Use cases

WISMODamaged & missing itemsSubscriptionsProduct adviceRefunds & returns
Case studies Integrations Sign in
Guides 10 min read

AI resolution rate: what is realistic for an online store in 2026

AI resolution rate for online stores: what share of tickets an AI agent resolves end to end, why published figures differ and how to measure it honestly.

Bryan Delmee
Written by Marketing, Engaige
Share

AI resolution rate is the share of support tickets an AI agent closes alone, start to finish. For online stores, published figures run from about 5% of all tickets to a median of 45% of the tickets the AI touches (top quartile above 65%), and 60% of all tickets at one named retailer. Most of the gap is what each figure is divided by.

Which automation or resolution rate is realistic, and how it is measured, is one of the questions buyers ask us most. This page answers it and sits in our guide to AI agents in customer service.

What is AI resolution rate?

AI resolution rate is a fraction: the tickets an AI agent closed alone, over either the tickets it handled or every ticket you received. A ticket belongs in the top of that fraction only if it passes three checks: what the customer asked for was actually done in your systems, no colleague replied or acted, and the customer did not write again about it inside a window you set beforehand.

Both bases are legitimate. Divided by the tickets the AI handles, the rate shows how good the agent is at what it takes on. Divided by all incoming tickets, it shows how much work left your team.

In the strict definition used here, three outcomes are not a resolution: an AI reply (answered), a conversation the customer left (contained) and a ticket a person finished (assisted).

Deflection is a different measure again. Deflection counts contacts that never reached a person; resolution counts requests that were actually completed. Our metrics guide covers deflection rate in detail.

What AI resolution rate is realistic for an online store?

On the tickets an AI agent handles, about half is the published median. As a share of all tickets, it depends on how many ticket types the agent covers. Among our customers, 60% of all tickets is a realistic median, where “resolved” means the agent takes the action, replies and closes the ticket with no person involved. Without that definition a percentage says little. For comparison: across 131 shops in one study it was 4.9%.

Published figures and what they divide by

SourceFigureDivided byWhat counts as resolved
Chatarmin, 20264.9%All resolved tickets, 131 shopsAI alone, no person and no workflow
Chatarmin, same study20.3%Tickets the AI touchedSame
Gorgias, 2026Median 45%, top quartile above 65%Tickets the AI touchedClosed without a human agent message
Otrium, an Engaige customer60%All 120,000 annual ticketsResolved end to end

Chatarmin and Gorgias report on their own customers; Chatarmin states that its sample “is not representative of e-commerce as a whole”.

On 16 November 2026 we put this exact question to four AI assistants, in English and in Dutch. The eight answers ranged from 5 to 20% of all tickets at the start to 70 to 84% of total volume.

Our view: build your plan on the all-ticket figure, and read any 80% as a share of something until the base is named. That high on all incoming tickets is not typical in the platform data above. The highest all-ticket figure in the table is Otrium, where 60% of the 120,000 tickets it receives in a year are resolved end to end with Engaige.

With Engaige, we are able to automatically resolve 60% of the 120,000 support tickets that we receive annually.

Tessa van der Lof Tessa van der Lof Head of Operations, Otrium · full case study

What the rate is worth in money is covered in our post on support ROI.

Why do published AI resolution rates differ so much?

Published rates differ for three reasons: the base (all tickets or only those the AI touched), the definition of resolved, and what else is counted in, such as workflows, assisted tickets and contained conversations.

The base. One identity explains most of the gap: share of all tickets resolved = share of tickets the AI handles × resolution rate on those tickets. In Chatarmin’s data, AI touched 24.2% of tickets and resolved 20.3% of those, which is 4.9% of all tickets. Gorgias’s metric documentation gives two rates for one agent: one over the total workload, one over the tickets it handled.

The definition. Three help desk vendors each define a resolution in their own documentation.

Vendor documentationWhen a conversation counts as resolved
Intercom, Fin AI Agent outcomesThe customer confirms the answer helped, or leaves without asking for more help. Removed again if the customer returns to the conversation.
Zendesk, automated resolution tiersThree tiers: assisted escalation, contained resolution, and verified resolution after an LLM check. Email is evaluated 72 hours after the first email, messaging by default 2 hours after the first message.
Gorgias, metric documentation (linked above)72 hours have passed without a hand-over to a person.

None of these is wrong; some are broader than the strict definition above, so the percentages cannot be compared.

