Outbound AI Calls for Ecommerce: Plays, Timing, Rules and Cost

Abandoned-checkout calls are the outbound AI calls with a clear estimated lift so far; calls after a delivery, a failed payment or a lapse in reordering are worth testing, not assuming. Call abandoned checkouts 5 to 15 minutes after they stall, schedule the rest inside legal calling hours, call only people whose written consent covers AI calls (the calling rules), and judge the program by the orders it adds over a holdout group.

The timing comes from Loop Voice's September 2026 benchmark of 18,361 shoppers across 7 stores: checkouts first called 5 to 15 minutes after abandoning had a 22.9% five-day order rate, against 4.1% when the first call came 15 to 60 minutes later. In the same benchmark, post-purchase upsell calls showed no clear lift (+0.3 points, 95% CI -0.6 to 1.3), and failed-payment and win-back flows were not live yet. Loop Voice publishes this page and is one of several tools that place these calls; the benchmark is its own, and an estimate rather than a controlled test.

Loop Voice flows start from events such as abandoned checkout, order fulfilled, second order (VIP) or a custom event, and campaigns (in beta) call a Shopify or Klaviyo segment, abandoned checkouts or inactive customers inside a daily calling window and pace you set.

Only the checkout timing has benchmark data behind it; treat the other timings as starting points to test. Loop Voice's example checkout flow waits ten minutes, calls, texts the checkout link, waits, calls again with a discount and sends a last note, and flows stop the moment the customer orders, says stop, or after a negative call.

A failed payment can also turn up inside a checkout flow: in one real call, an automotive parts store's order had been canceled after a declined payment, so the agent texted a link to reorder and stayed on until it confirmed.

Viewed product, added to cart and added to list can start flows too; keep those calls to customers who already gave call consent, and prove them against a holdout first.

In the benchmark's abandoned-checkout flow, 14,036 shoppers were called and 16.5% picked up at least once. The table keeps orders within 5 days apart from answers, which are a different measure.

Three rules follow:

1. Call 5 to 15 minutes after the abandon. Let the flow place the first call automatically, not a person working a list.
2. Don't write off after-hours checkouts. Called the next day, they ordered at 6.5%, better than in-hours checkouts first called 15 minutes to 4 hours later.
3. Retry in the early evening. On weekdays, calls placed 5 to 9 p.m. were answered 7.0% of the time against 5.4% for 8 a.m. to noon, counted per call.

The benchmark covers 7 stores, and the timing groups were not randomly assigned. In the 5 to 15 minute group more shoppers ordered (22.9%) than answered (20.5%), so some of those orders came without a conversation, and fast calls may reach shoppers who were only briefly interrupted. Use the table to decide when to call and your own holdout to learn what the calls add.

Among 824 answered benchmark calls with a recorded reason for not buying, 33.4% of shoppers were browsing, 25.6% named discount or price, 11.7% a site or checkout error and 5.5% a product question. Baymard's survey of US online shoppers points the same way: 42% had abandoned a cart because they were just browsing, and setting those aside, 17% named website errors or crashes. A good call finds out which case it is with its first question and handles only that one.

Jay's Modern Hearing order is the site-error case: the agent stayed on the line while he got the site working on another computer and rebuilt the order, until an add-on offer appeared at the last step.

The agent called the add-ons optional, didn't pitch them, and stayed through each snag until the confirmation appeared.

The second call is post-purchase. A check-in for a dog-chew brand found a packing mix-up; the agent opened a support ticket for the right chews, and only then brought up the next order.

Fix first, offer second: the 10% code came after the problem was handled, and the agent had Leon read it back.

The same habits run through Loop Voice's public call library:

• Says what it is. A furniture shopper asked, "are you a human or are you AI?" The agent: "I'm an AI assistant. Yep. Totally happy to help either way, though."
• Sends the link. The same call closed with "I just texted you the code ... and I included that checkout link again too."
• Lets the shopper go. A gardening shopper answered "I'm all set." when the agent offered to stay on the line while he checked out, and the agent thanked him and said goodbye.

In February 2024 the FCC ruled that calls made with AI-generated voices are "artificial" under the Telephone Consumer Protection Act, so AI calls carry the consent, identification and opt-out duties of prerecorded robocalls (FCC declaratory ruling 24-17). The practical version is below, not legal advice; our TCPA compliance guide for voice calls goes deeper on consent.

