Every late scan, wrong address and return-window question lands on the phones at once. VoiceRun agents answer order status, change and return calls on the first ring, read your order system and the carrier’s scans, apply your policy, and write the result back. When a caller wants a person, the agent can pass the order and the reason with the transfer.
Built for VPs of customer care, e-commerce operations and retail technology at big-box, home-goods and direct-to-consumer retailers.
Order care · address change
Cyber Monday 11:20 AM265K-365K
seasonal hires NRF forecast for holiday 2025, down from 442,000 the year before
National Retail Federation, Holiday Forecast, 20259%
of all returns are fraudulent, so every return call is also a fraud decision
NRF and Happy Returns, 2025 Retail Returns Landscape42%
of U.S. consumers prefer the phone for telling a company about an incorrect delivery
ContactBabel via Contact Center Pipeline, 20262.7M
U.S. customer service representative jobs in 2025, projected to shrink 5% by 2035
U.S. Bureau of Labor Statistics, Occupational Outlook, 2026The calls
Most start as a question and end as an action in a system: a cancel, a ship-to edit, a return label, a booked window. Each of these runs as code against the systems your care team already uses.
Finds the order by phone number, order number or email and ZIP, then explains what “label created,” “in transit” and “delivered” mean from the carrier’s last scan and the promised date.
Systems the agent works in: OMS · Carrier tracking
Reads whether the order has been released to the warehouse. Before release it cancels or edits the order. After release it requests a carrier intercept and says plainly what might not work.
Systems the agent works in: OMS · Carrier tracking
A classic account-takeover signal, so the agent verifies by risk tier, screens the new address and edits before release. Anything your rule flags goes to a person with the case written up.
Systems the agent works in: OMS · Fraud screening
Checks the window, condition, final-sale and oversize rules, then starts the return, books a freight pickup when the item won’t fit a box, and reports where the refund stands.
Systems the agent works in: Returns platform · OMS
Books and reschedules appointment windows in the carrier’s system, which owns the truck, and confirms what a crew needs: an adult at home, the room of choice, the stairs.
Systems the agent works in: Carrier scheduling · OMS
Answers “where is the truck” from the live ETA and captures gate codes, elevator reservations, parking and placement, then writes structured driver notes back to the carrier.
Systems the agent works in: Carrier scheduling · Delivery notes
Checks the adjustment window against the current price and issues the credit inside your limits. Protection plan questions and claims go to the plan administrator with the order attached.
Systems the agent works in: OMS · Pricing · Plan administrator
Answers whether an item is in stock at a store and whether a buy-online pickup order is ready, so store teams stop answering the phone between customers.
Systems the agent works in: Store inventory · OMS
Calls ahead of the complaint about a carrier miss, a backorder or a failed delivery, with the new date and the options. Also confirms tomorrow’s delivery window.
Systems the agent works in: OMS · Carrier tracking
Why it's hard
In a demo, an order call is one question. On the floor it is an answer stitched from several systems, followed by an action inside a policy that your best reps bend by feel.
The reality
The OMS says released, the WMS says not picked, the carrier says label created, and the storefront shows a cached status. An agent that trusts the wrong one promises things you can’t deliver.
How VoiceRun handles it
Which system wins is decided per field in code: the OMS for order state and cut-off, the carrier for scan events, payments for holds. The agent reads them at call time and never answers from a cached storefront page.
Code-first logic
The OMS says released, the WMS says not picked, the carrier says label created, and the storefront shows a cached status. An agent that trusts the wrong one promises things you can’t deliver.
Which system wins is decided per field in code: the OMS for order state and cut-off, the carrier for scan events, payments for holds. The agent reads them at call time and never answers from a cached storefront page.
Code-first logic
Tested before launch
Each rule your care team follows becomes a scripted caller. Before a release reaches a customer, simulated callers place real voice calls through the long tail: the cancel that’s too late, the address change that looks wrong, the sofa that comes back on day 41.
Simulated callers place real voice calls to the agent, evaluators grade every conversation against what should have happened, and a release that regresses never reaches your callers.
$ vr simulate staging --name peak-week --wait
7/7 simulated calls passed · evaluator: judge
The platform
Start with the call type that fills your queue every peak, and grow into the whole line on one platform and one call record.
