AI Ticket Resolution Rate: What Counts as Resolved
Your AI ticket resolution rate is the share of tickets the AI closes completely on its own. A ticket only counts if your team never touched it, the customer got what they asked for, and nobody reopened it within a set window. Most published rates skip at least one of those three conditions. That’s why the numbers vary so much from one tool to the next.
This guide covers the formula, the counting rules that change the result, what the sourced data says, and a way to test a resolution rate on your own Shopify tickets. If you’re comparing tools right now, look at what each one can actually do on your orders. See how Relay resolves Shopify tickets end to end, from refunds to address changes, within the permissions you set.
How the AI resolution rate is calculated
The basic formula:
AI resolution rate = tickets fully resolved by AI ÷ total tickets in the same period × 100
The formula is simple. The two numbers inside it are not.
The denominator. Is it every ticket that came in, or only the tickets the AI picked up? Take a store with 1,000 tickets in a month. The AI handles 600 of them and fully resolves 300. Measured against tickets handled, the rate is 50%. Measured against all tickets, it’s 30%. Same month, same AI, same work. The tool just reports a different denominator.
The numerator. What counts as “fully resolved”? A ticket marked closed is not the same as a ticket that’s solved. If the customer gave up, or came back the next day through another email, the AI didn’t resolve anything.
Before you trust any rate, ask for both definitions in writing.
What counts as resolved: a 4-point test
A ticket counts as resolved by AI only if all four points are true:
- No human touched it. No agent replied, edited a draft or took an action on the order.
- The request was fulfilled. The customer got the tracking link, the refund, the new address or the answer. Saying “I’ve passed this on” isn’t a resolution.
- No reopen within the window. A 72-hour window is a common choice: if the customer writes again about the same issue within 72 hours, the ticket is no longer resolved.
- No repeat contact through another channel. The same customer writing again about the same order counts as a reopen, even if it lands as a new ticket.
The fourth point is the one most dashboards miss. In a June 2026 field experiment on 680,676 chats, Alibaba found that 43% of chats led to a new attempt within seven days. A ticket that looks closed may simply have moved somewhere else.
Resolution vs deflection, containment and automation rate
These terms get mixed together in sales pages. Here is what each one measures.
| Metric | What it counts | What it misses |
|---|---|---|
| Resolution rate | Tickets fully solved by AI, with no reopen | Nothing, if the 4-point test is applied |
| Deflection rate | Customers who found an answer before opening a ticket | Whether the answer actually solved the problem |
| Containment rate | Conversations that never reached a human | Customers who gave up or left frustrated |
| Automation rate | Tickets where AI did at least part of the work | Tickets where a human still had to finish |
| First contact resolution | Tickets solved in one exchange, by AI or a human | Who did the work |
Containment and automation can never come out lower than resolution, because they count tickets the 4-point test throws out. When a tool shows you a single high number, check which of these five it really is.
What is a good AI resolution rate?
There’s no shared benchmark, because every vendor counts differently. Here is what independent, named sources say.
- Consumers see less than dashboards show. In a March 2026 survey of 2,000 consumers by NewtonX, only 24% said their last AI support interaction was fully resolved by the AI alone.
- Closed doesn’t mean solved. Alibaba’s June 2026 experiment on 680,676 chats found that 43% led to a new attempt within seven days.
- Self-service has been low. Gartner found in August 2024 that only 14% of customer service issues were fully resolved through self-service.
- Forecasts are ambitions, not results. Gartner (March 2025) expects agentic AI to resolve 80% of common customer service issues on its own by 2029.
The gap between what tools report and what customers say tells you something. Apply the 4-point test, and treat any rate measured on your own tickets as more reliable than a benchmark.
Which Shopify tickets AI can resolve on its own
Benchmark pages don’t break the number down for an online store. For a Shopify merchant, the resolution rate depends far more on your ticket mix and on what the AI is allowed to do than on the model behind it.
| Ticket type | Can AI fully resolve it? | What it needs |
|---|---|---|
| Where is my order | Yes | Live order and tracking data from Shopify |
| Address change before shipping | Yes | Permission to edit the order, plus a rule on fulfillment status |
| Order cancellation | Yes, before fulfillment | Permission to cancel and a clear cutoff rule |
| Return or exchange request | Often | Your return policy, eligibility checks per product |
| Refund | Often, within limits | Permission to refund and a cap per order |
| Forgotten discount code | Yes | Permission to issue codes within a limit |
| Product questions | Yes, if the answer is documented | A knowledge base that covers sizing, materials, care |
| Damaged item, chargeback, angry customer | Partly | A human for the decision, AI for collecting facts and photos |
The pattern is simple. An AI that can only reply caps your rate at questions with a written answer. An AI that can read the order and act on it (cancel, refund, edit the address) can close the order tickets that fill a store’s inbox. Order tracking is a good place to start: our guide on automating where is my order emails on Shopify shows the triggers.
How to measure your own AI resolution rate
Run this on four weeks of tickets, before and during any trial:
- Fix the denominator. Count every ticket received, not just the ones the AI picked up.
- Apply the 4-point test to every ticket the AI closed. Use a 72-hour reopen window.
- Match repeat contacts by customer email and order number, not by ticket ID.
- Split the rate by ticket type. A rate of 90% on order tracking and 10% on returns says more than one blended number.
- Read CSAT on resolved tickets only, with the same counting rules for every type. Our guide to CSAT counting rules by ticket type explains how.
- Sample false resolutions. Each week, read a handful of tickets the AI marked as resolved. Count the ones where the answer was wrong or incomplete.
Relay shows resolution rate and time saved in its stats, so you can run this test on your own volume during the trial instead of relying on someone else’s benchmark.
Why your AI resolution rate is low, and how to raise it
If your rate is low, the cause is usually one of these:
- Gaps in the knowledge base. List the questions the AI escalated last week. Each repeated one is a missing article.
- No real actions. If the AI can’t refund, cancel or edit an order, every one of those tickets goes to a human. Give it permissions with limits.
- No confidence threshold. An AI that guesses creates false resolutions and reopens. Set it to escalate when it’s unsure.
- No correction loop. When your team fixes an AI answer, that fix should update the rules or the knowledge base.
Raising the rate shouldn’t mean holding on to tickets that need a person. Some tickets should always go to your team: high value orders, angry customers, safety issues, anything outside your rules. When a customer asks for a person, use a clear handoff. Our guide on replying when a customer asks for a human agent has ready replies. A clean escalation protects your rate, because it stops a false resolution before it turns into a reopen.
FAQ
What is a good AI resolution rate?
There’s no shared benchmark. Gartner expects agentic AI to resolve 80% of common issues by 2029, but that’s a forecast. Today, NewtonX found only 24% of consumers said their last AI interaction was fully resolved. Measure your own rate with a fixed denominator and a reopen window before comparing.
What is the difference between deflection and resolution?
Deflection counts customers who didn’t open a ticket after seeing an answer. Resolution counts tickets the AI fully solved with no reopen. Deflection doesn’t tell you whether the problem was actually solved.
How do you measure false resolutions?
Track reopens and repeat contacts on the same order within a fixed window, such as 72 hours. Then read a weekly sample of AI-resolved tickets and count the wrong or incomplete answers.
Does the AI resolution rate drop over time?
It can. New products, new policies and new ticket types arrive, and an AI that isn’t updated starts escalating or guessing more. A weekly correction loop and an up-to-date knowledge base help keep it stable.