If you run support for a bank, lender, insurer, or healthtech, the question is not which AI agent has the best demo. It is which one your compliance team will sign off on, and which one fits how you actually deploy software.
Sierra and Lorikeet are both serious AI customer service platforms with real enterprise traction, and both can resolve customer issues end-to-end rather than just deflecting them. They are built around different assumptions, though, and for regulated support those assumptions matter more than feature checklists. This is a decision guide, not a ranking. It lays out the criteria a regulated buyer should weigh, how each platform tends to fit, and which scenarios point to one over the other.
The four criteria that decide most regulated deals: guardrails and pre-launch validation, audit and observability, workflow flexibility, and the deployment and pricing model.
Sierra is a strong general-purpose enterprise agent platform with outcome-based pricing and a high-touch deployment model.
Lorikeet is purpose-built for complex and regulated industries (fintech, financial services, healthcare, insurance, gaming), with defense-in-depth guardrails, deterministic plus natural-language workflows, and per-resolution pricing.
Neither is a default winner. The right choice depends on how regulated your workflows are, how much you want to own configuration in-house, and how your finance team prefers to buy.
Last updated: June 2026
Regulated support has a failure mode that generic CX does not. When an AI agent gets a return policy wrong, you issue a refund. When it mishandles a KYC unlock, a card dispute, a claims question, or a responsible-gambling flag, you can end up explaining yourself to a regulator. That is why the evaluation criteria below lead with control and provability, not with deflection rate or response time. Both Sierra and Lorikeet are credible. The goal here is to help you decide which one matches your risk profile and your operating model.
What Each Platform Is
Sierra is an enterprise AI agent company founded by Bret Taylor and Clay Bavor. It builds branded, conversational AI agents that resolve customer issues across channels, and it is best known for popularizing outcome-based pricing, where you pay when the agent fully resolves a case. Sierra serves a broad range of enterprises and has scaled quickly, with high-touch implementation support during deployment.
Lorikeet is an AI customer support platform built specifically for complex and regulated businesses. It frames its product as a concierge that resolves issues end-to-end across voice, chat, email, SMS, and WhatsApp, rather than a deflection chatbot. Around 80% of its customers are US financial institutions and fintechs, with additional traction in healthtech, insurance, and gaming. Its design center is regulated workflows and the controls that surround them.
Both resolve issues end-to-end. The difference is one of focus. Sierra optimizes for a broad enterprise market with a polished agent and a clean commercial model. Lorikeet optimizes for the specific demands of regulated industries, where the hard part is not the conversation but the controls, the auditability, and the edge cases.
At a Glance
A quick orientation before the detail. Confirm every line with each vendor directly, because positioning and capability both move quickly in this category.
Sierra · Built for: broad enterprise CX across many industries · Pricing model: outcome-based, pay on full resolution · Deployment: high-touch, vendor-led implementation · Best fit: a polished general-purpose agent where regulated journeys are part of a wider mandate.
Lorikeet · Built for: complex and regulated industries (fintech, financial services, healthtech, insurance, gaming) · Pricing model: per resolution (about $0.80 chat/email/SMS, about $1.00 voice, Coach about $0.10/ticket), customer defines a resolution, escalations not charged · Deployment: forward-deployed PM and engineer, sandbox in 20 to 30 minutes, operational in about a month, configuration you can own · Best fit: regulated teams where compliance is the dominant decision driver and the hard tickets are the point.
The Evaluation Criteria for Regulated Buyers
Most buying guides start with channels and resolution rate. For a regulated business, those are downstream of four things: whether you can constrain the agent's behavior, whether you can prove what it did, whether you can encode your real workflows, and whether the way you buy and deploy fits your team. Score both platforms on these, in this order.
1. Guardrails and Pre-Launch Validation
The first question a compliance team asks is not "how good is it?" but "how do we stop it from doing the wrong thing, and can we prove that before launch?" In a regulated business the cost of a confident wrong answer is a complaint or an examination, not a churned customer.
What to evaluate: Can you constrain the agent with explicit rules (scripted disclosures, dollar thresholds, jurisdiction-specific responses, mandatory escalation triggers)? Can you test those constraints before go-live and read a pass or fail report? Is there layered protection, or a single prompt doing the work?
