What ecommerce customer support tasks can be automated?
Support automation modes describe how much of a resolution happens without a person. In drafted mode the software investigates and writes the reply, and an agent sends it. In autonomous mode the software resolves and responds without an agent opening the ticket. In autonomous-with-escalation, it begins autonomously and hands over when it recognises the case is beyond its remit or sensitive.
Why this happens
Most writing on this subject presents a fixed list of automatable tasks, which is a description of a particular product rather than of the problem. Whether a refund should be issued without a person looking is not a question about software capability. It depends on the value, the customer, the policy and the retailer’s appetite for risk, and those differ from one business to the next.
How it works
Every request type has the same shape
Identify what is being asked, retrieve the operational data that answers it, apply the retailer’s policy, decide, act, update the helpdesk, respond. What differs between a delivery query and a damage claim is which data is needed and how much judgement the decision carries.
The mode is set per scenario, not per product
A retailer might resolve delivery queries autonomously, draft refunds above a threshold for an agent, and always escalate anything from a customer flagged as vulnerable. All three run in the same system on the same day.
Some scenarios need no response at all
Marketing outreach, supplier pitches and spam arrive in every support inbox and none of it needs a reply. Classifying and closing that volume is automation nobody counts, and it is a meaningful share of most inboxes.
Some need only an acknowledgement
Where a case will take time to resolve, a templated first response that sets expectations is better than silence and does not need a person to write it.
And some should always reach a human
Complaints, legal threats, anything involving a vulnerable customer. These should be configured to escalate regardless of how confidently the software could answer.
A worked example
IllustrativeThe same request handled three ways, by three retailers with different policies.
- A customer reports a £40 order arrived damaged, with photographs attached.
- Retailer A resolves autonomously: under their policy, damage claims below £50 with usable evidence are refunded without review.
- Retailer B drafts: the same investigation runs, the reply and the refund are prepared, and an agent approves before either goes out.
- Retailer C resolves autonomously but escalates, because this customer has three damage claims this quarter and their policy routes repeat claimants to a person.
One request type, one investigation, three outcomes. The difference is policy, not capability.
| Request | Data required | Decision | Action |
|---|---|---|---|
| Where is my order? | Order, carrier scan history | Delivery state against the promised date | Explain, replace or escalate |
| Cancel my order | Order, fulfilment state | Whether it can still be stopped | Cancel or explain why not |
| I want to return this | Order, return policy | Return eligibility | Create the return |
| I want a refund | Order, policy, customer history | Refund eligibility | Issue or route for approval |
| It arrived damaged | Order, captured evidence | Whether the damage supports the claim | Refund, replace or escalate |
| Wrong item sent | Order, evidence, despatch record | Where the error occurred | Replace or refund |
| My parcel never arrived | Tracking, policy, claim history | Delivery evidence and remedy | Refund, replace, and file a carrier claim |
| Product question | Catalogue, specifications | Whether the answer is documented | Answer or route to a specialist |
| Angry or complaining | Ticket, history, sentiment | Whether judgement is needed | Escalate with the investigation done |
| Marketing or spam | Message content only | Whether a reply is warranted | Close without response |
Common mistakes
- Asking which tasks can be automated. Almost all of them can, and the answer tells you nothing about which ones should be in your operation.
- Setting one mode for everything. A business that automates nothing until it trusts everything automates nothing for a long time.
- Automating the hard cases first to prove the technology. Start with the volume, because that is where the capacity is.
- Forgetting the non-response categories. Spam and marketing outreach take up real inbox space, and closing them correctly is the easiest automation in the list.
- Leaving escalation implicit. The cases that must always reach a person should be configured explicitly rather than left to a confidence threshold.
What we see in the data
At one live account, 60% of total inbound ticket volume is handled without an agent opening the ticket. That figure covers the whole inbox rather than a selected ticket type, and it rose steadily as the retailer moved individual scenarios from drafted mode into autonomous mode.
How this was measured →Checklist
- ✓Categorise a month of tickets by what was actually being asked, not by the queue they landed in.
- ✓For each category, list the systems an agent has to open to answer one.
- ✓Decide the mode for each: drafted, autonomous, or autonomous with escalation.
- ✓Identify the scenarios that need no response at all, and the ones that need only an acknowledgement.
- ✓Write down the cases that must always reach a person, and configure them explicitly.
- ✓Start with your highest-volume category in drafted mode, then move it to autonomous once the drafts are consistently right.
Questions
What ecommerce customer support tasks can be automated?
Delivery queries, cancellations, returns, refunds, damage and wrong-item claims, missing parcels, product questions, and the classification of spam and marketing outreach. Complaints and sensitive cases can be investigated automatically but should be escalated to a person rather than resolved.
What are the modes of support automation?
Three. Drafted, where the software investigates and writes the reply for an agent to send. Autonomous, where it resolves and responds without an agent opening the ticket. And autonomous with escalation, where it begins autonomously and hands over when the case proves to need judgement or is sensitive.
Can refunds be issued automatically?
Technically yes, and many retailers do so within defined limits. Whether you should depends on value thresholds, customer history and your own policy. A common pattern is autonomous below a value threshold, drafted above it, and always escalated for customers with an unusual claim pattern.
What should never be automated?
Anything involving a vulnerable customer, legal threats, and formal complaints. These can still be investigated automatically, so the person picking them up starts with a full picture rather than a blank ticket, but the response should come from a human.
How much of an inbox is typically automatable?
It depends on ticket mix and on how much of the surrounding operation is reachable through APIs. At one live account, 60% of total inbound volume is handled without an agent, and the share rose as individual scenarios moved from drafted mode into autonomous mode.
Sources
- Zendesk — developer API documentationPrimary
- eDesk — help centrePrimary