AI Cold Calling for Ecommerce: Why Warm Outbound Works Better, and What Gets You Flagged

Type AI cold calling into Google and you'll find page after page written for B2B sales teams. Book more demos. Let the bot take the rejection. That's fine if you sell software to companies. If you run an online store, your question is different: should an AI voice call shoppers and try to sell to them?

We don't think true cold calling belongs in ecommerce. By cold, we mean calling people who've never bought from you or visited your checkout. That kind of calling runs into some of the toughest phone rules in the markets most stores sell to, and spam filters watch for exactly that pattern.

Outbound AI can still earn its place in a store, though. Picture a customer who just placed a COD order, or quit a checkout halfway, or missed a delivery. They know your store. The call is about something they started. People call that a "warm" call. Keep one thing in mind as you read: warm describes the relationship, not the consent. A customer who knows you hasn't necessarily agreed to an AI call, and this article keeps those two ideas apart.

A quick note on how to read this. Some parts of this article tell you what a specific source says, such as an FCC ruling, a regulation, or a regulator's guidance. We link each one and try not to read more into it than it says. Other parts describe what we'd do ourselves to keep risk down. Those are our recommendations, not legal requirements, and following them won't make a calling program compliant on its own. Whether you can make a particular call depends on things we can't see from here: what the call says, what type of number you're calling, the technology you use, and which federal, state, or national rules apply. This isn't legal advice. Before you launch, ask a lawyer who knows telemarketing law to review your setup for every country you call.

What AI cold calling means for an ecommerce store

With AI cold calling, an AI voice agent dials the number and runs the whole conversation. It doesn't play a recording. It listens, answers questions, and can act while it talks, for example by texting a checkout link or fixing an address.

B2B cold calls usually reach someone at work, about something their company might buy. Ecommerce calls reach a shopper, usually on their own mobile, often after dinner. People put up with far fewer surprise calls at that hour, and the law protects them more.

Stores that use outbound AI well tend to follow the same rough flow. Something happens in the store first: a shopper abandons a checkout, a COD (cash-on-delivery) order comes in, or a courier can't deliver a parcel. The agent calls soon after, while the customer still remembers. Because it reads the order from the store, it already knows what they bought and where it's going. It handles whatever comes up, whether that's a sizing worry, a wrong address, or doubt about the price. It can send a checkout link on the spot. Then it logs the outcome against the order so your team can see who needs a human follow-up.

Notice that nobody in that story is a stranger.

Why true AI cold calling rarely fits ecommerce

Look at the consent row again. A past purchase gives you a relationship. It doesn't hand you permission to call someone with an AI voice. You generally need to obtain and document the consent required for that call; what counts depends on the market and call type.

Three things make consumer cold calling with AI hard to defend. Start with the law. The FCC has said AI-generated voices count as artificial voices in the US (we cover this below), and calls that use them generally need prior consent. A purchased list rarely comes with any. Then there's the money side: a stranger has no reason to stay on the line with a store they've never heard of. Last, deliverability. When you fire lots of unexpected calls at unrelated numbers, you start to look like the callers that labeling systems hunt for.

So if you planned to load a bought list into an AI dialer, we'd pause and call a lawyer first. Everything after this point is about warm outbound, because that's where we see outbound AI doing real work for stores.

The warm outbound AI calls stores actually use

Here's how we sort the common call types. The last column shows how cautiously we'd treat each one. It isn't a legal classification. What the caller actually says decides whether a call counts as marketing, and your lawyer should make that call. Service calls don't get a free pass either. You still need the right consent for the number and the technology you use, and you should confirm how the rules classify them.

Here's a mistake we see stores make: they slip a discount or a "while I have you" upsell into an order confirmation call. That one line can change how the rules treat the whole call. Boring service calls are one of the cheapest risk controls you'll ever get.

Why stores bother with voice at all

An email can remind a shopper about their cart. A call can answer the question that stopped them. If someone left because they weren't sure about the fit or the delivery date, two minutes on the phone can settle it.

