Industries  /  Retail & e-commerce

Retail & e-commerce

The same question,
several thousand times.

Where is my order. Can I return this. Do you have it in my size. High volume, low complexity, arriving on whichever channel the customer happens to prefer — and every one of them answered by a person today is margin.

Why retail support is different

The contact profile is unlike anything else on this site, and it changes what good looks like.

01

Volume spikes are brutal

A promotion, a stock-out, a delivery failure or a viral post can multiply contact volume in an afternoon. You cannot hire for a peak that arrives without notice, and queueing through it costs you the customers you just paid to acquire.

02

Customers arrive everywhere

Instagram DMs, WhatsApp, web chat, email, and the phone — often the same person across two of them. Treating each as a separate inbox is how a customer gets two different answers.

03

Most contacts are lookups

Order status, returns policy, stock, delivery times. These are database questions dressed up as conversations, and they are the ones that consume a support team's day.

What the AI handles

Not deflection — resolution. A customer who gets an accurate answer in ten seconds is better served than one who waits nine minutes for the same sentence.

  • "Where is my order?" Looked up in your system through a permissioned connection and read back with the tracking detail — not "please check your email".
  • "Can I return this?" Your actual policy, including the exceptions, from your own content.
  • "Do you have it in stock?" Live, if your catalogue exposes an API.
  • "When will it arrive?" Your delivery windows, by destination.
  • "I want to change my address." Taken and written back where your system allows it, with confirmation and an idempotency key so it cannot be applied twice.
  • "Cancel my order." Handled if it is still cancellable, escalated if it is not.
What goes to a person
  • A complaint, or anyone whose tone turns — detected and routed on the first signal.
  • A damaged or wrong item where a judgement call about goodwill is needed.
  • Anything involving a refund above a threshold you set.
  • A high-value or repeat customer, if you want them always to get a person — that is a routing rule, not a hope.

Your agents stop spending the day on order lookups and spend it on the conversations where a person actually changes the outcome.

One customer, one thread, every channel

The part most tools get wrong, and the part customers notice immediately.

Instagram & Messenger

DMs land in the same queue as everything else rather than in a phone somebody checks between tasks.

WhatsApp

A first-class channel, with conversations that resume days later on the same thread.

Web chat & SMS

Free per message, so a chat-heavy month does not produce a surprise bill.

Email

Tickets with a sticky owner and a response target, not a shared inbox two people reply from.

Agents carry separate capacity per channel — one voice call and several chats at once — so a phone conversation is never interrupted by a message. How the channels work together

The economics

Retail is the one industry here where the argument is usually cost per contact rather than revenue recovered.

A person answering "where is my order" costs several dollars. The AI costs cents.

Web chat and email are free per message here — you are never charged per contact on the digital channels. AI text handling bills against tokens at $5 per million, which for a short order-status exchange is a fraction of a cent. Voice is 28¢ a minute.

And because there is no per-seat fee, seasonal cover costs nothing on the software side. A team that doubles for Black Friday pays the same platform fee in November as in June.

Seasonal team, same fee
MonthPeoplePlatform
June8$69
November18$199
December24$199
January10$69
Plan changes with the seat cap, not with headcount. On a per-seat platform each of those rows is a different invoice.

What retail teams ask us

Does it connect to Shopify, WooCommerce or our own platform?
The AI reaches your systems through MCP — a permissioned connection where every call is schema-validated, classified read or write, and logged, with writes requiring confirmation and carrying an idempotency key. If your platform has an API, it can be wired. Tell us what you run and we will be straight about the fit.
What stops it inventing a delivery date?
It answers from your content and from lookups, not from inference. If it cannot find the answer it escalates rather than guessing — and you can read exactly what it said, because every conversation is transcribed and graded against your QA form.
How does it handle a spike?
The AI has no queue of its own — concurrency is a platform matter, not a staffing one. The measured figures are on the security page: 1,600 simultaneous voice calls and 1,000 simultaneous chats on the production box.
Can we keep our current phone number and helpdesk?
Your numbers, yes — bring your own carrier or take numbers from us. If you want to run this alongside an existing helpdesk rather than replace it, start with one channel and keep the rest where they are.
Start with one channel

Put the AI on chat and watch the queue drop.

Chat is free per message and the lowest-risk place to start. If it works there, move the phone.

Google or Microsoft · 30-day free trial · $15 credit · no card required