Lorikeet AI: The Complete Guide (2026)

This is a neutral, hands-on guide written for support and operations leaders who are actively shortlisting AI support tools. We cover the good and the awkward, and where a different approach might fit your stack better.

What is Lorikeet AI?

Lorikeet's homepage, an AI support agent for complex, high-stakes support.

Lorikeet (Lorikeet CX, at lorikeetcx.ai) is an Australian-founded startup building what it calls a "universal customer concierge" — an AI agent designed to resolve the hard tickets, not just the easy ones. It was reportedly founded (around 2023) by Steve Hind, a former Stripe product lead, and Jamie Hall, a former Google engineer with a machine-learning research background. That pairing — a payments-operations background and a machine-learning research background — is a useful lens for understanding the product: it is built by people who spent years inside a company (Stripe) where getting a customer support answer wrong could mean moving the wrong amount of someone's money. The company has leaned hard into complex, high-stakes support from day one, and originally operated under the name Optech before rebranding to Lorikeet.

The core pitch is that most AI support tools are good at Tier 1 deflection (password resets, order status, "where are my docs") but fall apart on Tier 2 and Tier 3 work — the tickets that require looking something up, making a decision, and then actually doing something across multiple systems. Lorikeet targets exactly those cases: replacing a lost card, changing a booking, processing a refund, verifying a customer's identity, disputing a transaction, or escalating to a human with full context already gathered.

The customer base tells the same story. Lorikeet reports that roughly 80% of its customers are (as of 2026) US financial institutions and fintechs, and it says it has passed security reviews at major US banks. It publicly cites regulated fintech deployments reaching around 85% automation while holding or improving CSAT — a figure worth treating as a vendor claim rather than an independent benchmark, but a directionally telling one for the verticals it targets: financial services, healthcare, insurance, gaming, and marketplaces.

Funding and trajectory

Lorikeet has raised aggressively behind this thesis. It reportedly came out of stealth in October 2024 with a seed round (variously reported at ~$5–7M USD) (backed by ex-Google and ex-Stripe operators and investors), raised roughly $9M USD more in early 2025, and in August 2025 announced a $35M USD Series A led by QED Investors — a leading fintech-focused VC — with participation reportedly including Blackbird, Square Peg, Skip Capital, Capital49, Operator Partners, Airtree, and Athletic Ventures. That reportedly brings total funding to roughly $75M USD raised inside about ten months. The fintech-heavy investor base (QED in particular) is a fair signal of where the product is strongest and where its go-to-market energy is concentrated.

How Lorikeet AI works: the Intelligent Graph

The technical differentiator Lorikeet talks about most is its Intelligent Graph architecture and a concept it calls "Pockets of Determinism."

Most LLM-based agents are fully generative: you give the model tools and context, and it decides what to do at runtime. That is flexible but hard to audit and prone to hallucination — a serious problem when the "action" is moving someone's money.

Lorikeet's design is different. It breaks automation into tightly scoped, structured sub-workflows. A natural-language workflow agent orchestrates the conversation and decides, turn by turn, which step or "node" in the graph to move to next — this is the "dispatch" behaviour: the agent reads the customer's intent and dispatches the conversation to the correct branch of the graph rather than free-associating a plan from scratch. When it reaches a point where it needs to take a sensitive action, it hands off to a bounded sub-workflow that has its own validation, its own guardrails, and its own audit log. Each node combines deterministic procedural logic with narrowly scoped LLM calls, so the model is used for the fuzzy, language-shaped parts (understanding what the customer wants, drafting a reply) while the rigid, high-stakes parts (does this account qualify for a refund? has identity been verified?) run as code with predictable outcomes.

The phrase Lorikeet uses for this is "Pockets of Determinism." The idea is that you don't have to choose between a fully generative agent (flexible but hard to trust) and a rigid decision tree (trustworthy but brittle). Instead you surround the flexible parts with deterministic checkpoints, so a hallucination can't leak into an action that moves money or changes a medical record. Lorikeet claims this yields deterministic, consistent resolution with high action accuracy, and it publishes a technical deep-dive on the architecture.

