- How much can AI actually handle?
- Automatable vs. non-automatable tickets
- Resolution rates by ticket type
- Why WISMO decides most of the number
- How your store profile changes the answer
- What's capping your resolution rate
- Answers vs. actions: 40% to 65%
- How to raise your resolution rate
- How to measure it honestly
- Where Bookbag fits
How much of your support can AI actually handle?
A well-configured AI agent handles 40–65% of ecommerce support tickets without any human involvement. Stores with a lot of order-status volume, documented policies, and an agent that can take actions (not just answer questions) land at the top of that band — 65–70%+. Stores with no live order data or judgment-heavy ticket mixes sit lower, around 25–45%.
That range is wide on purpose. "What percentage of tickets can AI handle" is the wrong question if you ask it in the abstract. The right question is what percentage of *your* tickets can AI handle — and that comes down to three things: the mix of ticket types you actually receive, whether the agent is connected to live store data, and how it's configured to act and escalate.
The good news is that none of those three are mysterious. The factors that push the number up or down are predictable, and most of them are within your control. This piece breaks down the realistic ranges by ticket type and store profile, the math that makes WISMO the deciding factor, and the specific levers that move a 40% agent to a 65% one.
One caution before the numbers: be skeptical of vendors quoting 90%+ across the board. That figure usually counts every conversation the bot replied to as "handled," including the ones where it gave a generic answer and the customer gave up or filed a second ticket. A resolution rate that survives scrutiny — measured as tickets actually closed without a human and without a repeat contact — sits in the bands above for almost every ecommerce store. If a quote sounds too clean, ask exactly how it's counted.
Industry-typical AI resolution rate for ecommerce is 40–65% of contacts handled with no human. Order-heavy stores with clear policies and action capability reach 65–70%+. Stores missing order data or running on static FAQs land at 25–45%. The single biggest variable is whether the agent can look up live orders.
Which tickets are automatable — and which aren't
The cleanest way to predict whether a ticket type can be resolved by AI is to ask where the answer comes from. If it comes from data — an order record, a tracking number, a return policy, a product spec — it's automatable. If it comes from human judgment, discretion, or relationship context, it's only partly automatable, and sometimes not at all.
Order-status questions are the textbook automatable case. A customer asking "where's my order" wants two facts: the current shipping status and the estimated delivery date. Both live in your order data. The agent retrieves them and explains them in plain language. There's nothing to decide. Return eligibility within a written policy is the same shape — the rules are deterministic, so the agent applies them.
Now contrast that with a customer who's upset about a product that didn't meet expectations and wants an exception to your 30-day window. The agent can handle the informational half (here's the policy, here's how returns normally work), but the resolution — whether to bend the rule this once — is a judgment call that belongs to a human. These tickets are partially automatable at best, and the honest goal is a clean handoff with full context, not a forced resolution.
| Category | Automatable? | Why |
|---|---|---|
| Order status / WISMO | Highly automatable | Answer comes straight from order data |
| Return eligibility (within policy) | Highly automatable | Policy rules are deterministic |
| Shipping timelines and carrier updates | Highly automatable | Data-grounded and factual |
| Basic product questions (size, materials, fit) | Automatable | Answer lives in the product catalog |
| Return / exchange initiation (action) | Automatable with integrations | Needs action capability, not just answers |
| Simple billing and account questions | Mostly automatable | Account lookup plus written policy |
| Complaints needing empathy and exceptions | Partially automatable | Informational part yes; the call needs judgment |
| Disputes, chargebacks, suspected fraud | Not fully automatable | Needs investigation and human authority |
| Custom, wholesale, or B2B requests | Not automatable | One-off judgment and relationship context |
AI resolution rates by ticket type
Resolution rate varies enormously by ticket type, and the spread inside each type is mostly a configuration story. An agent with live order data, written return rules, and the ability to take actions performs near the top of each range. An agent running on static FAQ pages with no order integration performs near the bottom — and worse, it frustrates customers by failing on the exact questions they ask most.
