Lorikeet

What it is

Lorikeet is an AI customer concierge built for support work that has consequences — filing an insurance claim, reissuing a card, filling a prescription, processing a refund — rather than answering questions from a help-center article. The agent runs across chat, email, SMS, voice, and WhatsApp, executes multi-step workflows against your systems of record, escalates what it cannot close, and logs each step it took. Founded in Sydney by Steve Hind (CEO) and Jamie Hall (CTO), it launched publicly in October 2024 and raised a $35M Series A in August 2025 led by QED Investors, with Blackbird, Square Peg, Skip Capital, Airtree, and Operator Partners participating, taking total funding past $50M. The named customer list skews fintech and healthtech: Airwallex, Flex, Step, Breeze, Eucalyptus, Magic Eden, Linktree.

Two agents ship. The Concierge agent handles the customer conversation with memory that carries across channels, so someone who started in chat does not re-explain on the phone. Coach, added in January 2026, scores quality and runs root-cause analysis on every ticket instead of the 2-5% that come back through CSAT surveys. Workflows are authored in natural language with deterministic guardrails on the steps that move money or write to records.

Why it shows up in CX and support stacks

• It publishes per-resolution prices. Two tiers with rates on the page, in a segment where Decagon, Sierra, and Ada all require a sales call to learn the number. You can model the invoice before you take the meeting.
• Action-taking is the product, not a module. The pitch against deflection-first agents is that the tickets which cost the most are the ones an article cannot close. Lorikeet’s workflows call into Stripe, Shopify, Twilio, and internal APIs under scoped read or write permissions, with mock tools available so you can test a workflow before it can touch anything real.
• The meter excludes escalations. Billing fires on a successfully resolved ticket, so a conversation the agent hands to a human is not charged. That is the opposite of Ada’s per-conversation model, which bills the failures at the same rate as the wins.
• There is an official MCP server. Documented at docs.lorikeetcx.ai/mcp/mcp-server, it connects Claude, Claude Code, or ChatGPT to the account — trace a ticket’s workflow execution to find the root cause, author workflows in natural language, validate tool configurations, and run simulated customer scenarios. An ops team that already works in Claude Code can regression-test agent behavior there instead of clicking through a vendor console.
• It sits on the help desk you already run. Zendesk (including Messaging), Intercom, Salesforce, Front, HubSpot, and Help Scout on the ticketing side; Confluence, Guru, Notion, and Google Docs for knowledge; Genesys Cloud for telephony with call-forwarding for Amazon Connect, Dialpad, Five9, Vonage, and AirCall. Lorikeet Chat and an embeddable SDK cover deployment with no help desk at all.

Pricing reality

Published, which is the unusual part in this category.

• Start — $1,500/month billed annually ($18,000/year, 18,000 credits). Chat, email, and SMS resolutions draw 0.95 credits; voice draws 1.50 for calls up to 3 minutes; routing and analytics tagging 0.30 per ticket; automated QA 0.30 per ticket.
• Scale — $4,000/month billed annually ($48,000/year, 48,000 credits). Chat, email, and SMS 0.80; voice 1.20; routing and tagging 0.25; automated QA 0.25.
• Enterprise — custom rates and a custom allowance.

Neither published tier carries a per-seat charge, an implementation fee, or a platform fee. The credit allowance is denominated one-for-one with dollars — 18,000 credits for $18,000 — so a Start-tier chat resolution costs $0.95 and Scale takes it to $0.80. In volume terms, Start covers roughly 18,900 chat resolutions a year, about 1,580 a month, if you spend nothing on tagging or QA; Scale covers 60,000, about 5,000 a month.

That undercuts Intercom Fin’s $0.99 per outcome and sits well below Zendesk’s $1.50 per automated resolution on committed volume. The tradeoff is the floor: Fin charges no setup or platform fee and applies a small monthly outcome minimum instead, so at low volume Fin is cheaper no matter what the per-unit rate says. Against the enterprise pole the gap runs the other way — Decagon’s reported annual platform fee of around $50,000 lands before any usage, and Sierra’s outcome contracts run $150K-$350K+ a year.

Best for

Support and CX ops leaders at fintech, healthtech, and insurance companies whose tier-1 volume is multi-step and consequential — claims, identity verification, card and account actions, prescription flows — and who need the agent to write to systems of record under scoped permissions rather than deflect to an article. It is the right pick specifically in the 1,500-5,000 monthly resolution band, where the published rates beat what a sales-led enterprise contract would quote and the $18,000 floor is still payable.

Skip it if your ticket mix is help-center Q&A, where a deflection-first tool costs less for the same result; if monthly resolution volume sits in the low hundreds, where Fin’s per-outcome billing with no annual floor wins on arithmetic; or if procurement needs a Fortune 500 reference list and a four-figure headcount behind the contract.

Versus the alternatives

The incumbents are the help desks you already pay for. Intercom Fin and Zendesk AI have the installed base and add no new vendor to onboard; pick them when most of your tickets are answerable and the AI attached to your existing desk clears your bar. The fastest-growing entrant is Sierra — $950M raised in May 2026 at a $15.8B valuation, with over 40% of the Fortune 50 claimed as customers — and it wins when procurement wants the largest vendor in the category and will fund a six-figure outcome contract to get it. Decagon is the closest like-for-like on action-taking at enterprise volume, with a $250M Series D in January 2026 at a $4.5B valuation: pick Decagon above roughly 10,000 monthly resolutions, where a negotiated enterprise rate beats published tiers, and Lorikeet below it, where published rates and no platform fee beat a $50,000 entry. Ada is the pick when you want the invoice decoupled from how well the agent performs.

If none of them fit, the honest fallback is to keep humans on the complex queues and run a cheap deflection layer on the FAQ tier only. Buying an agent for consequential workflows you have not yet mapped produces a shorter escalation path, not a resolved ticket.

Watch-outs

• QA and tagging draw from the same credit pool as resolutions. At Start rates, automated QA costs 0.30 per ticket against 0.95 per resolution, so if the agent resolves less than about a third of your tickets, running QA across all traffic costs more than the resolutions themselves. Guard: model credits as (tickets × resolution rate × 0.95) + (tickets × 0.30 tagging) + (tickets × 0.30 QA) against last quarter’s real volume before picking a tier, and get the overage rate and the treatment of unused credits at renewal in writing — the pricing page publishes neither.
• “Successfully resolved” is a sentence on a web page, not a defined SLA. The promise that you do not pay for tickets you are unhappy with is the commercial core of the model, and as published it has no dispute window, no arbitration mechanism, and no stated criteria. Guard: get the definition, the window, and the credit mechanism into the order form, and instrument your own resolution and reopen rates from week one so any disagreement is settled with your data rather than the vendor’s dashboard.
• Both published tiers are annual-only, with no trial or pilot terms listed. Guard: negotiate a paid pilot scoped to one channel and your top 3 intents by volume, with a resolution-rate exit that fires before the annual term starts, and use the MCP server’s simulation path to replay real past tickets through the workflows before granting any write access.
• Sydney-based with roughly $50M raised, in a category where a rival raised $950M this year. On a multi-year commitment covering regulated workflows, vendor scale is a procurement question, not a snob one. Guard: ask for the current SOC 2 report and, where PHI is in scope, a signed BAA before contracting rather than accepting a trust-center badge; require data-residency terms and an export path for workflow definitions and transcripts in the agreement itself.