AI outbound calls for ecommerce: what actually works in 2026
AI outbound calls let a voice agent phone your customers about abandoned carts, failed payments and win-backs. What works, what it costs, what to check.
An AI outbound call is a phone call placed by a voice agent instead of a person. For an ecommerce brand, the agent dials a customer from a segment you choose, has a real conversation about their cart or their order, and can create the order and text the payment link before hanging up.
The technology stopped being the hard part in 2025. What separates a campaign that recovers revenue from one that annoys your list is which calls you hand to an AI, which you keep for a person, and how you set the rules. This guide covers all three, with the exact settings that matter.
Key takeaways
AI outbound works best on high-volume, low-negotiation calls: order confirmations, abandoned checkouts, failed payments, replenishment reminders, win-backs.
It works worst on high-AOV and relationship calls: VIP outreach, wholesale, anything where the customer expects to be recognised.
Pricing is per minute, not per seat. On Consio an outbound call costs $0.10 per minute plus $0.30 per minute of AI, so a two-minute AI call runs about $0.80.
The live handoff to a human is what makes the rest acceptable. If the agent cannot transfer mid-conversation, expect complaints.
Consent and calling rules are the real constraint, not voice quality. Opted-in customers, do-not-call exclusions, calling windows in the customer's own time zone, an automatic opt-out path.
Measure reach and pitch separately. Reach problems and pitch problems have opposite fixes, and most teams misdiagnose which one they have.
Why the phone came back for ecommerce
Email and SMS got cheap, then got crowded. A phone call is the one channel where a customer with a question gets an answer in ten seconds, and the only channel where you can handle an objection while the cart is still open.
The numbers from brands running the phone channel are not marginal. Houndsy generated $55,879 from the channel in a single month, and CEO Pavan Bapu calls it "one of our highest performing revenue channels, full stop.
Both of those campaigns were run by human agents, which is worth being precise about. The phone works. AI is what makes the phone affordable at volume.
What an AI outbound call can actually do in 2026
Most voice products sit in the gap between "the AI can hold a conversation" and "the AI can complete a transaction." Here is what a production ecommerce voice agent does on an outbound call, using Consio as the reference implementation.
It dials with context already loaded. The cart, the full order history, previous calls, SMS exchanges from the last 90 days, and your policies are in the agent's context before the phone rings. It does not run a lookup after the customer says hello.
It searches your catalogue properly. Search runs across product names and descriptions, so "the blue merino one" resolves even when the customer does not know the SKU name. It reads live stock levels, pitches what you can actually ship, and offers alternatives when something has sold out.
It creates the order. The agent builds a Shopify draft order during the call, applies the discount code you configured, and texts the checkout link to the number it is calling, even when you have no email on file. This is the single capability most vendors do not have.
It sends things mid-call. Checkout links, product links with the exact variant, tracking information. Over MMS it sends photos, return labels and PDFs (JPG, PNG, GIF, PDF, up to 5MB per file).
It cancels orders when that is the right outcome. Unfulfilled orders inside a cancellation window you configure, with a refund and a restock, restrictable by order tag.
It transfers to a person, live. To a teammate or an external number, with the conversation context attached. The external leg appears in the transcript under its own section, so a transferred call is not a black box.
It ends dead calls fast. A customer who picks up and says nothing gets one prompt, one warning, then a clean hang-up inside 20 to 25 seconds. Older agents burned minutes on silence.
It logs itself. Disposition, recording, transcript and an AI summary land on the customer timeline. Completed-call events, call summaries and attributed orders fire as webhooks into your own systems, and helpdesk tickets are created automatically in Zendesk, Gorgias, eDesk, Richpanel or Kustomer after outbound calls as well as inbound ones.
If a vendor cannot demonstrate the order creation and the live transfer, you are looking at a deflection bot with a dialer bolted on.
How an AI outbound campaign is actually configured
This is the part buyers never see in a demo deck, and it is where campaigns succeed or fail. Every setting below is a real field in a Consio campaign.
The list: any Shopify or Klaviyo segment. Abandoned checkouts, churn risk, failed payments, VIPs, replenishment, browse abandonment, B2B. You point the campaign at the segment rather than uploading a CSV.
Continuous or one-time execution. A continuous campaign re-syncs the contact list with the latest data from the selected segments every two hours, so new abandoned carts enter the queue on their own. A one-time campaign pulls the list once at the start and never updates it. Cart recovery wants continuous. A seasonal win-back wants one-time.
The caller: human agents or AI agent. On the same campaign you choose either human agents, where your team picks up scheduled calls from the campaign queue, or AI agent, where the AI places every call automatically. The segments, the script and the reporting are identical on both.
The agent type. Outbound calls are placed only by an agent of type Outbound Sales Agent, which ships with its own base prompt tuned for product discovery and conversion. Your inbound agent stays exactly as it is.
The script. An opener, a follow-up and a closing script per campaign, editable, with AI-generated guidance so the pitch and the objection handling stay consistent across every call and every caller.