The table below shows typical and strong resolution rates by ticket type. "Typical" is what an averagely configured store sees in the first month or two. "Strong" is what a store hits once order data is wired in, policies are documented, and the agent can act. The overall blended number is the weighted average across your real ticket mix — which is why two stores running the same software can report 42% and 64%.
Read these as ceilings you work toward, not switches you flip. Order-status resolution sits high because the task is narrow and the answer is unambiguous. Billing sits lower because it spans simple lookups (easy) and genuine disputes (not). Complex complaints barely move on resolution rate by design — the win there isn't autonomous closure, it's the agent gathering details, setting expectations, and routing the ticket to the right person so the human starts halfway done instead of from scratch.
| Ticket type | Typical resolution rate | Strong resolution rate |
|---|---|---|
| Order status / WISMO | 80–95% | 90–97% |
| Return / exchange eligibility | 65–82% | 78–90% |
| Shipping timelines | 75–90% | 85–95% |
| Basic product questions | 60–80% | 75–88% |
| Return initiation (with action) | 55–78% | 70–85% |
| Billing / payment questions | 50–70% | 65–80% |
| Complex complaints | 10–25% (partial) | 20–35% (rest escalated) |
| Overall blended | 40–65% | 60–70%+ |
These ranges reflect general findings across ecommerce support queues. Benchmarks suggest first-contact resolution averages around 70–75% for ecommerce teams overall (human plus AI), with top quartiles reaching 80–85%. Your own numbers depend on catalog complexity, return rate, and how much of your volume is order-status. Treat published figures as a sanity check, not a target handed down from above.
Why WISMO decides most of the number
If you only optimize one thing, optimize WISMO. "Where is my order" is the single largest ticket category for most ecommerce brands — industry studies consistently put it at 25–50% of total volume, clustering around a third. It's also the most automatable category there is, with strong resolution rates of 90%+ once an agent can read live order data.
That combination — huge share, very high ceiling — means WISMO does more to move your blended resolution rate than every other category combined. Work the math and it's obvious why two otherwise similar stores report wildly different numbers.
- WISMO is data-shaped: the answer is a fact in your order record, not a judgment call.
- It's repetitive and high-volume, so automating it frees the most human hours per point of resolution.
- Proactive shipping and delivery notifications cut WISMO contacts before they're ever filed — deflection upstream of the agent.
- An agent that can't read orders caps your whole resolution rate at roughly half of what's achievable, no matter how good its writing is.
Say WISMO is 40% of your tickets. With order data connected, the agent resolves ~92% of it autonomously — that's 37 points of your blended rate from one category. Without order data, the agent resolves ~5% of WISMO (the generic "check your tracking email" replies) — under 2 points. Same agent, same store: a 35-point swing in total resolution rate from one integration.
How your store profile changes the answer
Two stores with identical software and identical configuration can still land 20 points apart, because their ticket mixes differ. A store that ships physical goods with a simple catalog gets a flood of order-status and shipping questions — exactly the automatable kind. A high-consideration store selling complex or high-AOV products gets more pre-sale, fit, and judgment questions, which automate less cleanly.
The table below is a rough map of where different store profiles tend to land on blended resolution rate, once reasonably configured. Use it to set expectations, not as a ceiling — the levers later in this piece move every one of these numbers up.
| Store profile | Typical blended rate | What drives it |
|---|---|---|
| Standard DTC physical goods | 55–70% | High WISMO and shipping share, simple catalog |
| Apparel & footwear | 50–65% | Heavy returns and fit questions; gains from return automation |
| Subscription / replenishment | 55–68% | Account, billing, and skip/swap actions dominate |
| High-AOV / considered purchase | 40–55% | More pre-sale judgment and white-glove expectations |
| Complex catalog (electronics, parts) | 40–58% | Compatibility and spec questions need rich product data |
| Marketplace / B2B mix | 35–50% | More one-off, relationship-driven requests |
If returns and exchanges are a big share of your volume, the difference between an agent that explains the policy and one that actually issues the label and processes the exchange is enormous. For apparel, return automation alone can move blended resolution 10–15 points.
