You’ve built a real operation. Amazon, eBay, Shopify, a support queue that doesn’t stop. The business works. AI customer service has probably crossed your radar, and if you’ve dismissed most of what you’ve seen, fair enough. Most of it is hype written for tech enthusiasts, not for sellers running high-volume marketplace operations.
Here is what actually works, and what doesn’t. Written for the established multichannel seller, not the early adopter.
TL;DR
- AI handles high-volume, predictable queries well: WISMO, return triage, pre-purchase questions.
- AI does not handle complex disputes, apologies, or account-health queries. Those need a human.
- The maths work at meaningful ticket volume with a WISMO-heavy queue. Model it before committing.
What AI genuinely handles well
WISMO queries. “Where is my order?” has a factual answer. AI that connects to live order data reads the tracking status and responds without a human touching it. For sellers with a WISMO-heavy queue, this is where the volume reduction happens. Electrical World (2,000-3,000 tickets/month, multi-marketplace) deflects 80%+ of post-sales queries using eDesk. High volume, predictable query mix.
Drafting responses in seconds. A human agent reads, thinks, types, reviews, sends. AI reads, pulls order data, drafts, and presents in under three seconds. Even in Agent-Assist mode (humans still review before sending), the time saving per ticket adds up across a queue.
24/7 coverage without staffing it. Amazon’s 24-hour response window applies every day. AI doesn’t have a shift. SLA compliance on Amazon becomes a configuration question rather than a staffing one.
Consistent application of channel rules. Amazon Buyer-Seller Messaging guidelines, eBay Resolution Centre requirements. A human handling high volume makes mistakes. AI trained on those rules applies them the same way every time.
What AI genuinely cannot do
Guarantee accuracy without a human review step. AI is a good drafter. It is not a reliable autonomous sender in the first 90 days of any deployment. Sellers who started with Agent-Assist mode, ran it for 30 days, and only then moved to HandsFree on validated query types have better outcomes than those who switched everything to autonomous from day one.
Handle complex disputes. A customer disputing a charge or escalating to an Amazon A-to-Z claim needs human judgement. AI can classify the message and route it to the right person. That is genuinely useful. It cannot make the call.
Send an apology autonomously. An automated apology on the wrong dispute can be cited as an admission in an A-to-Z claim review. This is a hard line: complaint-language messages go to a human, every time, no exceptions.
Cover genuine edge cases. Fraud signals, repeat buyers with history, novel situations outside its training. AI handles the predictable majority. Humans handle the unpredictable tail. Neither replaces the other.
When the maths work and when they don’t
For a seller at 500 tickets a month with half as WISMO or return-status queries, 65% AI resolution costs $322/month (325 tickets at $0.99 each). Compare that to agent time on those tickets. The calculation clears at any reasonable agent rate.
At 50 tickets a month with mostly unique enquiries, the setup and calibration time does not pay off.
The honest version: it works when volume is high enough and the query mix is predictable enough that automation has something to work with. Zendesk’s CX Trends 2025 research shows AI adoption in support scaling fastest in operations with high-volume, structured query types, which is exactly the ecommerce WISMO profile.
For sellers at the right volume, the ecommerce support automation guide covers the setup sequence, and the AI agent support software breakdown covers what to look for in a platform.
Book a Free Demo to run the maths against your actual ticket volume before deciding.
Pricing: Essential $39 / Growth $89 / Professional $119 / Enterprise custom per agent/month. AI resolutions (Ava): $0.99/€0.99 per automated ticket. Pricing verified August 2026.
Key takeaways
- AI handles WISMO, return triage, and pre-purchase queries because the answers are factual and the data is available.
- AI does not handle complex disputes, apologies, or account-health queries. Keep those on a permanent human-only list.
- Start in Agent-Assist mode. Run 30 days with AI drafting and humans approving before enabling HandsFree on anything.
- The maths work at meaningful volume with a WISMO-heavy queue. They don’t work at low volume or with a complex enquiry mix.
- Scepticism is healthy. The sellers who get the most from AI support treat it as a tool to verify before trusting.
Action plan:
- Categorise last 30 days of tickets: WISMO, return triage, pre-purchase, complex disputes. The first three are your automation target.
- Multiply your monthly WISMO and return-triage volume by $0.99 and compare to current agent cost for those tickets.
- If the maths work, start Agent-Assist on WISMO only and review every AI draft for 30 days before expanding.
FAQ
What can AI actually handle in ecommerce customer support?
AI handles high-volume, predictable queries well: WISMO enquiries, return-window triage, pre-purchase product questions, and tracking updates. It works by reading live order data from connected marketplaces and generating accurate, channel-compliant responses. It works best with a supervised period in Agent-Assist mode before sending autonomously, and at ticket volumes where predictable query types make up a significant share of the queue.
What should AI never do in ecommerce customer support?
AI should never send apologies autonomously, respond to Amazon A-to-Z claims or suspension notices without human review, or make judgement calls on complex disputes. These query types carry real account-health risk if mishandled. Identify them upfront and exclude them from AI automation permanently. They are not categories to add to automation later once the system has “learned.” They are exceptions by design.
Book a Free Demo to see how eDesk Ava handles the CAN list and blocks the CANNOT list automatically.
