The main support number rings your team as usual. After four rings, when all agents are busy, or after 6 pm, your phone system forwards to a VoiceGenie number. Other requests become a callback ticket with a summary.
Every order call answered, every refund decision made by your team
The AI answers at once and finishes order status, returns and address changes. Your team keeps refunds, damaged goods and upset customers. Together they cover more calls, with better service, than either alone.
- Order status on the first ring
- Can say it is an AI, every call
- Refunds stay with your team

Where is my order?
Order 48213 shows in transit, two days past the promised date.
Start a return
The jacket is too small, day 12 of a 30-day window.
Cracked on arrival
A lamp base broke in transit. An agent decides replace or refund.
Stop my subscription
A coffee plan customer says cancel. The AI offers a pause once.
Why online store phones will run on AI and people together
37% of online shoppers had a late package last year, and 55% prefer a phone call for an urgent order problem. US contact centers answer in 99 seconds on average, and 8.9% of callers hang up first. A people-only team cannot answer every tracking call in peak week. An AI-only line cannot judge a cracked lamp or a fraud hold.
- Answers every where-is-my-order call on the first ring, nights included
- Starts returns and fixes addresses before the label prints
- Quotes your written policy the same way on every call
- Decides replace or refund on a damaged or missing item
- Spots fraud and repeat claims that a script would miss
- Saves a loyal customer with a fair exception and a kind word
- Night and peak-week calls are answered, so fewer shoppers give up or dispute the charge
- The agent opens a damaged-item call with order, photos and summary, so the story is told once
- Your team stops reading tracking numbers aloud and spends the day on claims and upset customers
If the answer sits in your order system or return policy, let the AI finish it. Start with tracking, return labels and address changes.
Refunds past your limit, damaged goods, fraud holds and chargebacks need an agent with authority. Route them with the order and transcript.
Measure whether the order issue was finished and the customer did not call back. A short call that ends in a second call is not a win.
Can AI handle ecommerce customer service calls like order status, returns and cancellations?
Yes, for calls that are lookups and rule-based actions: order status, address changes before shipping, return labels, simple cancellations and policy questions. Your team should take refunds outside policy, damaged or lost orders, fraud holds and anyone who asks for a person. Start with order status on overflow and after hours, read-only, and expand after a measured pilot.
- Order status and returns are mostly lookups, so they suit AI first
- Refund exceptions, damaged goods and fraud holds stay with your team
- Card numbers should never reach the AI, the transcript or the recording
- No federal click-to-cancel rule is in force in Nov 2026; ROSCA and state laws are
Returns, late parcels and cancellations are where the pressure sits.
No public study splits ecommerce calls by reason, and the widely repeated share for 'where is my order' has no traceable source. These figures show where the pressure sits. Tag your own calls for two weeks before you plan anything.
Forecast US retail returns in 2025: 15.8% of sales and 19.3% of online sales (NRF and Happy Returns)1
of 1,348 US online shoppers had a late package last year; 23% were marked delivered but never got it (Narvar, 2026)2
of 5,000 US mobile subscribers prefer a phone call for an urgent order issue, up from 30% in 2020 (First Orion, late 2021)3
of voluntary subscription cancellations come at the first two renewals (Recharge, 20,000+ brands, Jul 2025 to Mar 2026)4
How AI and your team improve every stage of the shopper's life with your store
A shopper calls at many points, not only after something breaks. The AI answers each call at once and does the lookups. Your agents take the moments that need judgment, with the full history in front of them.
Read down a column for one stage, or across a row to see what the AI and your team each do from first contact to win-back.
Which ecommerce calls should AI own, and which stay with your team?
Sort calls by what the answer needs. A fact in your order system, or a rule you can write down, the AI can finish. Money outside the rules, a safety issue, a fraud question or an upset customer goes to a person. No public study gives the mix by reason, so tag two weeks of your own calls first.
One damaged-item call, from first ring to refund decision.
A 6:40 pm call about a cracked lamp. The AI does the lookup, the checks and the paperwork. A person makes the one decision that costs money.
Answers and says it is an AI
It names the brand, says it is an AI assistant and that the call is recorded, then asks for the order number. Saying 'agent' at any point reaches a person.
Verifies and finds the order
Order number plus ZIP code or email on file. It pulls the order, tracking, delivery date and return policy, and reads the item back to confirm.