Kovax publishes its own numbers for these calls. Its abandoned cart recovery page says merchants recover "around 55% of abandoned carts" and see "average order value rise about 27% from in-call upsells." Its COD confirmation page lists "95% COD orders confirmed" and "50% lower RTO" (return to origin). Treat those as vendor claims, not industry benchmarks. The pages don't explain how Kovax measured them, over what period, across how many stores, or what each percentage is a share of. For example, the 55% could mean all abandoned carts or only customers who picked up. Your products, customers, and call setup will shape your own results, so run a pilot and measure it.

If you're just starting, service calls like COD confirmation and delivery exceptions usually carry less risk. They're short, customers tend to expect them, and they protect money you already track. Less risk still isn't zero risk, so get the consent right and confirm how the rules classify them. Our guide to RTO in ecommerce shows what a failed delivery costs you.

What the rules say: US, UK, and India

This section reports what specific sources say. It doesn't summarize the whole law anywhere. Every country has exemptions, conditions, and local rules we leave out, and regulators keep changing them. Use it as a list of questions for your lawyer.

United States

This is a partial summary. It draws on the FCC ruling, the FCC's rules, and the TCPA statute, all linked below. It leaves out exemptions, conditions, state laws, and court decisions that may apply to you. Ask a telemarketing lawyer to check each point against your actual calls.

In February 2024, the Federal Communications Commission (FCC) issued a declaratory ruling on the Telephone Consumer Protection Act (TCPA). It held that the TCPA's references to artificial or prerecorded voices encompass current AI technologies that generate human voices. Under the ruling, callers who use these technologies need the called party's prior express consent unless an emergency purpose or an exemption applies.

The kind of consent you need depends on the call and the line. The FCC's TCPA rules at 47 CFR 64.1200 require prior express consent for artificial or prerecorded voice calls to wireless numbers. If those calls advertise or telemarket, the rules require prior express written consent. Advertising and telemarketing calls that use an artificial or prerecorded voice on residential lines need prior express written consent too.

"Written consent" means something specific here. The same rule describes a signed written agreement in which the person clearly authorizes the seller to make these calls to a particular number. The disclosure has to tell them they don't need to sign to buy anything.

The rule also says the message must name the business behind the call at the start and give a phone number. Telemarketing messages of this kind to residential lines must also offer an automated way to opt out.

Calling hours apply to solicitations too. The rule bars telephone solicitations to residential subscribers before 8 a.m. or after 9 p.m., in the called person's local time.

And people can sue. The TCPA lets an individual claim their actual loss or $500 for each violation, whichever is greater, and a court can raise that to as much as three times for willful or knowing violations. Courts decide case by case how to count violations and whether to raise the award. High-volume callers worry for a simple reason: per-violation damages pile up fast across thousands of calls.

Why abandoned cart calls are hard to pin down

People often ask us whether abandoned cart calls need written consent. We can't give a one-line answer. The content matters, since a call that pushes a purchase looks more like telemarketing than one that only flags an order problem. The number type matters, because the rules handle wireless, residential, and business lines differently. The technology matters too: an artificial voice or an autodialer brings in different rules. And some states add their own telemarketing laws on top of the federal ones.

We treat cart calls as telemarketing and collect written consent. That's our cautious policy, and we explain it further down. We're not claiming the law always demands it.

Our outbound sales playbook for ecommerce covers the separate rules for live human callers, including the Do Not Call registry.

United Kingdom

This summarizes one regulator's guidance. Ask UK counsel how the rules classify your calls and what consent you need.

The Information Commissioner's Office (ICO) handles automated and live marketing calls differently under the Privacy and Electronic Communications Regulations (PECR). Its telephone marketing guidance says you must not make an automated marketing call unless the person agreed to that specific type of call. General marketing consent doesn't cover it, and neither does consent to live calls.

Live marketing calls follow a different rule. You must not call a number on the Telephone Preference Service (TPS) or Corporate TPS unless that person specifically agreed to hear from you. In practice, you screen your list against TPS and against your own do-not-call list.

The guidance leaves one question open for us: does a two-way conversational AI call count as an "automated call" under PECR? We assume it does, and we'd plan that way until a lawyer tells you otherwise. That's our cautious assumption, not a ruling. If you take the same view, naming automated or AI calls in your consent request is one way to make the consent specific. But one sentence in a checkbox doesn't settle it. How you ask, how you record the answer, and how people withdraw consent all decide whether it holds up.

