×

Lorikeet Overview for Paddle

Lorikeet is an AI customer support agent for complex, regulated businesses. This deck covers how it applies to Paddle's buyer support operation.

Prepared for Mosthafa · July 2026
Alex Holder, Lorikeet

Where Lorikeet fits

Lorikeet resolves the buyer tickets your NLU bot escalates: refunds, cancellations, and transaction lookups. It is charged per resolved ticket.

Handles the escalated tail

Refund eligibility checks, subscription disambiguation, transaction lookups, and multi-transaction accounts. This is the 15,000 to 20,000 tickets a month that bypass the NLU bot or bounce out of it into the human queue.

Identifies the right transaction

It reads the product and details the buyer names and matches them against the buyer's purchase history, rather than defaulting to the most recent charge.

Charged per resolved ticket

Lorikeet is charged only when it closes a ticket. There is no per-message fee and no per-seat licence.

Maintained in plain English

Workflows are written in plain English and edited like a document. When a seller onboards or a checkout flow changes, you edit the text and simulate it against historical tickets before publishing.

State of play at Paddle

250,000 to 280,000 tickets a month across buyer and seller support. About 90% are deflected by the NLU bot. A 23-person buyer-support team handles the rest. Volume varies and reached about 350,000 earlier this year.
80 to 85% of buyer tickets are refunds and cancellations. The rest are invoice requests, transaction lookups, and queries about unrecognised charges. The difficulty is finding and identifying the correct transaction, not classifying intent.
15,000 to 20,000 tickets a month reach the human team as escalations, reopens, or messages that bypassed the bot. During volume spikes the queue falls weeks behind.

Where Lorikeet sits:

Between the NLU bot and the human team, resolving escalations, reopens, and bypassed tickets. It reads the conversation from Zendesk.
Transaction search that matches the product a buyer names against their purchase history, rather than assuming the most recent charge.
Failover: it can run as the primary resolver, or as a fallback if the NLU layer is unavailable.

Demo use cases

Two buyer-support use cases, built as a working sandbox with synthetic data, live workflows, and an agent you can inspect and edit.

01

Refund request, end to end

An authenticated buyer requests a refund. The agent looks up their transactions, identifies the correct one, checks it against the refund window, processes the refund to the original payment method, and confirms with a reference number.

Refunds and cancellations are 80 to 85% of buyer tickets.

02

Cancellation with transaction disambiguation

A buyer with several subscriptions across different sellers says "cancel my video subscription." The agent matches the product named to the correct subscription rather than the most recent one, confirms, then cancels.

Tests whether the agent identifies the right transaction from limited detail.

Appendix

Lorikeet overview, the team, investors, customer voice, architecture, POC plan, security, and pricing.

Lorikeet builds an AI support agent for complex businesses

Lorikeet resolves customer issues by understanding the request and taking actions across the systems that hold the answer. It works across chat, email, voice, and SMS, 24/7.

Re: Refund for my PixelPress purchase

Paddle

To: Jordan Rivera

Hi Jordan, I've found your PixelPress Photo purchase from 18 July ($29.00). It's within the refund window, so I've refunded it.

The $29.00 will land back on your original card in 5 to 10 business days. Your refund reference is RF-40021. Anything else I can help with?

Email

Lorikeet

0:08

"Hi Jordan, this is Paddle support, confirming your Lumafy Video Studio subscription is cancelled and won't renew. You'll keep access until August 2nd. Nothing else is needed on your end."

Voice

There's a charge from "PADDLE.NET" on my card I don't recognise. Is this legit?

L

Paddle is the payment provider for lots of software purchases, which is why it shows as Paddle on your statement. Let me find the exact charge for you.

Thanks

L

Found it. $19.00 to Lumafy Video Studio on 2 June, your video editing subscription. Want a receipt, or shall I help you cancel it?

Chat

Our people have decades of experience with AI and building great products

CEO and Co-founder

Steve led product teams at Stripe and Watershed, building tooling to enable complex processes like carbon accounting and financial reporting at scale.

CTO and Co-founder

Jamie was a research tech lead at Google Brain, leading research on factual grounding in large language models. Third named author on Google's breakthrough 2022 LaMDA paper, and fourth named author on the 2020 predecessor Meena paper.

Our team comes from leading companies applying AI and driving large-scale corporate transformation

Lorikeet is backed by top global investors

We've raised over $50M from investors who backed Nubank, Klarna, Canva, Airwallex, Supabase, Midas, Clearscore and more.

Announcing our

$35M

Series A

Led by QED Investors

Blackbird

Square Peg

Capital49

Skip Capital

Airtree

Operator Partners

www.lorikeetcx.ai

Customer feedback from head-to-head evaluations

"We tested AI solutions head-to-head and Lorikeet was a winner in every metric."

