Published 27 October 2026
Email and SMS own most of the abandoned cart recovery conversation, and for good reason: they’re cheap, automated, and easy to measure. But anyone who has run a cart recovery program for more than a quarter knows the diminishing returns are real. Open rates on recovery emails have been sliding for years, and the customers who haven’t unsubscribed from marketing texts are often the ones least likely to respond to another one. A phone call isn’t a replacement for that program, but it reaches a slice of customers and a class of problem that text-based channels consistently miss.
Why some abandoned carts aren’t a messaging problem
Not every abandoned cart is abandoned for the same reason. Some are genuine browsing, no real intent, and no amount of follow-up will convert them. But a meaningful share are stalled for a specific, fixable reason: a shipping cost that surprised the customer at checkout, a payment method that failed silently, a size or variant question that never got answered, or a promo code that didn’t apply the way the customer expected. These are exactly the cases where a short conversation resolves in thirty seconds what an email thread would take two days to untangle, if it gets resolved at all.
Order confirmation calls serve a related but distinct purpose. For higher-value orders, a quick confirmation catches address errors, duplicate orders, and fraud flags before they turn into a return or a chargeback, which is a more direct cost saving than cart recovery and one that’s easier to measure cleanly.
Where voice AI actually fits, and where it doesn’t
The honest case for voice AI here isn’t that it outperforms a skilled human agent on persuasion. It’s that it makes a channel economically viable that wasn’t before. No retailer was going to staff a team to call every cart above fifty dollars; the math never worked. An AI agent that can place that call, confirm the likely issue, offer a specific resolution like a corrected shipping quote or a working promo code, and hand off to a human for anything it can’t resolve, changes that math because the marginal cost per call drops to something that makes sense even on mid-value carts.
Where it doesn’t fit is anywhere the conversation needs real negotiation, emotional judgment, or a decision outside a defined set of options. If a customer wants to negotiate a discount outside policy, or raises a complaint about a previous order, the agent’s job is to recognize that and route to a human immediately rather than attempting to handle it. Getting that handoff boundary right matters more to the program’s reputation than almost anything else, since a bad automated call experience spreads faster than a good one.
Designing the call so it doesn’t feel like a robocall
The failure mode that kills these programs isn’t technical, it’s tone. A voice agent that opens with a generic script and can’t adapt when the customer gives an unexpected answer reads instantly as a robocall, and customers hang up or get irritated. The calls that work are narrow and specific: they reference the actual cart contents, they ask one clear question tied to a likely blocker, and they’re built to end the call quickly if the customer isn’t interested, rather than pushing through a script. Brevity and specificity read as respect for the customer’s time; a generic script reads as a sales tactic.
Timing matters as much as script quality. Calling within the right window after abandonment, typically within an hour or two for intent-driven abandons, catches the customer while the purchase is still front of mind. Calling a day later after they’ve already bought elsewhere or lost interest wastes the channel and annoys the customer for no return.
What to track if you pilot this
The metrics that matter are recovery rate relative to a holdout group that gets only the existing email and SMS flow, not recovery rate in isolation, since some of the orders a call appears to recover would have come back anyway. Call answer rate and average call length are useful operational signals, since a program converging toward very short calls and low answer rates may be targeting the wrong segment or calling at the wrong time. Opt-out and complaint rates deserve just as much attention as recovery numbers, since a channel that recovers revenue while quietly damaging customer sentiment isn’t actually a win.
Starting narrow
The programs that work best start on a deliberately small slice: a single product category, a specific cart value range, or a specific abandonment reason that’s already well understood, like a known shipping cost issue at checkout. That narrow scope makes it realistic to tune the script and timing against real results before expanding, rather than trying to build one script that covers every type of cart abandonment at once. It also keeps the opt-out and complaint numbers visible early, before a bad script choice reaches a large volume of customers.
If you’re running a cart recovery program that’s plateaued on email and SMS, or you’re processing high enough order volumes that order confirmation calls could meaningfully reduce fraud and return costs, it’s worth scoping a narrow pilot against your actual numbers. Start a project conversation with us and we can work through whether the math makes sense for your order volume and average cart value.
Related