What's actually capping your resolution rate
When a store is resolving fewer tickets than the benchmarks suggest is possible, the cause is almost never "the AI isn't smart enough." It's one of a small, repeatable set of configuration gaps. Fix these in order and the number climbs.
No live order data integration
WISMO is 25–50% of volume. If the agent can't pull real-time order status, it can't resolve any of it — it just tells customers to check their email. This one gap alone caps resolution at roughly 40–50% of what's achievable, and it's the first thing to wire up.
Vague or undocumented policies
An agent can only apply rules it knows. If your return and refund policy is unwritten, inconsistent across pages, or full of "case by case" language, the agent has nothing deterministic to act on, so it escalates every eligibility question. Documenting clear policies is usually the highest-leverage task after order data.
Answers only, no action capability
An agent that can say "yes, you're eligible to return this" but can't generate the label still hands every return to a human. The informational half is done; the resolution isn't. Action capability — initiating returns, issuing small credits, updating order notes — is what separates a 40% agent from a 60%+ one.
Over-cautious escalation thresholds
Some setups escalate at the first hint of uncertainty, which quietly tanks resolution rate. The right calibration is to escalate when the agent genuinely can't resolve the issue — not whenever a ticket looks slightly complex. A well-tuned agent resolves what it can confidently handle and hands off cleanly when it can't.
Answers vs. actions: the gap between 40% and 65%
Here's the distinction that explains most of the spread in this whole topic: answering a question is not the same as resolving a ticket. A chatbot that recites your return policy has answered the question, but the customer still has to go do the thing. They'll be back. That's a partial resolution dressed up as a full one — and it shows up later as a repeat contact.
An agent that takes actions closes the loop in one session. The customer asks to return an item; the agent checks eligibility against your rules, generates the label, emails it, and updates the order — done, no human, no second contact. The same applies to address changes before fulfillment, subscription skips and swaps, applying a small goodwill credit, or cancelling an unshipped order. Each of these converts a ticket that used to require a human into a fully resolved interaction.
This is the practical line between a chatbot and an agent. The chatbot deflects by answering; the agent resolves by acting. When people report resolution rates above 60%, action capability is almost always the reason.
There's a revenue angle here too, which is easy to miss when you're focused on cost. An agent that recommends products, recovers a stalled cart, or answers a pre-sale fit question doesn't just resolve a contact — it can drive a sale. The same conversation that would have been a cost center under a human becomes a small revenue line. That doesn't change your resolution rate, but it changes how you should value the agent that produces it.
- Order tracking and delivery-status lookups, answered from live data.
- Returns, exchanges, and refunds within merchant-set rules and caps.
- Address or item changes on unshipped orders.
- Subscription management — skip, swap, pause, change frequency.
- Small goodwill credits or refunds under a threshold you set.
- Cart recovery and product recommendations that turn support into revenue.
A high "resolution" rate built on answering without acting is a mirage. If the agent tells a customer what to do but doesn't do it, the ticket reopens. Watch repeat-contact rate alongside resolution rate — a real resolution doesn't bounce back in 48 hours.
How to raise your AI resolution rate
If you're below the benchmark, the path up is systematic, not magical: find your biggest escalation categories, fix the root cause, repeat. In rough order of impact, here's what moves the number.
- 1Connect live order data first. It's the single highest-impact change available, because WISMO is the biggest and most automatable category. Without it, nothing else gets you past ~45%.
- 2Document your return and refund policies in plain rules. Replace "case by case" with specific conditions (windows, exclusions, restocking fees) the agent can apply deterministically.