What else is counted in. In the Chatarmin data set, total automation including no-code workflows was 32.9% of resolved tickets, where AI alone resolved 4.9%.

How do you calculate AI resolution rate honestly?

Count a ticket as resolved by AI only when the request was completed, no person replied or acted, and the customer did not return about it within a window you fix in advance. Then report the rate over both bases, next to coverage.

  • Resolution on handled tickets: AI-resolved tickets ÷ AI-handled tickets × 100.
  • Resolution on all tickets: AI-resolved tickets ÷ all tickets received × 100.
  • Coverage: AI-handled tickets ÷ all tickets received × 100.

Four rules keep the count honest:

  1. Fix the re-contact window and state it. Vendor windows run from 2 hours to 72 hours.
  2. Count assisted tickets separately. A hand-over is never a resolution.
  3. Check the action in the back end, not only in the reply.
  4. Read a weekly sample of AI-closed tickets against the rubric you use for your team.

To isolate the AI’s contribution when several changes run at once, measure per ticket type, before and after that type goes autonomous.

A high rate can give a false sense of success when customers drop off or accept a wrong answer. So pair the rate with re-contact and CSAT on AI-closed tickets. Zendesk warns of the same in its own guide (Zendesk, 2026). In Gorgias’s data, that CSAT is roughly level with human-resolved tickets of the same type.

Which tickets can an AI agent resolve end to end, and which not?

An AI agent can resolve the tickets where the answer or action follows from your data and a written policy, such as order status, return requests, cancellations before dispatch and policy questions. Tickets that turn on judgement, an exception or goodwill still need a person, and that share is yours to set.

The last 10 to 30% of tickets is often very case-specific or needs a human touch. Those are the edge cases, and it takes time before you can automate them. Gorgias arrives at a similar estimate for most e-commerce brands: 20 to 30% of volume truly needs a person.

An answer is not a resolution. For returns, quoting the policy is an answer; registering the return is a resolution. For order status, a tracking link is an answer; reading the carrier status and acting under your late-parcel policy is a resolution.

If your volume is mostly complaints, warranty or custom work, expect a lower all-ticket rate and manage the rate per ticket type. Where a person approves one step, count the ticket as assisted.

What makes an AI agent resolve more tickets?

Three things you control: what the agent can read and do in your systems, how much your written policies let it do alone, and how it improves after go-live. One you do not: your ticket mix, which sets the ceiling.

Gorgias’s research on its own customers points the same way: the brands at the top differ in intent coverage, authority to act, system integrations and a narrow escalation policy.

Depth of access. An agent that only reads help articles can answer. One connected to your store, carriers and returns tools, such as PostNL, DHL or Returnless, can act. On Shopify, Engaige combines live Admin API data with your storefront content. Before it answers, it has time to read the order, check your policy and act.

Customer · email

Hi, where is my order? It's been eight days and the tracking hasn't moved.

Shopify logo

Shopify

Order 48219 · fulfilled

Policy Where's my order Action Lookup order
Picqer logo

Picqer

Still processing Contradicts Shopify

Action Get shipment
DHL logo

DHL

Returned to depot · 3 delivery attempts

Settles it: the parcel is not with the customer

Knowledge Delivery times per carrier Action Get tracking

Reasoning…

Not deliverable. Reship to a confirmed address, booked in Picqer.

Knowledge Depot returns excluded Action Reship order

Suggested reply

Your parcel went back to the depot after three delivery attempts, so it will not arrive. I have booked a reshipment to the address you confirmed, arriving Thursday.

Every answer shows its working: which systems Engaige checked and why it replied as it did. Example data.

Policies and authority. A question with no written policy ends in a hand-over. With Engaige you decide per step whether the agent acts on its own or drafts for your team, with thresholds. Our guide to agent guardrails goes deeper.

Improvement after go-live. In Agent Assist, every reply your team sends is compared with what the agent suggested; that is one of the signals Engaige AI learns from. You approve the changes it proposes.

How fast does the rate climb after go-live?

If you start in Agent Assist, the rate starts at zero resolved with a growing share assisted, then rises one ticket type at a time. Each type you switch to autonomous widens the share of tickets the agent handles, which is what lifts the all-ticket rate.

With Engaige you go live in Agent Assist in hours. Nothing closes without a person there, so those tickets are assisted. You then switch ticket types to autonomous one at a time, as the agent matches your team on each. Brands that move quickly reach autonomy in about a week, and launching autonomous from day one is also supported. The rate keeps climbing after that, as Engaige AI keeps learning from hand-offs and emerging trends.