Takes the facts
Lamp base cracked on arrival. The AI texts a link for photos, notes the delivery date and asks whether the customer wants a replacement or a refund.
Applies your rule
Inside the window and under your auto-approve limit it can issue a label and a replacement. Over the limit, or a third claim this year, it hands off.
Warm hand-off with context
The Human Agent Desk gives the agent a one-line summary, the transcript, the photos and the caller's mood. The customer does not repeat the story.
Decision, record, follow-up
The agent approves. AI Assist proposes the disposition and notes, accepted in one click. The order note, ticket and confirmation SMS update through your integration.
Start with order status on overflow and after hours, read-only.
Start narrow: one call type, one entry point, twelve weeks, and written conditions for what comes next. Order status moves no money, so a wrong turn is cheap to catch and fix. Refunds and cancels come after.
Order status and tracking calls, on overflow and after hours
It is read-only, so the AI cannot refund or cancel by mistake. One checkable outcome: the right answer from your order and carrier data. It also clears calls that ring out at night. Returns and order changes are stage two.
Baseline and scope
- Tag two weeks of calls by reason; log unanswered, voicemail and abandoned calls by hour of day
- Connect read-only order lookup and carrier tracking; check every field the AI will read aloud
- Write the do-not-answer list: refunds, fraud holds, damaged goods, safety, legal threats, chargebacks
- Set proof of identity: order number plus ZIP or email, and no account detail without it
Staff test
- Staff place 40 scripted calls, 10 of them traps: wrong ZIP, order not found, 'delivered but not here', a furious caller
- Read every transcript; fix wrong dates, long pauses and any call that should have reached a person
- Test the warm hand-off in hours and the callback ticket after hours, with the summary showing
- Check the opening line: AI named, recording stated, a person on request
Live on overflow and after hours
- Forward only four-ring overflow and calls after 6 pm; the main queue stays with your team
- Review every transcript daily in week 5, then a sample of 30 a week
- Fix the knowledge base weekly: carrier delays, split shipments, backorder wording
- Report to the team: calls answered, resolved, handed off, hung up, callback tickets
Expand one step
- Add one call type, return start or address change before shipping, with the same audit
- Raise the share of calls sent to the AI only if the exit criteria hold two weeks running
- Write the peak-season plan: overflow thresholds, a banner message, staffing for exceptions
- Keep damaged goods and subscription cancels with the team until the audit shows they are safe to blend
Expand when all of these are true
- At least 90% of audited order-status calls match the order and carrier data (30 calls a week)
- Zero calls where the AI promised a refund, credit or exception it was not set to give
- Fewer than 5% of AI calls end in a hang-up before an answer or a hand-off
- Repeat contact on the same order within 7 days is no higher than your own baseline
Measure resolved orders and repeat contacts, not how human the voice sounds.
Write the definitions down and take the baseline before anything goes live. Public phone benchmarks for ecommerce are thin, so several baselines come from contact centers in general or from platform data. Targets are planning ranges, not promises.
Planning ranges are targets to test in a pilot, not promises. Measure your own baseline for two weeks before you change anything.
What a mid-sized online store recovers when fewer calls go unanswered.
This example is a brand with a small support team and calls that arrive after hours. Change the inputs to your own call log. The calculator counts the AI minutes, the staff time freed and the orders kept from calls that used to ring out.
A worked example. Move the sliders to match yours.
AI at $0.14 a minute with volume credit applied, plus Insights seats for the people who take handed-over calls. Phone line charges are your carrier's and are not included. Illustration, not a quote. Open the full calculator
Where these numbers come from
- Staff cost: BLS median pay for customer service representatives is $44,770 a year 7. With benefits and payroll costs added (about 1.4 times, my estimate) that is roughly $5,300 a month before seat, software and supervision.
- Missed calls: US contact centers report 8.9% abandonment 6. Brands with no night or weekend cover miss more, so the example uses 15%. Check your own log by hour.
- Capacity and AI share: 45 calls a person a day and a 45% AI finish rate are planning figures, not benchmarks. Phone agents also spend time on email, chat and wrap-up.
- Outcome: 70 is a placeholder for the margin on one order. Most missed calls are support, not sales, so only 10% count as an order kept or closed. Cart abandonment averages 70% 8, which is why a pre-purchase answer matters.
- Outbound checks can cost orders: 64% of shoppers in one vendor survey said they would probably not finish an order if asked to call to confirm it 9. Use them only where fraud loss is real.
