Skip to content
replytor

Support agent for retail and e-commerce

Answers from your actual stock.

Your customers ask about availability, shipping and returns. replytor answers those questions from your own data, names the source behind every answer — and hands over to your team the moment it is unsure.

Book a conversationHow the agent works

Example from the email channel
Customer

Hello, is item 4051-B still available, and how long does shipping to Austria take?

replytor

Hello, item 4051-B is available — seven in stock right now. Shipping to Austria takes three to five working days and costs €9.90. Do get in touch if you have questions about fitting it.

Sources stock:4051-B · 14:02 shipping:AT · listing 4051-B

Four questions fill your inbox

In retail it is nearly always the same four. And none of them can be answered from an FAQ text — only from systems.

“Is it still available?”

The answer lives in today's stock, not in last week's listing copy.

“Where is my delivery?”

The answer lives in the order, together with tracking number and dispatch date.

“Will it fit my device?”

The answer lives in the item description, the data sheet and your resolved cases.

“Can I send it back?”

The answer lives in your return terms — and those differ by sales channel.

Why FAQ bots fail in retail

A conventional chatbot learns text. But text cannot answer the very questions that fill the inbox. A learned stock level is yesterday's number, wrong today — and it still reaches the customer in a friendly, convincing sentence.

So replytor draws a hard line between knowledge and facts as of now. Manuals, policies and resolved cases are indexed. Stock, orders and ticket status are queried at the moment of answering and never learned.

That is not a setting, it is how the thing is built. A number the agent quotes is as old as the answer — not as old as the last import.

What happens on every request

  1. The message arrives

    From email, chat, shop or ticket system. A duplicate delivery never produces a second answer.

  2. Rules run before the model

    Cancellation, legal threats, refunds, complaints: these never reach a language model. They go straight to your team.

  3. Sources are consulted

    Documents and resolved cases searched, stock and orders queried live — always scoped to your account.

  4. The draft is checked

    Every statement must point to a source. If none is found, no draft is written at all.

  5. Send or hand over

    Cited and confident: the answer goes out. Unsure: your team gets the case with the reason.

What replytor will not do

Not out of caution — there simply is no tool for it inside the agent. These cases are handover reasons.

Never moves money

No refunds, no credit notes, no cancellations.

Never promises a price

No discounts, no goodwill, no exceptions.

Never changes contracts

Cancellation and withdrawal always reach a human.

Never promises a date

A backlog item is not a delivery date.

Watch first, send later

Every new customer starts in observation mode. The agent reads along, writes its drafts as internal notes and sends nothing. You see, on your own inbox, what it would have answered, what it would have cited and where it would have handed over.

Only when those numbers hold up does one channel go live, then the next. And even then the stop switch stays: one switch per channel that halts sending immediately.

See it on your own inbox

Observation mode measures against your real requests without a single message reaching your customers.

Book a conversation