How a complex ticket actually flows. Take a fintech example — "I was charged twice for my subscription." A deflection chatbot would surface a help-center article. Lorikeet's graph instead: (1) classifies the intent and dispatches to the duplicate-charge workflow; (2) calls the billing/payments API to pull the customer's recent transactions; (3) runs a deterministic check for whether a genuine duplicate exists within the refund window; (4) if it qualifies, verifies identity through a bounded sub-workflow; (5) issues the refund via an action against the payment system; (6) writes the result and its reasoning to an audit log; and (7) only escalates to a human if any checkpoint fails or the case falls outside policy. Each of those steps is individually auditable, which is the property regulated buyers care about most.

The trade-off is that this workflow-centric model needs more upfront design than a "point it at your help center and go" chatbot. You are building and maintaining bounded flows, mapping them to your internal APIs, and defining what counts as a valid action — not just uploading docs and flipping a switch. Lorikeet mitigates this with implementation support (its higher tiers include ongoing engineering help), but the mental model is closer to "we build a supervised automation with you" than "self-serve setup in an afternoon."

Key features

• Multi-step workflow automation — the headline capability: run procedures that span multiple systems and end in a real action, not just a reply.
• Deep integrations and actions — connects to your help desk, business systems, and internal/custom APIs to actually do things on the customer's behalf.
• Omnichannel coverage — chat, email, SMS, WhatsApp, and voice from one agent.
• Auditability and guardrails — bounded sub-workflows with their own validation and audit logs, aimed at regulated industries.
• Human handoff with context — escalates to agents with the full conversation and workflow state, rather than dropping the customer.
• Compliance posture — HIPAA BAA availability on higher tiers, plus a security posture that has cleared reviews at major US banks, reflecting the healthtech and fintech focus.

Integrations

Lorikeet is built to layer onto an existing support stack rather than replace it, and its integration list reflects the regulated-enterprise buyer:

• Ticketing / help desks: Zendesk, Intercom, Front, and Kustomer — so Lorikeet can read, respond to, and resolve tickets inside the tool your agents already live in.
• CRM and telephony: Salesforce, Talkdesk, Twilio, and Amazon Connect, which is what enables the voice channel and identity/account context.
• Knowledge bases: Notion, Confluence, Google Drive, and Guru for grounding answers in your own documentation.
• Internal and custom APIs: the most important category for its target use cases — Lorikeet's actions (issue a refund, replace a card, change a booking, verify identity) run against your own systems, which is how a "resolution" becomes a real state change rather than a canned reply.

The honest caveat: because Lorikeet is a newer company, its catalogue of one-click, pre-built connectors is smaller than that of long-established incumbents. Much of the real work — and value — lives in wiring up your internal APIs, which is a build, not a checkbox.

Deployment and onboarding

Lorikeet is not a self-serve, sign-up-and-go product. Deployment is a collaborative build: you and Lorikeet map your highest-value complex workflows, connect the relevant systems and APIs, define the guardrails and what a valid resolution looks like, and then expand coverage over time. Its Scale and Enterprise tiers explicitly bundle ongoing engineering support for exactly this reason. Expect a real implementation project measured in weeks, not an afternoon — appropriate for the stakes, but worth planning for.

Lorikeet AI pricing

Refreshingly, Lorikeet publishes pricing — still uncommon in this category — and ties it to outcomes rather than seats. The model is per resolution: you pay when Lorikeet successfully resolves a ticket, with no per-seat licences and no charge for tickets it escalates to a human. Two details make this model more customer-friendly than most usage pricing:

• The customer holds the veto on what counts as a resolution. If a "resolved" ticket didn't actually solve the problem, you're not billed for it. That directly aligns the vendor's incentives with real outcomes rather than raw automation volume.
• Escalations are free. If Lorikeet correctly recognises a ticket is beyond its remit and hands it to a human with full context, you don't pay for that interaction — so the agent isn't penalised for being appropriately cautious on high-stakes cases.