Calling hours in the customer's time zone. A window such as 09:00 to 19:00, evaluated against the contact's local time, not yours.
Dial attempts and contact caps. How many times a contact is called back, and a ceiling on how often one person can be contacted during a campaign.
Per-attempt voicemails. A different pre-recorded voicemail for each attempt, dropped automatically the moment voicemail is detected. The dialer then sends a follow-up SMS or moves to the next call.
Exclusions. Do-not-call exclusions applied at segment level, plus a state-level toggle to exclude likely Texas residents.
Call pacing. Calls are placed one at a time to protect your number's reputation with carriers, which matters more than it sounds once you are dialling thousands of contacts a week.
The use cases that pay for themselves
Use case | Why the phone beats email and SMS | Best caller |
|---|---|---|
Order confirmation | Short script, zero negotiation, catches address and variant errors before fulfilment | AI |
Abandoned checkout | The objection is answerable in one sentence, and the payment link goes out mid-call | AI below your AOV threshold, human above |
Failed payment and dunning | Urgent and mechanical, worth far more inside the first hour than the next day | AI |
Replenishment reminder | Large lists, low stakes per call, obvious value to the customer | AI |
Win-back on lapsed buyers | Base too large for a human team to work at all | AI |
Post-purchase upsell | The AI qualifies interest, a person closes it | Both |
VIP outreach and wholesale | The customer expects to be recognised, and the conversation is a negotiation | Human |
The last row matters more than it looks. Being explicit about where you do not use AI is what makes the rest of the programme credible to your own team, and to your customers.
Where AI outbound still fails
Anyone selling you AI outbound without this section is selling you a pilot that dies in week three.
Messy contact data. Numbers get reused across profiles and one customer often has two Shopify accounts. If the platform does not merge profiles and hold multiple numbers and email addresses per contact, your connect rate suffers and your attribution lies to you. Consio merges automatically on confirmed contact info and holds several Shopify and Klaviyo profiles under one contact, with protection against over-merging on shared or dummy numbers.
Variant matching. "The blue one, medium" is easy for a person and hard for an agent working a catalogue of forty SKUs. Catalogue search has improved a lot, and it is still the thing to test on your own products before you scale.
Tone on a premium brand. A transactional agent on a luxury product reads as a downgrade. This is a prompt and voice problem rather than an unsolvable one, and it needs a real review pass with whoever owns your brand voice.
The unanswerable call. Some conversations should not be attempted. The agent's job there is to recognise it fast and transfer, rather than improvise.
No baseline. Without a control group that receives no call, you cannot separate recovery from coincidence. This is the most common failure and the cheapest to avoid.
AI, human, or both
Three modes cover almost every ecommerce campaign.
AI runs it. Volume, 24/7 coverage, marginal cost per call. Order confirmations, dunning, replenishment, win-backs, and the bulk of abandoned checkouts.
Humans run it. Judgement, negotiation, relationship. High-AOV carts, VIPs, wholesale, anything where the customer's first sentence is unpredictable.
AI opens, a human closes. The agent works the list and transfers live the moment the conversation is worth a person. Most brands land here after a month, and it is the configuration to insist on before signing anything, because it is the only one where a bad AI call does not become a lost customer.
The practical test: set an AOV threshold. Below it the AI calls, above it a person does. Then move the threshold based on your own numbers rather than a vendor's benchmark.
What AI outbound calls cost
Two pricing models exist and they behave very differently at scale.
Per minute. You pay for talk time. On Consio that is $0.10 per minute for the call plus $0.30 per minute for the AI, so a two-minute AI call is roughly $0.80 and a thousand of them about $800, whatever time of day they happen.
Per month with a minute bundle. You commit to a block and pay overage beyond it. Ringly publishes $349 per month for 1,000 minutes and $799 for 2,500, with per-minute overage after that.
Per-minute is easier to model against revenue and easier to stop. Bundles are cheaper per minute if you reliably consume the whole bundle, and expensive if you do not.
Either way, the number that decides whether the programme survives is cost per recovered order, not cost per minute. Work it out before your first campaign from four inputs: average call length, connect rate, conversion on connected calls, and your AOV.
For comparison, a human outbound agent through a BPO typically runs $12 to $14 per hour plus a per-sale commission and handles 25 to 30 calls an hour. AI is an order of magnitude cheaper per attempt. Whether it is cheaper per dollar recovered is the comparison that actually matters, and it depends on your segments.
Which platforms actually place outbound calls
Half of the "AI phone agent" category for ecommerce does not do outbound at all. Checked September 2026 against each vendor's own site:
Platform | Outbound calls | AI, human, or both | Built for |
|---|---|---|---|
Yes | Both, on the same campaign | Shopify and ecommerce | |
Yes | Both, unified agent desktop | Enterprise, multi-vertical | |
Yes | AI only | B2B and B2C ecommerce | |
Yes | AI only | B2B and GTM teams | |
SMS, not calls | Human agents | Ecommerce cart recovery | |
No, inbound only | AI, escalates to your team | Shopify brands | |
Primarily inbound | AI, escalates to humans | Retail, healthcare, transport |
If your problem is missed inbound calls rather than outbound revenue, buy an inbound product. Two of the seven above are excellent at that and cannot dial out.
