Yes. An AI voice agent can call eligible customers after they abandon checkout, ask whether they need help, answer approved questions, and guide them toward the next step.

The important word is eligible. A phone number appearing in checkout data does not automatically mean a store should place a marketing call. The merchant must consider customer consent, applicable laws, calling hours, data use, opt-outs, and whether the call is appropriate for the situation.

Used responsibly, an AI call can solve a problem that a reminder email cannot: it can discover why the shopper stopped.

What actually triggers the call?

The workflow normally begins with an ecommerce event. A customer starts checkout, provides enough contact information, and leaves without completing payment. The store or integration then waits for a defined period and checks whether the order is still incomplete.

Before dialing, the system should apply rules such as:

  • Has the customer completed the order since abandoning?
  • Is the phone number valid?
  • Is the customer eligible to be contacted?
  • Is the current time appropriate in the customer’s location?
  • Has the customer opted out?
  • Has the store already attempted contact too many times?
  • Does the cart meet the store’s value or product criteria?

Only then should a call be created.

What does the AI say?

The opening should be clear, brief, and non-deceptive. The agent should identify the store and explain that the call concerns an incomplete checkout.

A useful conversation starts with permission:

“Hi, this is the automated assistant from [Store]. You recently started an order but did not finish checkout. Did you need help with anything?”

From there, the agent can follow the customer’s intent. If the customer says the call is unwanted, it should stop. If the customer asks a question, the agent should answer only from information the store has approved.

Which questions can an AI cart-recovery call handle?

Common examples include:

  • Delivery estimates
  • Shipping costs
  • Return and exchange rules
  • Product size or compatibility
  • Stock availability
  • Checkout or payment confusion
  • Whether a promotion applies
  • How to return to the incomplete checkout

The agent should not improvise policies or make promises outside its authority. For example, it should not invent delivery guarantees, issue unauthorized discounts, or claim a payment succeeded without reliable data.

Why call instead of sending another message?

Abandoned-cart email and SMS are useful, but they are mostly one-way until the customer chooses to reply. A call immediately gives the customer a way to explain what went wrong.

Voice is especially useful when:

  • The order value justifies a higher-touch recovery attempt
  • The product requires explanation
  • Shoppers frequently ask pre-purchase questions
  • Delivery timing matters
  • The customer attempted payment but encountered friction
  • The store sells products where compatibility or sizing creates uncertainty

For a low-value, impulse purchase, a call may feel excessive. For a high-consideration purchase, it may feel like useful assistance. The strategy should match the product and audience.

How to avoid sounding intrusive

The customer experience depends on restraint.

First, do not pretend the AI is a human. Second, make the reason for the call obvious. Third, keep the opening short. Fourth, make declining easy. Fifth, avoid repeated attempts.

The agent should also understand negative intent. Phrases such as “do not call me,” “I am not interested,” or “remove my number” should trigger an immediate stop and the appropriate suppression workflow.

Consent and compliance are part of the product design

Rules for automated and marketing calls differ by country and sometimes by state or region. Transactional support, marketing, and recovery outreach may be treated differently. The store is responsible for determining the lawful basis for each workflow.

A responsible implementation includes:

  • Documented consent rules
  • Regional calling restrictions
  • Time-of-day controls
  • Opt-out handling
  • Data minimization
  • Call disclosure where required
  • Retention rules for transcripts and recordings
  • A human escalation path

This article is operational guidance, not legal advice. A merchant should review the program with qualified counsel before launching in a new jurisdiction.

How Kiyoto approaches the use case

Kiyoto provides voice AI for two parts of the ecommerce customer journey: outbound cart-recovery conversations and inbound customer-service calls.

For cart recovery, the goal is to contact approved segments, identify friction, answer approved questions, and guide the shopper back toward purchase. For customer service, the same knowledge can support incoming questions about products, shipping, returns, and orders.

Connecting these workflows helps the brand maintain consistent answers before and after purchase.

A practical launch checklist

Before your first call, confirm:

  1. 1The triggering event is accurate.
  2. 2Completed orders are excluded.
  3. 3Consent and regional rules are implemented.
  4. 4The agent identifies the store and the automated nature of the call.
  5. 5Opt-outs stop future contact.
  6. 6The knowledge base contains only current, approved information.
  7. 7Unsupported questions escalate.
  8. 8Outcomes are tracked.
  9. 9A small pilot will be reviewed before expansion.

AI can call abandoned-cart customers. The better question is whether your store can do it in a way that is useful, accurate, and respectful. That is the standard the workflow should meet.

See how Kiyoto can support conversational cart recovery and inbound ecommerce customer service.