Lindsay Boland

Product Manager, Flex

"We ran POCs with Fin AI, Decagon... Lorikeet was a clear winner."

Jiaona Zhang

Former CPO, Linktree

"Considered Sierra and other well-known players... [Lorikeet] could handle nuance, de-escalate emotional moments, and follow SOPs without breaking a sweat."

Jessica Mishlove

Head of Customer Relations, Arbor

Read case studies

Journey of a Support Ticket

How Lorikeet's AI agent resolves customer issues end-to-end

1

Customer sends message

Via chat widget, email, WhatsApp, voice, or ticketing system

2

Channel Intake

Mode (chat / email / voice / SMS) · Escalation rules · Auto-close config

3

Brand + Config Loaded

Paddle voice · Buyer vs seller context · Language detection

4

Inbound Guardrail

Evaluate message → pass to triage, or escalate immediately

5

Intent Classification

AI matches customer intent against all available workflows

Matched Workflow

Run response SOP

FAQ Fallback

Search knowledge base

No Match

Escalate to human

6

Workflow Executes

Structured decision graph or Natural Language agent follows step-by-step SOP

Tools

Zendesk · Transaction lookup · Refund & cancellation APIs

Knowledge

paddle.net help · SOPs · Seller policies

Actions

Tags · Internal notes · Human escalation

7

Outbound Guardrail

Check AI response against policies → pass or escalate

8

Reply Sent to Customer

Brand voice · Formatting · Citations · Side effects (tags, CSAT, Slack)

If customer replies → loops back to step 5 with full context

If no reply → auto-resolved after idle period

Proof of concept

2 to 4 weeks, offline, free. Confirms Lorikeet executes against real Paddle buyer-support tickets
We mock data to simulate production situations, rather than integrate directly
No obligation to proceed if it does not meet your needs
Live pilot in your Zendesk workspace next, taking an agent seat to handle real tickets in production. You can scope a single workflow (for example, refunds) to start and expand from there

Train and test AI

Integration scoping

Legal / privacy / security review

Week 0 to 1

We will:

Train Lorikeet's tone and brand from your guidelines
Run response testing over 20 to 50 example tickets
Plan training based on your SOPs

We need:

Tone of voice, SOP, ticket corpus
Fast feedback on tickets

Week 1 to 2

We will:

Build, test and iterate on 1 to 2 agreed use cases
Share learning materials with your team
Produce documentation to aid implementation

We need:

Continued feedback on AI response quality

Week 2 to 3

We will:

Run Lorikeet 101 training session with your champion
Complete any necessary security and legal reviews
Agree on terms of contract including cost

We need:

Tech lead to align on integration plan

Week 3 to 4

We will:

Complete security and legal reviews
Kick off integration sync

We need:

Agree on commercial terms and sign

Lorikeet is built securely for enterprise

Compliance-grade architecture, day one. Designed for fintechs, marketplaces, and other regulated businesses handling sensitive payment and account data.

SOC 2 Type II
& ISO 27001

Both certifications independently audited. Your data is always protected and is never used for training.

GCP + VPC

Built on Google Cloud, running in a virtual private cloud. Audit-grade logging end-to-end.

Ex-Stripe security

Founding engineering team includes an ex-Stripe Staff Security Engineer. Security reviewed by industry leaders.

Handles sensitive PII at scale

Already deployed with FCA-regulated fintech and marketplace subscribers handling buyer PII, payment methods, and transaction data at international scale.

For Paddle's data posture: as merchant of record you hold buyer PII, payment methods, and transaction records across many jurisdictions (EU GDPR, UK GDPR, US state rules). We can walk your security and legal team through our controls during the pilot: data residency, access controls, and incident response.

View trust center →

You only pay for resolutions

You are charged per resolved ticket.

Pay only for resolved tickets

Lorikeet charges per ticket successfully resolved, as determined by you. Tickets it does not resolve are not charged.

No platform or per-seat fees

No setup costs, no per-seat licenses, no monthly minimums. The only line item is resolutions.

Volume discounts

Per-resolution price decreases as volume grows. Available across email, chat, voice, and SMS.

Aligned incentive structure

You buy credits from Lorikeet up front
Credits are only consumed when the AI reaches the desired outcome
You don't pay if the AI fails and the case is passed to a human
Credits sized to your forecast volume, buy top-up packs at a discount mid-contract, no surprise overages

At Paddle's volume: the 15,000 to 20,000 buyer tickets a month routed to Lorikeet fall in the Enterprise / Custom tier. Pricing is set per resolved ticket and sized to your workflow mix and forecast volume during scoping.

Alex Holder

alex.holder@lorikeetcx.ai