What the dollar figure leaves out
- Your team answers fewer repeat tracking questions and spends more time on damaged goods and upset customers
- Every call is recorded, transcribed and QA-scored, so a dispute over what was promised becomes a reviewable fact
- Peak weeks need not mean voicemail: overflow calls are answered on the first ring while your team works the exceptions
Rules for ecommerce calls: what you must get right.
Not legal advice. These rules shape what the AI may say, whom it may call and how it takes payment. Federal subscription rules are unsettled as of November 2026, so build to the strictest state you sell in.
TCPA: an AI voice counts as artificial10
The FCC ruled in 2024 that AI-generated voices are artificial voices under the TCPA. Outbound AI calls to cell phones need the called party's consent, so capture it at checkout in plain words and keep the record.
Order vs marketing calls; opt-outs11
Order-service calls need prior express consent; ads such as cart recovery need written consent. Telemarketing runs 8 am to 9 pm local. Stop requests must be honored within 10 business days; FCC order 26-67 changes opt-out rules from Nov 25, 2026.
FTC Mail, Internet or Telephone Order Rule12
Covers goods ordered by phone or online. With no promised date you need a reasonable basis to ship within 30 days. If late, send a delay notice with a cancel option; refund within 7 working days, or one billing cycle on your own credit.
Click-to-cancel: vacated, being rebuilt13
The Eighth Circuit vacated the FTC's 2024 rule in July 2025. The FTC restarted rulemaking in March 2026 and, as of September 2026, had proposed no rule. ROSCA still requires simple ways to stop online charges, and the FTC enforces it.
State auto-renewal laws14
California, as amended July 1, 2025, requires cancellation by the same method used to sign up, phone included, plus express consent, a yearly reminder and consent records kept at least three years. Other states differ, so use the strictest as your standard.
PCI DSS: keep card data off the call15
PCI SSC phone-payment guidance says to keep security codes out of the phone environment and unrecoverable after authorization, even in recordings. Send a payment link by SMS or use masked keypad entry so the AI, the transcript and agents never see digits.
Four questions to ask any voice AI vendor, including us
- Do you use our call audio or transcripts to train any model, and can the contract rule that out?
- How do card numbers stay out of recordings, transcripts and logs? Show the payment-link or masked-entry flow.
- Can the AI act only inside limits we set, such as refund caps, and is every action on an order logged?
- How do you honor stop-calling requests said in plain speech, across every campaign and number?
This is general information, not legal advice. Rules differ by state and country and change often. Confirm with your own counsel before you call.
The systems the AI reads from and writes to.
VoiceGenie typically connects to these through the open API, webhooks or Zapier, plus its HubSpot, Salesforce, Zoho and Google connections. Confirm each field the AI will read or change before the pilot.
Storefront and order data
- SHShopify
- BIBigCommerce
- WOWooCommerce
- ACAdobe Commerce
- SCSalesforce Commerce Cloud
Help desk and CRM
- GOGorgias
- ZEZendesk
- FRFreshdesk
- RARe:amaze
- HUHubSpot
- SASalesforce
Shipping, returns and subscriptions
- SHShipStation
- AFAfterShip
- NANarvar
- LRLoop Returns
- HRHappy Returns
- RERecharge
Payments and fraud review
- STStripe
- ADAdyen
- SPShopify Payments
- SISignifyd
- RIRiskified
One platform for the AI, the desk and the tools behind your agents
You do not need one vendor for the dialer, another for the AI agent, a third for the agent desk and a fourth for tools that help agents. VoiceGenie is one platform. You set it up once, and every input improves both the AI and your team.
Four tools, four set-ups
own training · own reports
own training · own reports
own training · own reports
own training · own reports
One platform, set up once
- Your phone numbers and carrier
- CRM and calendar connections
- Knowledge and call rules
- Recordings, QA and analytics
Five mistakes that cost ecommerce teams money or trust.
Questions people ask before they start
Can an AI agent really handle 'where is my order' calls?
Yes, when it can read your order system and carrier tracking. It finds the order after checking an order number plus ZIP or email, reads the status and delivery window, and offers a text link. When tracking shows a loss or a delivered-but-missing parcel, it opens the claim and hands the call to your team with the details.
How much does an AI voice agent cost for an ecommerce store?
