After-Sales Incidents with AI: From “It Arrived Broken” to a Fix Without an Email Chain

Monday, 9:15 a.m. A customer emails the general inbox: “My order arrived broken.” At 11 they send a photo over WhatsApp to the sales rep’s phone. At 1 p.m. they call because nobody has replied, and whoever picks up knows nothing about the email or the photo. On Tuesday someone finds the thread, asks for the order number (which was in the very first message) and the customer, quite rightly, gets annoyed. AI aftersales incident handling exists precisely so this scene doesn’t happen again.

In most small and mid-sized businesses, after-sales is a black hole: incidents come in through five different channels, everyone keeps their piece in their own inbox, and nobody can answer the customer’s simplest question: “Where are we with my issue?” It isn’t a lack of goodwill. It’s a process problem.

And it’s expensive. The sale is already made, but the relationship is decided now: a customer who feels ignored after a problem doesn’t buy again, and tells others.

Why after-sales turns into an endless email chain

An after-sales incident isn’t just a complaint: it’s a small process with several steps that are almost never written down. You need to identify the order, check the warranty, work out whether the damage came from shipping or from the product, choose between repair, replacement or refund, and update the customer at every step. When all of that lives in one person’s memory, the usual happens:

• Scattered intake: general inbox, a rep’s WhatsApp, the website form, phone calls. Each channel is an island.
• Context you have to dig up by hand: order number, purchase date, batch, customer history. Every lookup costs minutes.
• Invisible status: nobody knows whether the case is waiting on the carrier, the warehouse or a decision, so the customer calls to ask, which creates even more work.
• Different criteria depending on who answers: one customer gets a refund while another, in the same situation, is told to wait.

The cost of getting it wrong is measured: in a 2024 survey, 63% of consumers were willing to switch to a competitor after a single bad experience (Zendesk, CX Trends 2025). In after-sales, that bad experience is usually silence.

How AI aftersales incident handling works, step by step

You don’t need to replace your ERP or your helpdesk. The AI sits on top of what you already use and brings order to the flow:

1. Single intake. Email, WhatsApp, web forms and call transcripts land in one place. If the customer writes through two channels, the AI recognises it as the same case and doesn’t open two tickets.
2. Classification by type and urgency. Damaged product, wrong shipment, warranty fault, missing parts, return. The AI reads the message (and the photo, if there is one) and assigns type and priority: a stopped machine at an industrial customer isn’t the same as a dented box.
3. Context gathered automatically. From the name, email or order number it queries the ERP: what was bought, when, whether it’s under warranty, previous incidents, which carrier delivered it.
4. Proposed resolution based on your procedure. If at your company “shipping damage with a photo within 48 hours” means a replacement plus a claim against the carrier, the AI proposes exactly that, every time. The rules are yours, not the software’s.
5. A person decides what matters. Refunds, replacements, commercial exceptions and anything outside the norm go to a human with everything ready: summary, photos, history and proposal. Approving takes seconds.
6. Customer updated at every status change. “We’ve received your request”, “your replacement ships tomorrow”, “your refund has been issued”. Automatic messages through the channel the customer used, without anyone having to remember.

When the process is long (warehouse, shipping, invoicing a credit note), it can be split across several specialised agents working together, as we explain in multi-agent systems for end-to-end processes.

After-sales is not the same as a channel or as sales

It helps to separate three things that often get mixed up. The channel is where the customer talks to you: we cover it in WhatsApp Business with AI. The sales pipeline is how you manage a customer before they buy: see AI CRM automation. This article is about what happens after the sale, when something goes wrong and needs fixing in an orderly way. If most incidents reach you by email, you’ll also want to read automating email with AI.

What improves and how to measure it

The first thing you notice is that customers stop calling to ask, because they already know where their case stands. Then the team stops hunting for data and starts making decisions. To know whether it works, track two metrics before and after:

• First response time: from the moment the incident arrives until the customer gets more than an automatic acknowledgement. With AI, a reply with a case number and the next step is immediate.
• Resolution time: from intake to closure. The savings come from not losing hours gathering context or waiting for someone to find the thread.

Add a third if you want to fine-tune: the share of incidents resolved under the standard rule with no exception. If it’s high, your procedure is clear and the AI applies it well; if it’s low, the problem isn’t the tool, it’s the missing procedure.

When does automating after-sales make sense?

• Volume: a few incidents a day is already enough to pay off; with two a month, a shared spreadsheet will do.
• Several intake channels: if everything arrives in a single inbox, the first step can be simpler.
• A procedure, even an informal one: if you can explain what you do for each type of incident, the AI can apply it. If every case is improvised, write the rules first.
• Accessible data: orders and customers in an ERP or online store with an API, exports or database access.

It doesn’t make sense if incidents are very rare, if each one is a unique technical project, or if nobody will review what the AI proposes.

Frequently asked questions

Does the AI decide on its own whether to refund or replace a product?

It shouldn’t. The recommended setup is for the AI to propose a resolution based on your rules and for a person to approve refunds, replacements and exceptions. You gain speed without losing control over anything that costs money.

Do I need to change my ERP or my online store?

No. The AI connects to what you already use (ERP, online store, email, WhatsApp) through APIs, exports or connectors. What changes is the process, not the software.

What happens if the customer writes on WhatsApp and then calls?

The AI recognises the customer and the order and adds the new contact to the same case. Whoever takes the call sees everything that came before and doesn’t ask for the same details again.

Can it understand photos of a damaged product?

Yes. Current models describe the image, identify visible damage such as dents, breakage or missing parts, and attach it to the case. The final assessment, especially for warranty claims, is made by a person.

	Without AI	With AI
Incident intake	Five scattered channels	One single case
Order and warranty data	Looked up by hand	Pulled automatically from the ERP
Status for the customer	Calls to ask	Update at every status change
Resolution criteria	Depends on who answers	Your procedure, every time
Refunds and exceptions	Decided without context	Approved by a person with everything ready

Contact us and we’ll analyse your case for free →