India

This summarizes five TRAI press releases. India's rules have changed several times since 2024, twice in the last few months alone, so check the current position with your telecom provider and counsel.

India regulates commercial calls through the Telecom Commercial Communications Customer Preference Regulations (TCCCPR), and the Telecom Regulatory Authority of India (TRAI) keeps updating how they work. Here's the trail.

The crackdown started in August 2024. On 13 August, TRAI directed telecom operators to stop promotional voice calls from unregistered senders and telemarketers, including calls that are "pre-recorded or computer generated or otherwise." TRAI also told unregistered senders making commercial voice calls to move onto the Distributed Ledger Technology (DLT) platform.

In February 2025, TRAI amended the TCCCPR itself. Its 12 February release kept promotional calls on the 140 number series and set aside the new 1600 series for transactional and service calls. It also stopped senders from using normal 10-digit numbers for telemarketing. Penalties went up as well. After a first violation, the rules bar outgoing services on all of a sender's telecom resources for 15 days. Repeat offenders lose all their resources for a year and land on a blacklist. The same release says a sender can't ask a customer who opted out for consent again within 90 days.

Then, on 10 July 2026, TRAI clarified who gets the 1600 series: entities that RBI, SEBI, IRDAI, and PFRDA regulate (banking, financial services, and insurance), for service and transaction calls to existing customers, plus government bodies calling citizens. Businesses in any sector that make promotional calls must register under the TCCCPR and use the 140 series. Customers can block 140-series calls from one sector, or from all sectors, through the Do Not Disturb (DND) registry.

A month later, on 10 August 2026, TRAI announced a phased 1601 series for service and transactional calls from businesses outside banking, finance, insurance, and government. Phase I covers utilities and courier, logistics, and parcel delivery companies, and TRAI gave telecom providers 90 days to onboard them. Providers allocate the numbers directly to eligible businesses, not to intermediaries or aggregators, and businesses can't use them for promotional calls. The release doesn't mention ecommerce brands or online retailers.

The latest change came on 18 September 2026, when TRAI announced further TCCCPR amendments aimed at technology-driven enforcement. These define application-to-person (A2P) calls as voice calls that an application, software system, or automated platform starts without a person dialing, including calls with an artificial voice. Businesses must tell their telecom provider in advance that they make A2P calls, and from which numbers. TRAI treats undeclared calls as unsolicited commercial communication. The release also says older consents only count if the business got them in a verifiable way and registered them on the provider's digital platform. Providers will use AI and machine learning to flag numbers that look like spam.

So what does all this leave unsettled for an online store? Quite a lot. An AI voice agent's calls appear to fit the September 2026 definition of A2P calls, which would mean you declare them to your telecom provider before you call. Check that with your provider, and ask who files the declaration when your calls run through a voice platform. The number series question stays open too. The 1601 series covers courier and logistics companies in Phase I, which may matter if your delivery partner calls your customers. None of these releases say which series an ecommerce brand should use for its own service calls, like COD confirmation or delivery exceptions, or whether a later phase will include them. They don't spell out the full registration steps, headers, or consent requirements for your calls either. What they do make clear is that promotional calls, which would cover cart recovery offers, win-backs, and upsells, belong in the registered 140-series framework. Customers who've blocked your sector through DND won't get them.

A conservative operating approach (our recommendation)

None of what follows is a legal requirement, and following it won't make your program compliant by itself. It's simply what we'd do to keep risk low while a lawyer works through the details.

Start with who you call. We'd only call people who already have a relationship with your store, and we'd never touch bought, scraped, or rented consumer lists.

Next, decide how you'll treat each call. We treat anything with a sales angle as telemarketing. For US calls about abandoned carts, reorders, win-backs, and upsells, we'd collect consent that meets the written consent standard, even if you suspect a lower bar applies. Service calls stay clean: no offers on a COD confirmation or a delivery call.