Consent, do-not-call and the rules you have to follow
Handle this before the launch rather than after. It is the section that stops deals.
Calling your own opted-in customers about their own order is a different legal proposition from cold outreach, and the platform should give you the controls to keep it that way:
Consent. Call people who opted in, on a lawful basis you can point to. Do not run AI outbound against purchased or cold lists.
Do-not-call exclusions applied at segment level, plus state-level exclusion where a state maintains its own registry.
An automatic opt-out path. On Consio a customer who replies STOP or UNSUBSCRIBE has the conversation closed and is added to the outbound blocklist for both calls and SMS. If they message again later the restriction lifts automatically. Customers a human agent blocked manually stay blocked until an agent removes it.
Calling windows in the contact's own time zone, with a contact cap per campaign.
Recording controls. Pause and resume mid-call, and delete a recording after the fact when a customer asks.
One thing we will not tell you, because it is not ours to decide: whether you must disclose that the caller is an AI. In the US that obligation varies by state and the rules have been moving. Get a position from your own counsel, write it into the agent's opening line, and keep it consistent. Brands that lead with it tend to find it costs them nothing.
How to launch your first campaign in a week
Pick one segment, not five. Abandoned checkouts above your AOV median, or failed payments. Something with a single clear objection.
Set the guardrails before the first dial. Calling hours, attempt cap, DNC exclusions, opt-out. Not after the first complaint.
Write the opener and the objection handling. Two paragraphs: who is calling, why, and what the customer gets.
Choose continuous or one-time execution deliberately. Cart recovery is continuous. A seasonal push is one-time.
Run 100 calls and listen to 20 of them yourself. The recordings, not a sample of transcripts. You will hear things no dashboard reports.
Keep a control group that receives no call.
Fix the two worst moments, then scale the list.
Connecting Shopify and getting a number takes minutes. The week is script work and listening.
How to measure it properly
Volume metrics tell you nothing alone. Three layers do.
Reach. Answer rate and connection rate. If these are low, the problem is your data or your calling windows, and a better script will not fix it.
Pitch. Conversion on connected calls. If reach is fine and this is low, the problem is the conversation.
Revenue. Attributed orders, average order value, revenue per call. Choose the attribution model deliberately: first-touch credits the earliest qualifying interaction, last-touch the most recent, and a 3-day window is available for attributable events. Keep it consistent so month-to-month comparisons mean something.
The funnel in order: customers contacted → customers reached → sales attributed, with the conversion rate at each step. Where the drop-off sits tells you which layer to work on. Most teams who think they have a pitch problem have a reach problem.
Two more things worth knowing about the reporting. Orders attributed to an AI agent can be isolated with a filter, so you can compare an AI campaign against a human one on the same segments rather than arguing about it. And a weekly heatmap of call volume by hour tells you when your customers actually answer, which is rarely when it feels right to call.
FAQs
What is an AI outbound call?
A phone call placed by a voice agent rather than a person. In ecommerce the agent dials a customer from a segment, discusses their cart or order, and can create the order and text a payment link during the call.
Is an AI outbound call the same as a robocall?
No. A robocall plays a recording to a list. An AI outbound call is a two-way conversation with a specific customer about their specific order, and the agent can answer questions, take an order, or transfer to a person.
Can the customer ask to speak to a human?
Yes, and this should be non-negotiable in whatever platform you choose. On Consio the agent transfers the call live to your team or an external number with the conversation context attached, and the transferred leg appears in the transcript.
How much does an AI outbound call cost?
On a per-minute model expect roughly $0.40 per minute all in, which is $0.10 for the call plus $0.30 for the AI on Consio. The figure to plan against is cost per recovered order, which depends on your connect rate, conversion and AOV.
Where does the call list come from?
From your store. Consio points a campaign at Shopify and Klaviyo segments: abandoned carts, churn risk, VIPs, replenishment, browse abandonment, B2B. A continuous campaign re-syncs those segments every two hours so new contacts enter the queue automatically.
Can the AI take an order and get paid?
On Consio, yes. The agent creates a Shopify draft order during the call, applies your configured discount code, and texts the checkout link to the phone it is calling, even with no email on file.
How do I know whether the AI or my reps perform better?
Run both on the same segments in the same window, then filter attributed orders by AI agent and compare answer rate, connection rate, conversion and revenue closed side by side.
How long does it take to launch?
Connecting Shopify and getting a number takes minutes. A first campaign with proper guardrails and a script you have reviewed is realistically a week, and most of that week is script work and listening to recordings.
Content Marketing
