Customer service

When a refund goes wrong, can we show who decided and why?

Agents now answer tickets and settle refunds. LeafCraft lets them act alone where they have earned it, and sends large or unusual cases to the person you name.

One teamYour rulesYour memoryYour budget

Book a demo Experience now

LeafCraftto Service manager, Zone 3 · 10:42

Refund above the agent's limit

ItemCase CS-9981 · 7,400 · damaged on deliveryWhy youAbove the agent's limitAgent saysRefund in full. Photos match the delivery report.WaitingMoves to Head of customer experience, in 2 hours

Your reason (required): "Repeat customer. Courier claim raised."

ApproveSend backReject

Your reason is saved in the record, next to the agent's reason.

Decision request in Microsoft Teams · example dataLeafCraft console

What's changing in customer service

Agents refund, promise and close

Service agents now act on money and on commitments to customers.

A wrong refund is a fraud vector

An agent that refunds with no checker will be found by someone who wants refunds.

Customers and regulators ask why

A complaint about a decision needs the decision, the rule and the person, not a transcript.

Two risks the layer removes

The agents you can't see, and the model that gets retired.

Agents you don't know about

In customer service, agents arrive through vendor updates and team pilots. LeafCraft finds all of them, gives each an owner, and brings it under your rules or removes it.

Inside software you buySwitched on by vendor updates

CRMERPHR systemService deskOffice suite

Built by your teamsPilots, scripts and side projects

LangGraphCrewAINotebooksAutomationsCopilot studios

LeafCraft connects to your ecosystem and finds every agent

One registerexample counts

23found

9no owner yet

11governed

3retired

How discovery works

A model that gets retired

An agent built straight on one model has to be rebuilt and re-approved when that model is retired or repriced. On LeafCraft you switch the model for that agent in config and keep everything else. Months and a budget, saved.

Agent tied to one model

The model is retired or repriced
Prompts, tests and sign-offs rebuilt
Months of work, a new approval cycle
The history stays with the old tool

MonthsA rebuild budget

Agent on LeafCraft

Pick another model for that agent
Switching is a config change
Test it against your own record first
Rules, limits, owner and memory stay

HoursNo rebuild

TimeHours to switch, not months to rebuild

CostNo rebuild budget, no new approval cycle

MemoryThe record carries over to the new model

Your choice of model

Why it matters now

In customer service, the agents arrived before the rules did.

Customer service is where agents meet customers directly. The layer sets what each agent may promise, refund or change alone, sends the rest to a named person, checks every closed case with someone other than the agent that handled it, and keeps the record a complaint or a regulator will ask for.

How the layer works

Helpdesk and CRM agents

Chat and voice bots

Refund tools

Team lead laptops

No shared owner · no shared limit · no shared record

LeafCraftone layer, on your side

One ownerper decision, AI or person

Your ruleson every run

Your memoryboth reasons, your database

Your budgetset before work runs

The core infrastructure

What LeafCraft runs for customer service.

One team

Agents resolve within their limit. People decide the exceptions and the refunds above it. A separate check on every closed case.

How it works 02

Your rules

What each agent may promise, refund or change alone, and what goes to a named person. Silence is never a yes.

How it works 03

Your memory

Every case with every step and both reasons, so a complaint, a regulator or a customer can be answered from the record.

How it works 04

Your budget

Cost per case and per agent known before it runs, with limits that stop runaway loops instead of billing for them.

How it works

AI and people, one team

Your customer service team, with AI on the org chart.

Does the work

Decides

Checks

AI drafts. A person decides. Another agent checks.

The agent prepares the case. Your person makes the call and writes why. A separate audit agent checks the decision against your rules.

Same rules for every seatNo one checks their own workOne named ownerCost set before it runsBoth reasons recorded

No one checks their own workAgent or person. Maker and checker are never the same.

One named owner per decisionNever "the system".

Authority earned from the recordWidens with clean work. Narrows after mistakes.

People stay where they workSlack, Teams, WhatsApp (coming) or email. Their reply moves the work on.

What changes for you

Today

With LeafCraft

Refunds approved by the agent that proposed them

A separate agent or person checks every refund

Promises made in chat with no record

Every commitment has an owner and a reason on record

Escalations lost in queues

By impact, to named people, with a time limit

Cost per case unknown

Estimated before it runs, capped per agent

Silence is never a yesEstimated before it runsNothing erased after it's written

Authority · Refund agentEarned one task at a time, from the record

Level 2 of 4

Checked tasks with no override, last 12 weeksTrack record improving

AdvisesA person approves every refund

Done

Refunds alone up to a set amountAbove it, recommends and a person decides

Current

Higher limitUnlocks when the record meets your bar

Locked

Mistakes in the record narrow the limit again. A vendor update never widens it.

Product screen · example dataLeafCraft console

How it helps your people

Who it helps around customer service.

Industries where it runs

Consumer durablesWho checks the agent that approves warranty claims and dealer payouts?Read Electronics and IoTCan an agent change a setting on a customer's device without asking?Read B2B software companiesCustomers ask how our agents are governed. Can we show them?Read Banking, NBFCs and insuranceCan we show the regulator who did what?Read Advertising and mediaCan we trust the morning numbers?Read Logistics and supply chainWhat happens when an agent gets it wrong at 2am?Read

Leaders who ask first

COOs and supply chain headsWhat happens when an agent changes a PO or a supplier order at 2am?Read CFOs and finance leadersWhat does each AI decision cost, and can I show my auditor who approved it?Read Risk, compliance and auditWhen I ask who approved this, can anyone show me the full trail?Read Chief AI Officers and AI headsEvery new agent needs its own guardrails and tests. How do we scale past the first few?Read

Questions teams ask

Can our support team stay in their tools?

Yes. They get requests on Slack, Microsoft Teams, WhatsApp (coming soon), email or the built-in inbox, and their reply moves the work on.

Can we keep customer text out of traces?

Yes. One setting masks message content in traces. Timings, costs and agent names stay visible.

Does the agent learn from corrections?

Yes. When someone corrects an agent, it keeps the lesson and uses it the next time similar work comes up.

Bring one customer service process.

We'll map who does the work, who decides and who checks, set the cost limits and show you the record.

Book a demo Experience now