How the per-resolution rate works

Lorikeet's rates are researched (based on third-party estimates and Lorikeet's own materials, which shift over time) at roughly $0.80 per resolution for chat, email, and SMS, and roughly $1.00 per resolution for voice (voice costs more to run, so it's priced higher). Plans appear to be structured as committed resolution volumes billed annually — for example, a Scale plan reportedly covering around 48,000 resolutions per year for about $48,000/year, which works out to that ~$0.80 blended rate. Smaller entry commitments and custom Enterprise agreements sit on either side of that. The main cost drivers are your resolution volume, the channel mix (voice costs more than chat), and how many of your tickets Lorikeet actually resolves versus escalates.

Treat the specific figures as a snapshot to confirm directly with Lorikeet, since plan structures in this space change often; the shape of the model (per-resolution, escalations free, customer veto) is the durable part. The economics are attractive because you pay for outcomes, not seats — but model your real ticket mix carefully. If a large share of your volume is simple, one-touch FAQ deflection, paying ~$0.80 per resolution can be more expensive than a per-action or subscription tool, because you're paying a premium rate designed for complex work on tickets that don't need it. Lorikeet's pricing rewards you when your tickets are genuinely hard; it can penalise you when they're mostly easy.

Pros and cons

Pros

• Genuinely strong at complex, action-oriented Tier 2/3 support — a real, under-served gap in the market that most "deflection" tools don't touch.
• Deterministic, auditable architecture ("Pockets of Determinism") that regulated industries can actually defend to compliance and risk teams — each step is traceable.
• Outcome-aligned pricing with two buyer-friendly features: no per-seat fees, free escalations, and a customer veto on what counts as a resolution.
• True omnichannel, including voice via Talkdesk/Twilio/Amazon Connect integrations.
• Serious security posture — has cleared reviews at major US banks and offers HIPAA BAAs — which shortens procurement for regulated buyers.
• Founders and investors with deep fintech DNA (ex-Stripe, ex-Google; QED-led funding), which shows in product depth for financial workflows.

Cons

• Built for complexity, so it can be overkill — and, at ~$0.80/resolution, over-priced — if most of your volume is simple FAQ deflection.
• Workflow-centric design means real upfront build effort: mapping flows, wiring internal APIs, defining guardrails. This is an implementation project, not self-serve setup.
• Newer company with a narrower footprint than incumbents; smaller catalogue of pre-built, one-click integrations than long-established platforms.
• Best fit is fintech/healthtech/regulated. General e-commerce, SaaS, or high-volume-but-simple support teams may find broader or lighter tools cover more ground for less.
• Because it's a standalone agent that you build workflows into, switching costs and lock-in grow with every flow you invest in — worth weighing against a layer-on-top approach.

Who Lorikeet AI is best for

Lorikeet AI is a strong fit if you are a fintech, healthtech, energy, or marketplace company with a meaningful share of high-stakes, multi-system tickets — the kind where an AI needs to verify identity, move money, or change a booking correctly and prove it did. If your support is mostly Tier 1 deflection, or you want something that layers onto your existing help desk without a workflow-engineering project, other approaches will be lighter and cheaper.

Lorikeet AI vs alternatives

No single tool wins for everyone. The market roughly splits into three camps: specialist agents built for complex, regulated work (Lorikeet), AI agent layers that sit on top of the help desk you already run (Macha), and broad standalone enterprise CX platforms (Sierra, Decagon, plus incumbents like Intercom's Fin and Salesforce Agentforce). Here's an honest comparison, including Macha.

Where Lorikeet leans into deep, deterministic workflows for regulated verticals, Macha takes a different angle: rather than being a standalone agent you build flows into, Macha runs as an AI agent layer on top of the help desk you already use — Zendesk, Freshdesk, Front, Intercom, or Gorgias. You build agents in plain English, connect custom tools to take actions, and pay per AI action rather than per resolution.

