Skip to content

AI automation for ecommerce

In a mid-sized ecommerce operation, 60–80% of tickets are four questions: where's my order, when does it arrive, how do I exchange it, how do I return it. All four have answers sitting in a system and none require judgement.

The problem is those four questions arrive via Instagram, WhatsApp, email and site chat, and the support team spends the day copying tracking numbers between tabs instead of handling the cases that genuinely need a person.

Short answer

In ecommerce, most support volume is post-purchase and repetitive: where's my order, how do I change a size, how do I return it. That automates well and resolves 60–80% of tickets. Refunds and policy exceptions stay under human approval.

What we automate first

  1. 01

    Order status and tracking

    Identifies the customer, finds the order and answers with the carrier's real status, not the platform's.

  2. 02

    Exchanges and returns

    Checks the case against policy, generates the label and explains the steps. Anything outside policy escalates.

  3. 03

    Product and sizing questions

    Answers from the actual product data and the customer's purchase history, not from generalities.

  4. 04

    Cart recovery and post-purchase

    Re-engages an abandoned cart by answering the specific hesitation, rather than firing off an automatic discount.

What it integrates with

  • Ecommerce platform
  • Shipping and logistics
  • WhatsApp Business API and Instagram
  • Payment gateway
  • Helpdesk

What stays under human approval

  • Refunds and credit notes.
  • Exceptions to the returns policy.
  • Claims for damaged or undelivered orders.
  • Any customer who asks for a person.

Where to start

  1. 01Classify the last thousand tickets by type. The order of attack falls out of that.
  2. 02Automate order status — usually 40% of volume and pure data lookup.
  3. 03Add in-policy exchanges and returns.
  4. 04Only at the end, product questions, which depend on the quality of your product data.

When it isn't worth it

  • If the real problem is logistics — orders arrive late — automating support just delivers bad news faster.
  • If you sell few, very high-ticket orders, personal service is part of the product.
  • If your product data is thin, the agent can't answer product questions and will escalate nearly everything.

Related questions

Can it process a refund on its own?
Technically yes, and in the first version we recommend against it. A refund is irreversible and moves money: the agent assembles the complete case with verification done, and a person approves in one click. After three months of data you can automate it by amount tier.
Does it work for Instagram support?
Yes, and in several categories it's the highest-volume channel. The difference from WhatsApp is that many enquiries arrive as story replies with no context, so identifying the customer and the order takes more back-and-forth before anything concrete can be answered.

Next step

Tell us which process is eating your month.

Thirty minutes, no commitment. You leave with a map of the process and an estimate of how much can be automated. If we don't see a case, we say so on the same call.

Response
Under 24 business hours
Mode
Remote · LATAM, Spain, the United States and the Gulf
Languages
Spanish, English and Arabic
Message us on WhatsApp