- 3Turn on return and exchange initiation as an action. A customer who gets a label in the chat has a fully resolved ticket; one told to "contact us" does not. See how to automate returns and exchanges for the mechanics.
- 4Review your escalation queue weekly. Pull the top three to five categories being handed off and ask, for each: could better configuration have resolved this? Usually one or two could.
- 5Authorize small-amount credits and refunds. For disputes under a cap you're comfortable with (say $15), letting the agent resolve on the spot removes an entire escalation category.
- 6Tune the confidence threshold. If the agent escalates "just in case" on questions it should own, tighten its scope and let it respond on established patterns instead of bouncing them to a human.
How to measure resolution rate honestly
Resolution rate is easy to game and easy to misread, so measure it carefully. The number you want is the share of contacts the agent fully handled with no human touch — not the share where it sent any reply, and not the share where it simply closed the conversation. Those last two inflate the figure without reflecting reality.
Pair it with two guardrail metrics so a high number stays meaningful. Repeat-contact rate tells you whether "resolved" tickets are actually staying resolved. CSAT tells you whether customers were happy with how they were resolved. A resolution rate that climbs while repeat contacts and CSAT hold steady is real improvement; a resolution rate that climbs while repeat contacts rise is just deferred work.
Segment the number, too. A single blended figure hides where the agent is strong and where it's leaking. Break resolution rate down by ticket type and by channel — website chat, email, WhatsApp, Instagram DM — and the picture sharpens fast. You might find WISMO resolving at 93% while return initiation drags at 48% because a label step isn't wired up. The aggregate told you "64%"; the segments tell you what to fix on Monday.
- Resolution rate: contacts fully handled by the agent with zero human involvement.
- Repeat-contact rate: how many "resolved" customers come back within a few days about the same issue.
- CSAT on AI-handled tickets: were customers satisfied, not just answered?
- Escalation reason breakdown: the categories you hand off most are your roadmap for what to fix next.
The percentage of inbound support contacts that an AI agent resolves end-to-end with no human involvement, where "resolved" means the customer's issue was actually handled — not merely answered or auto-closed. It's the headline efficiency metric, but it only means something when measured alongside CSAT and repeat-contact rate.
Where Bookbag fits
Bookbag is an AI support agent built for Shopify and ecommerce, and the reason that matters here is everything above: the number you hit depends on order data and action capability, and Bookbag is built around both. It connects natively to Shopify, WooCommerce, and BigCommerce, so WISMO and order-status questions are resolved from live data on day one rather than deflected to a tracking email.
It takes real actions inside your rules — order tracking, returns, exchanges, refunds within your caps, subscription changes, address edits, product recommendations — which is what moves resolution rate from the answers-only band into the 60–70% range. When a ticket genuinely needs a person, it hands off to your help desk with the full conversation and order context attached, so the customer doesn't repeat themselves. Bookbag deflects up to around 70% of tickets autonomously for well-suited stores, and most merchants are live in under a day.
Pricing is flat and predictable — monthly plans with message-credit allowances and a spend cap you set, with no per-resolution fee. That matters for this topic specifically: tools that charge per resolution penalize you exactly when the agent does its job well. Bookbag doesn't.
Key takeaways
- A well-configured AI agent handles 40–65% of ecommerce tickets with no human; order-heavy stores with action capability reach 65–70%+.
- WISMO is the deciding category — 25–50% of volume and 90%+ automatable, so connecting live order data is the single biggest lever.
- The number depends on your ticket mix and store profile, not just the software; identical setups can land 20 points apart.
- Action capability (issuing returns, credits, subscription changes) is what separates a 40% agent from a 60%+ one.
- Judgment-heavy tickets — disputes, fraud, custom exceptions — are partially automatable at best; the right outcome is a clean, context-rich handoff.
- Measure resolution rate alongside repeat-contact rate and CSAT, or a high number just hides deferred work.