A past order is not written consent, so win-back and VIP campaigns should call only customers whose consent covers AI calls. Shopify's checkout opt-ins (Settings > Checkout > Marketing opt-in) cover email and SMS, and Shopify's help page on them doesn't mention calls, so add your own unchecked consent line naming AI-voice calls wherever you collect the phone number.

The FTC's guidance for sellers says the National Do Not Call Registry doesn't block calls to consumers who signed a written agreement, including their number, to get your calls; sellers who have to use the registry must sync with it at least every 31 days.

The 8 p.m. cutoff in Florida and Maryland falls inside the 5 to 9 p.m. window that answered better than mornings. If you call Florida or Maryland numbers, end the window at 8 p.m. local time, or end it there for everyone. A weekly cap of three or fewer keeps one agent within the 24-hour limit; caps are set per agent, so if several agents can call the same customer, count their caps together.

Never let the agent claim to be human; Loop Voice's guardrails include clear identity as an AI when asked.

Loop Voice uses credits only when someone answers: an answered conversation uses about 1.5 credits on average, unanswered calls use none, and answered calls that don't convert still use credits. On monthly billing that is about $0.95 to $1.49 per answered conversation, depending on the plan (Loop Voice pricing).

Yearly billing lowers each price, and extra conversations are billed at the plan's per-extra-credit rate.

Example: a store has 600 abandoned checkouts a month with a phone number and call consent, a $90 average order and a 40% gross margin. At the benchmark's 16.5% pickup rate for abandoned checkouts, about 100 of those shoppers answer at least once, about 150 credits (more if some answer both calls), which Growth covers and Starter doesn't.

At about 100 conversations, Growth costs about $4 per answered conversation, not $1.20, because the low figures assume the credits get used. Counting the plan price only, covering the $399 takes about 11 extra orders a month at $36 of margin each, a lift of about 1.8 points on 600 checkouts. For abandoned checkouts, the benchmark's estimated lift was +5.7 points (95% CI 4.4 to 6.9, calls and texts together, from 5 of its 7 stores, as the abandoned cart page reports): this store clears the bar even at the low end, but the estimate is not a randomized test, so its own holdout decides.

Loop Voice ties revenue to the Shopify order number, with an attribution window you set per workspace, and reports it per flow, campaign, channel and cart value band. That figure counts every order that followed a call or text inside the window, including orders that would have happened anyway, so it overstates what the calls add.

The benchmark's call and text conversion rates also count orders that followed a touch, not orders the touch caused. Across both live use cases, AI calls and texts added an estimated +2.8 points of 5-day conversion (95% CI 1.9 to 3.7) for the shoppers they reached, measured against enrolled shoppers the flows never reached in the same 6 stores, so even that is not a randomized holdout.

Loop Voice's four published customer stories report attributed revenue, the same kind of figure: Modern Hearing (hearing aids) $300,000+ with 11% of calls becoming a sale, Warrior Will Power (health and wellness) $200,000+ from 1,184 calls at 7.1%, CityMan USA (streetwear) $45,000+ at 5%, and CastleFlexx (mobility and recovery) $30,000+ at 8%.

To find your own incremental number, hold out a random share of eligible checkouts, change nothing else, and compare order rates over your attribution window; how to run a holdout test on cart recovery calls covers the setup. Loop Voice's A/B split between two agents on the same flow tells you which agent sells better, not whether calling adds orders.

Read these every week:

Loop Voice's pricing page says the calls work best with a $50+ average order, 100+ abandoned carts a month and US customers, with a minimum viable volume of roughly 50 to 100 abandoned carts a month; below 50, focus on traffic first. Count yours in Shopify admin under Orders > Abandoned checkouts; Shopify lists a checkout there once it sits incomplete for more than ten minutes after the shopper gives an email, and deletes ones older than three months.

Skip or pause outbound calls when:

• You have fewer than 50 abandoned checkouts a month. Too few calls to learn from, and too few orders to carry the plan.
• Orders are small. Example: at a $30 average order and a 40% margin, each order earns $12, so Starter's $119 needs about 10 extra orders a month, a 10-point lift on 100 checkouts, above even the top of the benchmark's abandoned-checkout range (+5.7 points, 95% CI 4.4 to 6.9).
• No consent covers AI calls. Fix consent capture first, and use email, plus texts where shoppers opted in, in the meantime.
• Most customers are outside the US. The pricing page's fit names US customers.
• Calls keep finding the same checkout bug. Fixing the checkout is cheaper than calling around it.