Train your agent the way you train a teammate

Write your procedures in plain English, the same way you'd write an SOP. Add the conditions, the lookups and the actions, and your agent follows them the same way every time.

Northwind / Overview
N
Preview as Website chatWhatsAppEmail

Refunds and returns

DRAFTPUBLISHEDRevision 3 · written here in the dashboard on Sep 19 · edited by Priya
1,600 of 12,000 characters
45 of 50 steps remaining
Written for Northwind Assistant
Refunds and returns
Draft waiting
Booking an appointm…
In use
Urgent outage reports
Set aside · from your assistant
Write another
When should the agent use this?76/500
Someone asks for their money back, or says an order arrived wrong or broken.
  1. 1

    Let them say what went wrong in their own words before offering anything. Use their name, {{contact.name}}, once — not in every line.

  2. 2

    Use @find_order to pull up the order they are talking about. Ask for the order number if they have it, otherwise the email they used.

  3. 3

    Say plainly what our returns window is and where their order sits inside it, as of {{system.local_date}}. Do not promise an outcome you have not confirmed.

  4. 4

    If a refund is the right remedy, say so and use @start_refund. If a replacement fits better, offer that first and explain why.

  5. 5

    If they are still unhappy, or the order is outside anything you can settle, use @bring_in_teammate rather than repeating yourself.

  6. 6

Add a step
Type @ to use an ability, {{ for a detail, Enter for the next step.
Write the “when” into the sentence itself — “If it’s after hours, use @create_ticket instead.”
Saved. Publishing makes this the wording the agent reads. Save draft Publish

Trusted by over 100+ businesses

Northwind
Lumen
Harbor
Atlas
Meridian
Cobalt Labs
Fernwood
Quill
Halyard
Orbit
Sable
Tidewater
Vantage
Wren & Co
Zephyr
Northwind
Lumen
Harbor
Atlas
Meridian
Cobalt Labs
4.8
  1. Greet customer and ask how to help.
  2. Ask for order number or email to locate order.
  3. Use ⚡ lookup_order to retrieve the order details.
  4. Confirm which item the customer wants to return.
  5. Check the return window eligibility.
  6. Check the customer information.
  7. Use ⚡ payment_info to retrieve payment details.

The difference between answering and doing

Most agents answer the question. A procedure gives yours the steps to finish it, the same way every time.

Complex tasks, handled end to end

A refund with eligibility rules, a plan change with proration, a claim that needs three checks. Too involved for an answer, too routine for a person.

The same way every time

Two customers with the same problem get the same process. Not a similar answer, the same steps, in the same order, with the same checks.

Change it in minutes

A policy changes on Monday. Edit the step, re-run your tests, and every conversation follows the new rule. No sprint, no ticket.

Written like language, run like logic

Every support team already has procedures, in a Google Doc or in the head of whoever has been there longest. Procedures are the same, except your agent follows them.

Write it like you'd write it for a person

Numbered steps in plain English. "Ask for the order number." "Check whether it's inside the return window." Anyone who can brief a new hire can write one.

Conditions and branching

If the order is inside the return window, do this. If it isn't, do that. Real conditional logic, written as a sentence rather than assembled from blocks.

Steps that must happen

Verify identity before showing an account. Check eligibility before promising a refund. Mark a step as required and the agent can't skip past it to reach a faster answer.

Limits at the step level

Set what the agent can decide alone, what needs approval and what always goes to a person. Different rules for a $10 refund and a $500 one.

No improvisation inside a procedure

Outside a procedure the agent reasons freely. Inside one it follows your steps. You choose which is which.

Anyone can read what it will do

A procedure isn't hidden behind a canvas or a config screen. Compliance, a new manager or a lawyer can read the steps and know exactly what the agent does before it does it.

Everything your agent can do, out of the box

Looks it up

Pull the order from Shopify or the invoice from Stripe, right where it’s needed.

Branches on data

What comes back from one step shapes the next, not what the customer said.

Pauses for approval

Where a human should sign off, it stops, asks, and continues on approval.

