Lorikeet wants your missing card to be its problem

A missing debit card presents a small philosophical problem for customer service. The customer wants another card. The support bot, having consulted its library, wants to explain how to request one. Both parties can leave the exchange having done exactly what they came to do. Only one of them still has a problem.

Lorikeet built its business around that gap. One early deployment handled lost, stolen and missing debit cards by checking eligibility, updating an address and dispatching a replacement. The interesting thing was the sequence. A satisfactory sentence came after the work, rather than standing in for it.

A card, a clinic, a rent payment

Founded in 2023 by Steve Hind and Jamie Hall, Lorikeet sells a customer concierge to businesses. It works across phone, chat, email, SMS and WhatsApp. Its customers include Airwallex, Eucalyptus, Flex and Linktree. The connecting thread is operational complexity: a request can require account information, permission checks and several actions across different systems.

Hind previously led product teams at Stripe and Watershed. Hall was a research technical lead at Google Brain, working on factual grounding in language models and contributing to Meena and LaMDA. That pairing helps explain the company’s preoccupation. Fluent conversation is useful; a system that can be trusted with the next step is the product.

The platform’s Concierge Agent retrieves customer context and carries out configured workflows. Specialist agents collaborate on complicated tasks. Coach, its operations agent, reviews conversations, helps configure workflows and runs simulations. Lorikeet connects to tools such as Zendesk, Intercom, Salesforce and HubSpot, then uses business APIs to do the actual work.

The route here was less tidy than the pitch. In a Product Talk interview, the team described months spent exploring reflection tools and information dashboards. A healthcare startup redirected them toward a plainer task: clearing its inbox. Their early prototype was a command-line script returning results in a CSV. The useful discovery came from a customer with a queue, rather than another dashboard.

The old bot failed before the new one arrived

Consider Magic Eden. The NFT marketplace had rising support demand and an AI tool whose customer satisfaction score was about 45%, compared with 78% for human agents. The difficulty was often specific: a transaction, a swap, an NFT that seemed to have disappeared. A helpful answer depended on finding out what had happened.

Through Intercom and API connections to block explorers, Lorikeet could check transactions on supported blockchains and guide users through troubleshooting. Magic Eden’s published customer story reports 74% AI satisfaction within the first month. That is a substantial improvement, still below the stated human score. The numbers describe this deployment, rather than a promise for every buyer.

Carmoola offers a different version of the same puzzle. Its knowledge-base and form-driven automation resolved roughly 30% of inbound questions. The UK car lender wanted something warmer and more capable, particularly on WhatsApp. After a proof of concept, it chose Lorikeet and named the resulting agent Katie.

Carmoola reports 40% inbound resolution on day one, rising to 60%. An outbound A/B test lifted one conversion metric by 60%. These are different measures: the first concerns completed support conversations; the second concerns a particular commercial outcome. Head of Customer Operations Lucinda Bentley says some customers now ask for Katie by name. A little personality is pleasant. Being available when somebody is choosing a car at 2am is considerably more useful.

The permission slip matters

Here is the catch with giving software something to do: it needs limits. Lorikeet describes granular permissions and gating around risky actions. Customers supply their operating procedures and decide what requires escalation. The quality of those decisions matters as much as the quality of the prose.

“If an AI gives the wrong answer, I need to know why.”Tom Pemberton, Linktree

At Linktree, workflows included account-access investigations and refunds tied to 72-hour policies. The agent escalates when customers ask for a human, after repeated failed attempts, or when frustration warrants it. A handoff can arrive with the investigation already assembled. That is a practical benefit for the person picking up the ticket.

Flex, which helps renters split payments, worked with Lorikeet on interpreting dates, clarifying vague payment questions and flagging high-risk conversations. Its rent-week chat volume is four times that of the rest of the month. The lesson is copyable: test difficult tickets, supply relevant account data, clarify ambiguous language and define the exit to a person before launch.

A finished job has a price

Lorikeet’s published Start plan costs US$2,100 a month, paid annually, plus US$0.99 per chat, email or SMS resolution. Voice resolutions are listed at US$1.50, assuming a three-minute average. Scale starts at US$5,100 monthly with lower unit rates. Larger deployments receive custom pricing.

The fee is only part of the buying decision. A company also needs usable knowledge, working integrations and people who can settle policy questions. Forward-deployed support comes with the plans, with different service levels. Resolution billing creates an incentive to finish requests, but buyers should still establish exactly which outcomes count and measure them against their own tickets.

Lorikeet competes with Sierra, Decagon, Intercom Fin and other support systems. Its case rests on complex workflows and auditable control. The customer evaluations are useful evidence, although a win at one business does not settle the choice at another. If the underlying data is unreliable, the policy unresolved or the requested action forbidden, a concierge cannot politely wish those obstacles away.

Now the customer may send an agent

In August 2025, Lorikeet announced a US$35 million Series A led by QED Investors. Its stated mission is to give every company a universal customer concierge. The September 2026 B2A launch extends that ambition to personal AI assistants acting for customers. A subsequent traffic-detection beta flags suspected agent conversations and begins with observation and review.

That adds an awkward question to the familiar refund request: who is authorized to ask? Lorikeet says agents face the same identity checks and action policies as people. The original missing-card problem survives this new channel intact. Somewhere, a customer still needs the card. The service earns its fee when the right thing happens next.