CogniAgent
Sign inGet started

Customer Service

Returns & Refunds Agent

Handles standard return requests in your website chat, using your own written rules rather than a model's judgement. It checks the customer against the order, applies the right policy for that product category and sales channel, and tells them plainly what they are entitled to — then records the return and emails your team. Refunds above a limit you set can never be promised by the agent; they go to a person.

How it works

A customer opens the chat and says they want to send something back. The agent asks for the order number, the email the order was placed with, which item and what is wrong with it. It says nothing about eligibility yet — the actor handling this part holds no tools at all, so it cannot promise a refund it may have to take back. Once the order number and email match the same order, it reads four things: the order, its individual lines, your return rules, and any return already open on that item. It then picks exactly one rule. Rules are matched most specific first — a rule for that sales channel and product category beats a channel rule, which beats a category rule, which beats your default. That is how a hygiene item, a coat and an in-store purchase behave differently without anyone editing a prompt. The return window counts from the day the item was DELIVERED, not the day it was ordered. A parcel that took a week to arrive does not eat a week of the customer's window. Then it says where they stand, in plain language: in time and here are your options, out of time and here is when it ran out, not returnable and here is the one thing that would change that, or not delivered yet. The refund figure comes from the item's own unit price multiplied by how many they are returning — read from your records, never estimated from the order total. When they choose, the return is recorded, your team is emailed with the customer's own words, and the customer gets a reference before they leave the chat. And the part that matters most: a refund above the limit you set is recorded as awaiting approval and the customer is told it has gone to your team. The agent is not trusted to respect that limit — it is enforced in code, at the only point that writes the record. In testing the model asked for a refund to be authorised and was overruled; the customer was correctly told it was awaiting approval. A refund promised and then withdrawn is far worse than one that took a day.

What's included

The key things this template does for you.

Applies your own written return rules — per product category and per sales channel — so a hygiene item, a coat and an in-store purchase behave differently without editing a prompt

Refunds above a limit you set can never be promised by the agent: the limit is enforced in code at the only point that writes the record, not by instructions

Counts the return window from the day the item was delivered, not the day it was ordered

Quotes the refund from the item's own unit price in your records — never estimated from the order total

Verifies the customer against the order before revealing anything, and never reveals which half of a wrong order-number-and-email pair was right

Prevents duplicate refunds — the same item on the same order always produces the same reference, so a repeat request updates one return instead of opening a second

Records every return and emails your team with the customer's own words, so nothing sits unseen in a spreadsheet

Builds its own Google Sheets workbook on the first run — Orders, Items, Policies and Returns, with worked examples that stay current however long after release you install it

Installation

What setting this up involves.

1

Answer the install questions: your business name and a short description, your timezone, the team that handles returns and the email their cases go to, your returns policy in plain English, and — importantly — the refund amount the agent may authorise on its own. Anything above that always goes to a person.

2

Connect Google Sheets and Gmail when prompted — two connections in total. Sheets holds your orders and rules; Gmail is only how the agent tells your team about a return. The website chat itself needs no connection.

3

Open the installed "Returns Setup" application and press Run, once. It builds a Google Sheets workbook named "<your business name> - Returns Data" with four tabs — Orders, Items, Policies and Returns — seeded with worked examples. Running it again is safe: it reuses the workbook it made.

4

Edit the Policies tab FIRST — that tab is what decides who gets a refund. POL-DEFAULT is the fallback; add a row per product category, and a channel row if in-store purchases follow different rules. A channel rule beats a category rule, which beats the default.

5

Replace the demo data. Orders is one row per order with the delivery date; Items is one row per line of that order carrying its category and unit price. The agent reads the category to pick the policy and the unit price to work out the refund, so an item missing from Items is one it cannot act on.

6

Add the chat widget to your website, then test it with one of your own orders — ideally one refund under your limit and one above it, so you can see the approval path work before a customer does.

~5 min to launch

Free to start. You will sign in first, then set it up.

Connect 2 tools, fill in a few details, and install it. No coding.

FUNCTION

Customer Service

TAGS

ReturnsRefundsEcommerceWebsite Chat

Built by CogniAgent

Official template