Start from templates

Refunds, identity checks, billing disputes, bookings. Ready to adapt to how you work.

Regression tested

Save your scenarios and re-run the whole set on every edit.

Reads the trace

Which steps ran, which conditions were met, and what each lookup returned.

Order return & refundPREVIEW

I want to return my order

I can help with that. Could you share your order number or the email on your account so I can look it up?

It’s NW-10482

Thanks. That’s the running shoes, delivered 12 April.

Just the shoes please

Done. Your prepaid label is on its way by email.

Order & returns regression9 questions · Testing as Maya ChenCompleted

Where is my order?Passed

How do I track my shipment?Passed

Can I get a refund on a sale item?Passed

How long do refunds take to process?Passed

Do you offer installment plans or financing?Passed

Do you refund shipping costs on returns?Passed

How do I cancel a subscription?Failed

Can I exchange my order?Passed

Where is my return label?Passed

Order return & refundTakeover

Show me the latest collection…

Completed 3 actions

Checked data sourcesReturn Policy

Procedure: Order return & refundShopify Get Order

Action: Generate Return LabelHere are 3 products…

Orange sneakerOrange sneaker
Pastel sneakerPastel sneaker

See it run before a customer does

Works with the tools you already use

Enhance Guzli by simplifying workflows and linking your preferred apps, with fresh integrations added regularly.

Trusted by over 100 brands.

Northwind

“Figuring out what to automate and what needs a human — Guzli has been a great partner.”

Priya Natarajan, Head of Support
Read customer story

“Figuring out what to automate and what needs a human — Guzli has been a great partner.”

Priya Natarajan, Head of Support
Read customer story
Lumen

“A genuinely capable clinical tool, deployed without an engineering team.”

Kenny Kane, CEO
Read customer story

“A genuinely capable clinical tool, deployed without an engineering team.”

Kenny Kane, CEO
Read customer story
Harbor

“Guzli really enables us to control AI without letting AI control us.”

Pierre Dubois, Data Director at Harbor
Read customer story

“Guzli really enables us to control AI without letting AI control us.”

Pierre Dubois, Data Director at Harbor
Read customer story
4.8

Guzli is committed to delivering the top customer support platform for those who demand excellence.

MomentumLeaderSPRING 2026EasiestSetupSPRING 2026HighPerformerSPRING 2026

Enterprise-grade security

We follow industry-leading compliance standards and best-in-class encryption protocols to keep your customer data safe.

GDPR badge

GDPR

Full compliance with EU data protection standards.

SOC 2 Type II badge

SOC 2 Type II

Independently audited for security and reliability.

HIPAA badge

HIPAA

Built to handle protected health information safely.

Your data stays yours

Your data is only accessible to your AI agent and not used to train models.

Data encryption

Data is encrypted at rest and in transit using standard algorithms.

Secure integrations

We use verified variables to ensure users access their own data.

Frequently asked questions

A written set of steps your agent follows for a specific situation, the same way an SOP guides a teammate. You describe when it applies, what to check, what to ask and what to do, and the agent works through it the same way every time.

No. You write the steps in plain English. Where a step needs data or a system, you attach a lookup or an action to that step, and the rest stays ordinary writing.

Each procedure says when it applies, in its own first lines. The agent matches the customer's situation against those conditions and picks the procedure that fits, so you never script which one appears where.

Yes. Steps can pull live data (an order, a booking, an account) and run actions in your connected systems (issue the refund, change the booking, update the subscription). The agent runs them at the step where they belong.

The agent handles it. Customers interrupt, change their mind and trail off; the agent answers what came up, then picks the procedure back up at the right step instead of starting over.

Yes. Run it in a preview conversation and watch it work through the steps, including the lookups and actions, so you see it run before a customer does.

Explore other features

One agent across the whole platform. The same knowledge, the same brand voice, wherever the conversation goes.

The world’s best customer experiences run on Guzli

Join thousands of teams using Guzli to build AI agents that deliver exceptional support at scale — across chat, email, and voice.