VoiceGenie's AI is $0.14 per minute, billed per second, with no minimum commitment and a free trial credit. The phone line is not included, so a short order-status call costs cents of AI time. Human agent seats are priced per person per month on annual billing: Desk $29, Insights $59 and Copilot $99.
How long does it take to set up?
As a planning range, allow about two weeks to connect order lookup and write the do-not-answer list, then two weeks of staff testing before any real caller. The pilot here runs twelve weeks in four stages. Timing depends on how easily your storefront and carrier data can be read through the open API, webhooks or Zapier.
Will customers hate talking to an AI?
Some will, so design matters. Gartner found 87% of customers say a route to a human is essential when companies use generative AI (2026). Say it is an AI in the first sentence, resolve what it can quickly, and hand over on request with context. In a late-2021 First Orion survey, 55% of US mobile users preferred a phone call for urgent order problems.
Is the FTC click-to-cancel rule in effect in 2026?
No. The Eighth Circuit vacated the 2024 rule in July 2025, and the FTC began a new rulemaking in March 2026. As of September 2026 no proposed rule had been published. ROSCA and state auto-renewal laws still apply, so let subscribers cancel by phone in the same call. Check the FTC docket for updates.
Can the AI take card payments or issue refunds?
Refunds, yes, inside limits you set, through your store integration; larger ones go to a person. Card numbers should not pass through the AI at all. Send a payment link by SMS or use masked keypad entry so digits never reach the recording or transcript. Ask any vendor to show that flow before you sign.
Can an AI call customers to confirm orders or recover carts?
Yes, with the right consent. An AI voice counts as artificial under the TCPA, so order calls need prior express consent and marketing calls such as cart recovery need written consent. Honor stop requests said in plain speech. Be careful: in one vendor survey, 64% of shoppers said they would not finish an order if asked to call.
How do we handle Black Friday and holiday peaks?
Use the AI as overflow so calls are answered on the first ring while your team works the exceptions. Launch and test it before the peak, not during it. Narvar found 37% of shoppers had a late package last year, so expect tracking and delay calls. Pre-write answers for carrier delays and split shipments.
Related guides
From the VoiceGenie blog
Hear it handle your calls before you change anything.
Talk to our AI agent in your browser, or start a trial and point one line at it. Your team keeps every call that needs them.
No credit card · No minimum commitment · Keep your carrierSources
- NRF and Happy Returns, via SDC Executive, National Retail Federation: returns expected to cost retailers nearly $850B in 2025 (2025).
- Narvar, 2026 Holiday Shopping Report (65% will use AI to shop; delivery problems) (2026).
- First Orion, After The Sale Ends: 2022 Retail Industry Survey Report (2022).
- Recharge, Subscription churn is front-loaded: the first reorder matters more than acquisition (2026).
- Macha, citing Gorgias Ecom Lab, Support tickets per order by ecommerce vertical (Gorgias Ecom Lab data) (2026).
- ContactBabel, US Contact Center Decision-Makers' Guide, 17th edition (2025).
- US Bureau of Labor Statistics, Customer Service Representatives, Occupational Outlook Handbook (2025).
- Baymard Institute, Average documented online shopping cart abandonment rate (2026).
- Digital Commerce 360, reporting a Sapio Research survey for ClearSale, 33% of US consumers drop retailers after a false decline (2020).
- Cooley LLP, on FCC Declaratory Ruling 24-17, FCC: AI-generated robocalls are illegal under the TCPA (2024).
- Cornell Legal Information Institute (eCFR text), 47 CFR 64.1200, Delivery restrictions (TCPA rules on artificial or prerecorded voice calls) (2026).
- Federal Trade Commission, Business Guide to the FTC's Mail, Internet or Telephone Order Merchandise Rule (2025).
- Pulse, The FTC's Click-to-Cancel Rule Was Vacated a Year Ago and Is Being Rebuilt From Scratch (2026).
- Barnes & Thornburg LLP, California Expands Automatic Renewal Law: New Requirements Now in Effect (2025).
- PCI Security Standards Council, Information Supplement: Protecting Telephone-Based Payment Card Data, v3.0 (2018).
- Brownstein Hyatt Farber Schreck, New TCPA Revocation of Consent Rules Take Effect on November 25 (2026).
- Gartner, Gartner Survey Finds 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent (2026).
- American Bar Association, Business Law Today, BC Tribunal Confirms Companies Remain Liable for Information Provided by AI Chatbot (2024).