The practical difference comes down to two questions. First, your ticket mix. If most of your volume is a broad blend of Tier 1 and Tier 2 work across the whole desk, a per-action layer is usually cheaper and faster to stand up than a per-resolution specialist tuned for high-stakes complexity. If instead your defining problem is a handful of genuinely hard, money-moving, compliance-sensitive workflows, Lorikeet's specialized architecture is exactly what it was built for. Second, how much you want to change your stack. Macha is designed to add automation without ripping out or migrating off your existing support platform, and without a multi-week workflow-engineering project before you see value. Lorikeet is a more involved build, but that build is the point when auditability and determinism are non-negotiable.

Both are honest choices. For deeply regulated fintech or healthtech workflows where determinism and a defensible audit trail are the whole requirement, Lorikeet is a strong, purpose-built option. For teams that want AI agents working across their existing help desk quickly, priced by what the agent actually does, Macha is the lighter path — and for many teams, the two aren't even competing for the same tickets.

Frequently asked questions

What is Lorikeet AI? Lorikeet AI is a customer support platform that deploys an AI "concierge" capable of running complex, multi-step workflows and taking real actions across systems — designed for high-stakes tickets in regulated industries like fintech and healthtech.

How much does Lorikeet AI cost? Lorikeet uses per-resolution pricing with no per-seat fees. Researched rates (based on third-party estimates, which shift over time) are roughly $0.80 per resolution for chat, email, and SMS and about $1.00 for voice, reportedly sold as annual committed volumes (for example, ~48,000 resolutions/year for around $48,000 on its Scale plan). Escalations to a human aren't charged, and you hold the veto on what counts as a resolution. Confirm current numbers directly with Lorikeet.

Who founded Lorikeet, and is it well funded? Lorikeet was reportedly founded (around 2023) by Steve Hind (a former Stripe product lead) and Jamie Hall (a former Google engineer with a machine-learning background). It has reportedly raised roughly $75M USD in about ten months, including a $35M USD Series A led by fintech VC QED Investors in August 2025.

How is Lorikeet AI different from a chatbot? Rather than answering FAQs, Lorikeet uses its Intelligent Graph architecture — with "Pockets of Determinism" — to run bounded, auditable workflows that take real actions, like replacing a card, disputing a charge, or changing a booking, with guardrails and audit logs at each step. The generative model handles the language; deterministic code handles the high-stakes decisions.

What does Lorikeet integrate with? It connects to help desks including Zendesk, Intercom, Front, and Kustomer; CRM and telephony systems like Salesforce, Talkdesk, Twilio, and Amazon Connect; and knowledge bases such as Notion, Confluence, Google Drive, and Guru. Crucially, it also integrates with your internal and custom APIs, which is how its actions actually change state in your systems.

Is Lorikeet suitable for regulated industries like fintech and healthcare? Yes — that's its core focus. Lorikeet reports that roughly 80% of its customers (as of 2026) are US financial institutions and fintechs, that it has passed security reviews at major US banks, and that it offers HIPAA BAAs on higher tiers. Its deterministic, auditable architecture is designed specifically to satisfy compliance and risk teams.

How long does Lorikeet take to deploy? Expect a collaborative implementation project measured in weeks, not a self-serve afternoon. You map your complex workflows, connect the relevant systems and internal APIs, and define guardrails and valid resolutions together with Lorikeet; its Scale and Enterprise tiers bundle ongoing engineering support for this.

What are the best Lorikeet AI alternatives? Alternatives include Macha (an AI agent layer that runs on top of your existing help desk with per-action pricing), Sierra, Decagon, Intercom's Fin, and Salesforce Agentforce. The right choice depends on your ticket complexity, your budget model, and whether you want to keep your current help desk or replace it.

Who is Lorikeet AI best for? It's best for fintech, healthtech, insurance, gaming, and marketplace companies with a high share of complex, multi-system Tier 2/3 tickets that require real actions and strong auditability. If your volume is mostly simple Tier 1 deflection, a lighter or per-action tool will usually be cheaper.

See how an AI agent layer works on your existing help desk. Macha adds AI agents on top of Zendesk, Freshdesk, Front, Intercom, or Gorgias — built in plain English, priced per action. Explore Macha's AI agents for customer service or see pricing.