Then handle consent channel by channel. Calls, texts, and WhatsApp can each carry different requirements, and so can each country. We'd give each channel its own checkbox rather than write one sentence that tries to cover everything, unless counsel confirms a combined approach works for every channel and market.

On the call itself, say it's an AI and name your store right at the start. Even where no rule demands it, customers trust you more when they don't have to work it out. Respect calling hours in the customer's time zone, not yours. And when someone opts out, stop everywhere: calls, texts, WhatsApp, all of it.

Finally, keep proof. Record when the customer gave consent, for which number, the exact wording they saw, and where you collected it.

An example consent request (US-oriented, not a template)

Here's what a cautious US setup for AI-voice sales calls might look like. Please don't paste it into your checkout as it stands. We wrote it for the US only, it doesn't cover texts, and nobody has reviewed it for your business. Ask a telemarketing lawyer to draft or approve your real wording.

☐ I agree that [Store Name] may call me at [phone number] using an automated system and an AI-generated voice about my cart and offers. Agreeing is not required to buy anything. I can opt out at any time by telling the caller.

We made a few choices on purpose. The box starts unchecked. It covers calls only, so you'd need a separate request for texts. It names the seller and the technology. And it says plainly that buying doesn't depend on ticking it.

An example call structure (abandoned cart, US-oriented)

A good cart call opens with the store's name: "Hi, this is the AI assistant from [Store Name], calling about the checkout you started earlier." If that opening doesn't make it obvious, the agent should say clearly that it's an AI. It asks whether now's a good time. Then it helps before it sells, with something like "Was there anything about the order you weren't sure about?" Once it has answered the customer's question, it offers the next step: "I can text you the checkout link now if you'd like." Texting may need its own consent. Before hanging up, it offers an opt-out, honors it on every channel, and gives a callback number.

COD confirmations should run shorter and stick to the order: the item, the address, and whether the customer will be home to receive and pay.

Consent is not deliverability: how numbers get flagged

People mix these two problems up all the time. Consent deals with legal risk. It has no control over how carriers display your calls. You can hold perfect consent records and still show up as "Spam Likely," and that label can make people less likely to answer.

Two systems shape how your calls appear on someone's phone.

The first is call authentication. The FCC's STIR/SHAKEN framework lets carriers digitally sign caller ID information so the receiving carrier can check that the call really comes from the number shown. It helps verify caller ID. But authentication varies by carrier and setup, and on its own it doesn't decide whether your calls look trusted or unlabeled.

The second is call analytics. Carriers work with analytics companies that score phone numbers. Three of them, First Orion, Hiya, and TNS, appear on the Free Caller Registry, where businesses can register outbound numbers at no cost. It's worth doing. Still, the registry itself says registration "does not guarantee redress."

The analytics companies keep their scoring private, so nobody outside them knows the exact rules. Labels tend to follow a few patterns: heavy volume on a handful of numbers, long runs of short or unanswered calls, lots of instant hang-ups or reports, calls to people with no link to your business, and a caller ID customers don't recognize.

Nothing guarantees a clean label, but some habits help. Call people who have a reason to expect you. Register your outbound numbers on the Free Caller Registry and use the same business name everywhere. Get numbers from a reputable provider that supports call authentication, and don't rotate numbers to dodge labels. Spread calls across the permitted hours instead of firing them all at once. Check each number's connect rate every week, since a sudden drop often signals a new label. And make opting out easy, because a customer who can't make you stop is more likely to report you.

AI cold calling software: what to look for

Most AI cold calling software targets B2B outreach. A store doing warm outbound needs a different feature set:

We checked each vendor's pricing page on 15 October 2026. Prices move often, so check again before you budget. Kovax is free to install with 50+ AI calls. Its Starter plan cost $49.99 a month with 166 minutes included at 30¢ a minute, and the Growth and Ultimate plans charged less per minute. Developer platforms charge less per minute but need engineering time to connect store data and build call flows. Retell AI listed $0.07 to $0.31 a minute depending on the model and voice. Bland AI listed $0.14 a minute on its Start plan and billed telephony separately. None of these prices cover legal review or the work of building proper consent capture.

