Lorikeet: Regulated AI Customer Support, Deterministic Workflows, and Pricing Explained
- 1 hour ago
- 5 min read
Lorikeet is an AI customer support platform built specifically for regulated industries — healthtech, financial services, insurance — where every automated action needs to survive a compliance review before it ever reaches a customer. Rather than positioning itself as a general support chatbot, Lorikeet combines deterministic, fixed-sequence workflows for steps that legally or contractually cannot vary with natural-language reasoning for everything else in the same conversation, and resolves multi-step tickets end-to-end across chat, email, voice, SMS, and WhatsApp rather than stopping at an informational answer.
The company reports that roughly 80% of its customer base is US financial institutions and fintechs, despite healthcare-focused search results and marketing surfacing it prominently as a healthtech tool — a detail worth knowing before assuming its healthcare-specific claims carry the same weight as its financial-services track record. For a regulated company evaluating Lorikeet, the deciding factor is whether deterministic-workflow compliance and pre-launch auditability — a privacy or compliance team signing off on agent behavior before go-live rather than after — is worth adopting a specialized platform over a general customer-support AI tool that treats compliance as an add-on.
··········
HOW DETERMINISTIC WORKFLOWS, SCOPED TOOL ACCESS, AND MULTI-AGENT DISPATCH ACTUALLY WORK.
Fixed-sequence steps combined with flexible reasoning, least-privilege integrations, and sub-agent outreach define the mechanism beyond a support chatbot.
Lorikeet's core architectural choice is running deterministic structured workflows alongside natural-language reasoning within the same interaction rather than choosing one approach for the whole system: a regulated step — verifying identity, checking eligibility, processing a refund — executes the same way every time it runs, while the surrounding conversation still handles open-ended questions with flexible reasoning. That combination is the direct answer to a common regulatory objection to AI support: a fully natural-language agent's behavior is harder to certify in advance, while a fully scripted bot can't handle the actual variety of real conversations.
Integration follows the same caution-first pattern: rather than broad API access, Lorikeet connects to systems of record — EHR, claims, billing, CRM — through least-privilege scoped tools and webhooks, each one individually approvable by a security team before launch instead of granted as a blanket permission. A "Team of Agents" capability extends this to outreach beyond the customer conversation itself: a sub-agent can be dispatched to contact a third party — a pharmacy, a provider's office — on the customer's behalf to resolve something stuck outside the company's own systems, then report back into the original conversation. Every tool call and reasoning step is logged for a replayable audit trail, and the company states contractual no-training agreements with its model providers (OpenAI, Anthropic, Google), meaning customer data isn't used to improve those providers' shared models.
........
Component | Mechanism | Function |
|---|---|---|
Deterministic + natural-language workflows | Fixed-sequence steps combined with flexible reasoning in one interaction | Keeps regulated steps certifiable while still handling open-ended conversation |
Scoped integrations | Least-privilege tools and webhooks into EHR, claims, billing, CRM | Each connection individually approvable before launch, not blanket access |
Team of Agents | Sub-agent dispatch to contact third parties on a customer's behalf | Resolves issues stuck outside the company's own systems |
Omnichannel engine | Chat, email, voice, SMS, WhatsApp, shared context, sub-1s voice latency | Avoids a customer repeating themselves when switching channels |
Audit trail | Every tool call and reasoning step logged, replayable | Lets a compliance team review exactly what the agent did, after the fact |
Compliance posture | SOC 2, BAA-ready, GDPR-aligned, PII/PHI redaction, RBAC | Built for pre-launch sign-off by a privacy or security team |
........
··········
WHY MOST DETAILED INFORMATION ABOUT LORIKEET COMES FROM LORIKEET ITSELF.
A buyer evaluating specific capability claims — especially clinical-safety behavior — currently has to rely largely on the vendor's own published comparisons.
A notable feature of researching Lorikeet in depth is that the large majority of detailed, specific content available — integration philosophy, pricing figures, feature comparisons against Decagon, Salesforce Agentforce, and Forethought — is published directly on Lorikeet's own site as content marketing, rather than appearing in independent, third-party technical reviews. That's not unusual for a newer enterprise category, and it doesn't mean the claims are inaccurate, but it does mean a specific and consequential capability — such as the stated ability to detect suicide disclosure or mental-health crisis language mid-conversation and automatically escalate to a specialist clinician — is currently a vendor-stated claim rather than one independently audited and published by a third party.
For a healthcare or financial-services buyer, that distinction should change the evaluation process rather than the interest level: capability claims this consequential need direct validation in a sandbox environment against a company's own real conversation patterns and edge cases, not acceptance based on a case-study mention or a self-published comparison table. Lorikeet's own sandbox setup is reported to take 20 to 30 minutes to stand up, which is a reasonable bar for a compliance team to test specific claims directly before relying on them in production, rather than taking any vendor's published comparison — including Lorikeet's own — as settled.
··········
HOW LORIKEET'S OUTCOME-BASED PRICING COMPARES TO OTHER REGULATED-SUPPORT PLATFORMS.
Lorikeet prices per resolved conversation rather than per seat, a structure shared by at least one direct competitor and different from Salesforce-layered or flat-annual alternatives.
Lorikeet charges roughly $0.80 to $0.95 per resolved chat, email, or SMS conversation, and $1.20 to $1.50 per resolved voice conversation, with a QA add-on around $0.25 to $0.30 per ticket, and states it does not charge for escalations that don't reach resolution. That per-outcome model ties cost directly to successful resolutions rather than to seats or raw conversation volume, which shifts the cost risk toward the vendor for conversations the AI can't actually resolve.
........
Platform | Pricing model | Reported starting price |
|---|---|---|
Lorikeet | Per resolved conversation, no charge for escalations | ~$0.80–$0.95/chat-email-SMS; ~$1.20–$1.50/voice |
Decagon | Per resolved conversation, enterprise white-glove builds | ~$0.80–$0.95/chat-email-SMS; ~$1.20–$1.50/voice (comparable range reported) |
Salesforce Agentforce | Consumption-based, layered on a Salesforce subscription | Requires existing Salesforce investment |
Forethought | Custom enterprise contract, bundles resolution, triage, and QA | ~$59,500 median annual (reported) |
........
Integration philosophy separates these platforms more than the per-resolution rate does: Lorikeet's scoped, individually-approvable connectors suit a company that wants each system connection reviewed and signed off before launch, while Decagon's white-glove enterprise build model suits a company that wants Decagon's engineers embedded during setup rather than configuring scoped permissions itself. Salesforce Agentforce only makes sense as a starting point for a company already standardized on Salesforce's data and governance model.
··········
THE DECISION RULE FOR EVALUATING LORIKEET.
Lorikeet earns serious evaluation specifically for companies in regulated industries whose hardest support tickets are multi-step, cross-system, and touch protected or sensitive data — eligibility checks, prior authorizations, claims, KYC, disputes — where a compliance team needs to approve agent behavior before launch rather than audit it after an incident. A company whose support volume is mostly simple, non-regulated questions gets little from Lorikeet's specific compliance architecture and would find a simpler, cheaper support AI adequate. Before committing budget or compliance sign-off, validate the vendor's most consequential claims — clinical-safety escalation behavior, deterministic-workflow reliability under edge cases — directly in Lorikeet's sandbox against real conversation patterns, since the bulk of detailed public information about how well those claims hold up currently comes from Lorikeet's own published comparisons rather than independent audits.
·····
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····