The roll-out itself is in our implementation steps. What the rate means for the size of your team is in our team comparison. The case for a lean team is in the guide for small teams.

How do you check a vendor’s AI resolution rate before you commit?

Ask what the percentage is divided by, what counts as resolved and who or what decides it. Then see it on your own tickets, not on demo data. Put these questions to any vendor, including us:

  • Which tickets is the rate divided by?
  • What counts as resolved, and within which window?
  • Are workflows, assisted and contained tickets in the number?
  • What is the rate per ticket type on a mix like ours?
  • Which named customer’s result can we open?

Applied to our own claims: Engaige automates up to 90% of supported tickets, so the base is the ticket types the agent supports. Otrium’s 60% is a share of all the tickets it receives in a year.

You can test every Engaige policy in the Playground before it goes live, including real past tickets you paste in, or run Engaige on chosen ticket types next to your current AI and compare both.

Where does Engaige fit?

Engaige (that’s us) is the e-commerce-native AI support agent that plugs into the help desk you already run, resolves across your stack and keeps improving itself. Keep your helpdesk. Replace the AI. Behind it is Engaige AI, which you brief in plain language and which learns from your conversations, your inputs and data analysis.

Hello Potential New Client

What would you like to get done today?

Could you analyse this week's conversations and show me some automation opportunities?

Ask Engaige…

Launch my agent Create a policy Investigate a conversation Setup integration

Could you analyse this week's conversations and show me some automation opportunities?

Searched and found 4 contact reasons · 2.4s

Your agent resolved 426 of 512 conversations. 86 were handed over, and one contact reason accounts for most of them.

512 Total conversations 86 Handed over 83% Resolution rate

What were the reasons for handover, and how can we improve?

Searched and found 1 policy · 1.6s

Delivery status hands over whenever tracking is older than 24 hours, which is 40 of those 46.

Fix it.

Editing Delivery status · 1.9s

Done. It now handles stale tracking itself. Those 40 a week stay with your agent.

What were the reasons for handover, and how can we improve?

Fix it.

Reply…

Engaige AI shows what to change in your policies and why; you approve every change. Illustrative figures.

You can open any conversation to see the full reasoning behind every answer and action. The caveat: Engaige is specialised in e-commerce, not a generic horizontal tool, and text-first by design. If phone is a big share of your mix, a voice specialist is the honest answer.

To see what the rate would look like on your own ticket mix, book a demo.

Frequently asked questions

What is a good AI resolution rate for e-commerce?

About half of the tickets the AI handles is the published median. As a share of all tickets it depends on coverage, from 4.9% across 131 shops in one study to 60% at the one named retailer in the table above.

Is an 80% AI resolution rate realistic for an online store?

Usually it is a share of the tickets the AI touched or supports, not of all incoming tickets. On all tickets it is not typical in published platform data, so ask for the base.

Who decides whether a ticket counts as resolved by AI?

The vendor’s definition does, unless you set your own. Vendors use a customer signal, an automated check or time without a hand-over. Set one rule and apply the same rule to the tickets your team closes.

What is the difference between AI resolution rate and automation rate?

Automation rate often includes no-code workflows, and sometimes assisted or contained tickets. AI resolution rate should count only the tickets the AI finished alone. In the Chatarmin data the two are 32.9% and 4.9% of resolved tickets.

Does a higher AI resolution rate hurt customer satisfaction?

Not when the ticket is truly resolved: in Gorgias’s data, CSAT on those tickets is roughly level with human-resolved tickets of the same type. A rate can also hide customers who gave up, so track re-contact and CSAT on AI-closed tickets beside it.

Does email reach the same AI resolution rate as chat?

Gorgias’s research names contact form and email as the two channels with the cleanest path to end-to-end resolution. Engaige is text-first by design: email, chat, WhatsApp, SMS and social DMs.

How long does it take to reach a high AI resolution rate?

It depends on how fast you switch ticket types to autonomous. With Engaige, brands that move quickly reach autonomy in about a week, and the rate keeps climbing after that.

Can I test an AI agent on my own tickets before I switch?

Yes. With Engaige you paste real past tickets into the Playground, or run it on chosen ticket types next to your current AI.

Keep reading

Built for the future.
Available today.

Get started or book a demo to experience your support agent from the future.