Kovax's outbound sales agent shows what a store-specific tool looks like. Store events, like a stalled checkout or a delivered parcel, trigger its calls. It follows up on WhatsApp in a single thread tied to the order, tags contacts by intent, and honors quiet hours by time zone. Like any tool, though, your setup and your habits decide whether you stay compliant.

For a wider look at voice platforms, see our guide to the best AI voice agents for Shopify stores.

When a person should make the call instead

Some calls need a human. High-ticket orders often close on trust in a real person. Wholesale and B2B buyers expect someone who can negotiate terms. Complaint recovery after something went wrong deserves a person too. And whenever a customer asks to speak to someone, hand the call over. Every time.

Pre-launch checklist

Questions for your lawyer

• ☐ How do the rules classify each call type in each market you call?
• ☐ Which consent standard applies to each call type, channel, and number type?
• ☐ Does your consent wording and capture flow work for each channel and country?
• ☐ Which rules beyond the federal or national ones apply? (US state telemarketing laws, UK TPS, and in India, current DLT registration, A2P declaration of AI calls, call categories, headers, and which number series your service calls should use)
• ☐ Do any recent changes to consent or lead-generation rules affect your checkout consent flow?
• ☐ Do your call scripts handle identification, AI disclosure, and opt-out correctly?

Jobs for your team (our cautious practices)

• ☐ Keep call lists to customers and checkout starters
• ☐ Store each consent with a timestamp, number, exact wording, and source
• ☐ Strip offers and upsells out of service calls
• ☐ Follow each customer's time zone for calling hours
• ☐ Stop calls, texts, and WhatsApp the moment someone opts out
• ☐ Register your numbers on the Free Caller Registry and check connect rates weekly
• ☐ Measure a small pilot before you scale

Not sure where to begin? Pick one service call type, like COD confirmation or delivery exceptions, and launch it once you've confirmed its consent and classification. Add sales calls after you've built consent capture and finished legal review.

One last reminder: no single step on this list makes a campaign compliant on its own. That includes a customer relationship, a Free Caller Registry listing, call authentication, and an opt-out line in your script.

Frequently asked questions

Is AI cold calling legal?
That depends on the market, what the call is for, what it says, the type of number, and the consent you hold. In the US, the FCC treats AI-generated voices as artificial voices under the TCPA, so these calls generally need prior consent, and telemarketing calls need written consent. The UK requires consent to automated marketing calls specifically. In India, promotional calls must come from registered senders on the 140 number series, and you have to declare automated calls to your telecom provider in advance. Calling consumers from a bought list with an AI voice carries real risk, so talk to a lawyer before you try it.

Can I use AI to call customers who abandoned their cart?
Possibly. What the call says, the type of number, and any state or national rules all shape how the law treats it. The cautious route is to treat these calls as telemarketing, collect consent that meets the US written consent standard, call within permitted hours, name your store, and offer an opt-out. Then ask a lawyer to confirm your setup.

If someone is already my customer, can I call them with AI?
A relationship isn't consent. A past order makes the call more welcome and easier to explain. Whether you can use an AI voice still comes down to the consent you've actually collected and recorded for that type of call.

Do I have to tell customers they're talking to an AI?
Rules vary by jurisdiction. In the US, artificial-voice calls must name the business responsible at the start. Legal requirements aside, we'd tell people up front anyway. Nobody likes discovering halfway through a call that they've been chatting with a bot.

Does getting consent stop my number showing as "Spam Likely"?
No. Consent handles legal risk, while carriers and analytics companies decide labels on their own terms. Call authentication and a Free Caller Registry listing can help, but neither controls how your calls appear. Your best levers are calling people who expect you and watching your connect rates.

How much does AI cold calling software cost?
When we checked vendor pricing pages on 15 October 2026, developer platforms like Retell AI listed about $0.07 to $0.31 a minute, plus the engineering work to set them up. Ecommerce-ready apps like Kovax listed around 25 to 30 cents a minute. Set aside a separate budget for legal review.

Curious how event-triggered outbound calls work on a Shopify store? The Kovax outbound sales agent page walks through it.

Related reading: Outbound sales for ecommerce: a 2026 playbook · Abandoned cart recovery: email vs SMS vs voice · RTO in ecommerce