count)
- Escalation rate (percentage)
- Add dimensions for the views you need:
- Time — daily or weekly trending
- Topic or intent — which categories the bot handles best
- Channel — messaging vs email vs web widget
- Add a calculated metric:
D_COUNT(Automated resolutions) / D_COUNT(Total AI conversations) * 100
Step 3: Add context metrics
Don’t look at resolution rate alone. Add to the same dashboard:
- BSAT (bot satisfaction) — Customer rating of the bot interaction
- Understood rate — Percentage of conversations where the bot found a relevant answer
- Average conversation turns — More turns often means the bot is struggling
Step 4: Filter out false positives
Exclude conversations where: - The customer abandoned after the bot’s first message (no confirmation of resolution) - The customer contacted support again within 24 hours through a different channel - The conversation was auto-closed by a timeout rule
These filters ensure your resolution rate reflects genuine solutions, not conversation abandonment.
Benchmarks: what good looks like
AI resolution rates vary significantly by deployment maturity and use case:
| Stage | Typical rate | What it means |
|---|---|---|
| New deployment (0–3 months) | 10–20% | Bot handles basic FAQs; most conversations escalate |
| Optimized (3–6 months) | 25–40% | Bot covers top ticket categories with trained intents |
| Mature (6–12 months) | 40–60% | Broad coverage with custom actions and integrations |
| Best-in-class | 60–80% | Deep product integration; bot executes actions, not just answers |
By conversation type:
| Type | Expected resolution rate |
|---|---|
| FAQ / how-to questions | 70–90% |
| Account/billing inquiries | 40–60% (with integrations) |
| Technical troubleshooting | 15–30% |
| Complaints / escalations | 5–10% |
If your rate is below your deployment stage benchmark, look at the knowledge base coverage and intent training first — they’re the most common gaps.
Common mistakes
Counting all bot interactions as resolutions. If a customer asks a question, the bot gives an answer, and the customer leaves without confirming, that’s not a resolution. It might be an abandoned conversation. Rely on Zendesk’s LLM verification rather than simple conversation-end triggers.
Ignoring channel bleed. A customer might get a bot response on chat, leave, and then email support about the same issue. Your chat AI resolution rate looks great; your actual resolution rate is lower. Track multi-channel re-contacts to catch this.
Optimizing resolution rate without watching CSAT. You can increase AI resolution rate by making it harder to reach a human agent. But that’s a terrible customer experience. Always pair resolution rate with CSAT and bot satisfaction scores.
Not segmenting by topic. Your overall AI resolution rate is meaningless if the bot resolves 90% of password resets but 5% of billing disputes. Segment by intent or topic to understand where the bot adds value and where it doesn’t.
Setting unrealistic targets. A 100% AI resolution rate is not the goal. Complex, emotional, or ambiguous issues need humans. The target is to automate the right conversations — the ones that are repetitive, well-documented, and low-complexity.
What to do when AI resolution rate drops
A sudden drop usually means one of three things:
- Knowledge base changes — Did someone update or remove articles the bot relies on? Check recent changes to your help center content.
- New ticket types — A product release or incident creates conversations the bot hasn’t been trained for. Check the “not understood” rate and review unmatched intents.
- Threshold or config changes — Did someone adjust the resolution confirmation window or escalation rules? Check your AI agent configuration.
For gradual declines, look at: - Topic drift — Are customers asking different questions than they were 3 months ago? The bot’s training data may be stale. - Customer expectations — As customers learn what the bot can do, they may start asking more complex questions. This naturally lowers the resolution rate for conversations that reach the bot.
Building an AI resolution rate dashboard
A useful AI performance dashboard includes four sections:
Volume: Total AI-handled conversations, automated resolutions, and escalations — trended weekly.
Quality: AI resolution rate, BSAT score, understood rate, and customer re-contact rate after AI resolution — trended weekly.
Efficiency: Average conversation turns for resolved vs escalated conversations, time to escalation, and cost per AI resolution vs cost per agent resolution.
Coverage: Resolution rate by topic/intent, showing which areas are automated effectively and which need human support.
Add this as a tab on your support metrics dashboard so the team sees AI performance alongside human performance metrics.
FAQ
How is AI resolution rate different from ticket deflection? Ticket deflection counts all tickets diverted from agents — including self-service, chatbot interactions, and auto-responses. AI resolution rate specifically measures conversations where the AI agent confirmed it solved the customer’s problem. Deflection is a volume metric; resolution rate is a quality metric.
Should I compare AI resolution rate to human first contact resolution? Yes, but carefully. AI resolution rate and first contact resolution measure similar outcomes through different lenses. A mature AI agent should have a higher resolution rate than your human FCR for the specific topics it handles, because it’s handling the simpler, more repetitive conversations. If it doesn’t, the bot isn’t adding enough value.
What’s the cost impact of improving AI resolution rate by 10 percentage points? Rough math: if your team handles 5,000 conversations/month and your AI resolution rate goes from 30% to 40%, that’s 500 additional automated resolutions. At a savings of ~$15 per resolution (the difference between AI and agent cost), that’s $7,500/month. See cost per ticket for the full calculation.
How does Zendesk verify that an AI resolution is legitimate? Zendesk uses an LLM-based check that evaluates the conversation context to determine whether the customer’s question was actually answered. This is more reliable than simple “no further messages” detection, though it’s not perfect — edge cases still exist, especially for complex multi-part questions.
See your AI agent performance alongside your team metrics — start free