Your systems
The truth about an order is spread across several systems. The agent reads each one for what it knows best and writes changes back through the same APIs your own software uses.
Order management
Reads order state and the warehouse release cut-off, and writes cancels, holds and ship-to changes.
Common Manhattan Active Omni, IBM Sterling, Oracle Retail, Fluent Commerce
Carrier & delivery
Reads scan events and promised dates, books and moves delivery windows, and writes access instructions for the driver.
Common Narvar, parcelLab, AfterShip, Carrier APIs and portals
Returns & fraud
Checks return eligibility and risk flags, starts the return, and requests refunds through your own refund flow.
Common Loop, Happy Returns, Signifyd, Forter
Helpdesk & CRM
Where the agent looks up the customer and writes the case: what was asked, what was done, what a person needs next.
Common Zendesk, Gorgias, Salesforce Service Cloud
Telephony
Keeps your numbers and carrier, answers first or takes overflow, and hands calls to your care team.
Common Twilio, Telnyx, Bring your own carrier
Named products are examples of what’s common in each category; none of them is a prebuilt connector. VoiceRun agents call any API or database your team exposes, and every connection is code your team can read. Your carrier, numbers and contact center stay where they are.
Control
A retail agent is only as safe as its limits. Yours are rules in code: what it may cancel, credit, refund and share, and what goes to a person.
The agent never asks for a card number. A payment step goes to your existing payment system, so card numbers stay out of the conversation and the transcript.
A status question and a ship-to change after payment don’t get the same checks. Your rules set what each one needs, and anything flagged goes to a person.
Restocking waivers, goodwill credits and price adjustments are capped by rules you write. Past the cap, the agent offers a person and improvises nothing.
A caller who asks for a person gets one, and the agent can pass the order, the reason and what was already tried with the transfer. You choose the other triggers: fraud flags, damage claims, repeated misses.
In its 2024 ruling, the FCC treated AI-generated voices as artificial voices under the TCPA. Delay and delivery calls are built to dial only when your consent record allows, inside the calling hours you set, with the AI disclosure spoken from code.
When a customer says the agent promised a refund, the transcript, each system action and the evaluation sit under one call ID, with the recording too when you turn it on.
Your code remains your code, your data remains your data, and the models we train for you serve only you.
Deployment options: our SOC 2 managed cloud, your VPC, your cloud, or on-premises with VoiceRun Edge, available under enterprise agreement.
FAQ
You decide which system wins for each field, in code: your OMS for order state and the release cut-off, the carrier for scan events, payments for holds. The agent reads them at call time and checks release state before any write. Split shipments, backorders and “label created” are scenarios in the simulation suite, because that is where order status goes wrong.
The agent doesn’t take card numbers. When a call needs a payment, it hands off to your existing payment system, so card numbers stay out of the conversation and the transcript. Whether that handoff goes to a secure payment line or to a person is your decision, and our Forward Deployed Engineers scope it with your payments team.
Verification is tiered by the risk of the request, and the tiers are your rules in code. A status check needs less than a ship-to change after payment, and a change your rule flags goes to a person with the case written up. Refunds can be limited to the original tender, and the simulation suite includes callers who try to talk their way around both.
Peak breaks more than the model: provider rate limits, telephony burst limits and your own order APIs. Agents can chain speech, language and voice providers with failover, simulated callers rehearse a release before it ships, and weighted traffic splits send a slice of real calls to a new release first. Most teams also freeze releases ahead of the peak window and rehearse the overflow path with their care team.
The carrier’s scheduling system stays the system of record. The agent reads availability and writes bookings through whatever API or portal the carrier exposes, then does the part a scheduler can’t: collecting gate codes, elevator reservations and who will be home, and writing them back as structured driver notes. Where a carrier offers no write access, the agent hands the case to a person with the details already captured.
No. The Audio Control Plane sits between your phone numbers and whoever answers them, so the agent takes the first answer or the overflow and your contact center keeps the calls that need people. You keep your numbers, carrier and rates, and a transfer to your reps stays on one call record.
Bring last year’s Cyber Week call drivers and the systems an order call touches. We’ll map which calls an agent should take first, and what it takes to rehearse them before the rush.