How they fit: Lorikeet builds this in as its core thesis, describing a defense-in-depth model that runs adversarial simulations and red-teaming before launch, message checks on inbound, guardrails on outbound, and 100% automated post-interaction QA through its Coach agent. The framing the team uses is "the LLM is the engine, we are the cockpit." For a compliance lead who wants to approve behavior rather than approve faith, that pre-launch test-and-report loop is the differentiator. Sierra also offers guardrails and operates safely at enterprise scale, and its outcome-based model gives it a commercial incentive to be accurate. Its public positioning emphasizes the agent experience and resolution outcomes more than a pre-launch, compliance-readable validation suite. If your compliance team requires a documented test run before unsupervised resolution, ask both vendors to demonstrate exactly that step and compare what you get back.
A useful way to score this: ask each vendor to walk you through a deployment where the agent declined to act because a guardrail fired, and to show the configuration behind that block. A platform that can show you a deliberate refusal, and explain the rule that caused it, is one whose behavior you can reason about. A platform that can only show you successes is asking you to trust the happy path. For most regulated teams the layered approach (catch bad inputs, constrain outputs, then verify after the fact) is worth more than any single accuracy claim, because no single layer has to be perfect for the system to stay inside its lane.
2. Audit and Observability
When something goes wrong on a regulated ticket, you need to point at the exact step where it went wrong, not hand over a chat transcript. Regulators and internal risk teams want a replayable record.
What to evaluate: Can you replay the full reasoning and every tool call on any historical ticket? Is QA sampled or complete? Can you trace a single decision end to end months later?
How they fit: Lorikeet leans hard on this. Its Coach agent performs 100% automated QA (described as the AI evaluating the AI), with root-cause analysis, a ticket quality score, and resolution verification. Coach can also run standalone at roughly $0.10 per ticket, which means you can use the observability layer even on tickets a human handled. That complete-coverage QA, rather than sampling, is what regulated risk teams tend to want. Sierra provides analytics and reporting appropriate to enterprise deployments and is transparent about resolution outcomes given its pricing model. If your bar is 100% automated QA with replayable reasoning on every ticket, make that an explicit line item in both evaluations and compare the depth you are shown.
3. Workflow Flexibility
Regulated workflows are rarely a single answer. They are sequences: verify identity, run a risk check, update a system of record, send a compliant disclosure, escalate if blocked. Some steps must be deterministic for compliance reasons. Others benefit from the flexibility of natural language.
What to evaluate: Can you mix strict, deterministic logic with flexible natural-language handling in one interaction? Can the agent chain multiple actions across your systems and recover when one errors? How is configuration expressed, and who can own it?
How they fit: Lorikeet supports both natural-language workflows and deterministic structured workflows, and lets you combine them in a single interaction, with configuration expressed in plain English. For a regulated team this matters: you can force a scripted disclosure or a hard dollar-threshold block deterministically while letting the rest of the conversation flex. It also dispatches sub-agents (its Team of Agents) to call third parties, for example contacting a merchant on a dispute or a pharmacy on a prescription question. Sierra is a capable agent platform that handles complex, multi-turn conversations and integrates with enterprise systems to take action. Buyers who want a high degree of deterministic control over specific steps, expressed and owned in-house, should test both against one of their genuinely hard workflows rather than a demo script.
4. Deployment Model and Pricing
How you buy and deploy shapes the relationship for years. This is where the two platforms differ most cleanly, and where preference rather than capability often decides.
What to evaluate: Outcome-based versus per-resolution pricing, who defines a resolution, whether escalations are billed, how much configuration your team owns versus the vendor, and how long to first production tickets.
How they fit: Sierra pioneered outcome-based pricing, where you pay only when the agent fully resolves a case and escalations to humans cost nothing. The appeal is incentive alignment. One honest caveat worth raising in any regulated evaluation: a model paid only on full resolution can favor the tickets that resolve cleanly, and in regulated support the hard tickets (KYC, disputes, claims) are the ones that matter most, so confirm how the vendor defines and counts a resolution. Lorikeet prices per resolution at roughly $0.80 for chat, email, or SMS and roughly $1.00 for voice, with Coach around $0.10 per ticket, escalations not charged, and the customer holding veto over what counts as a resolution. Its Scale plan is 48,000 resolutions for $48,000 a year. For ROI context, a human-handled ticket commonly runs about $1.25 to $4. Lorikeet deploys with a forward-deployed PM and engineer, a sandbox in 20 to 30 minutes, and operational status in about a month, with configuration your team can own afterward. Both are reasonable models. Pick the one your finance team and your operating team prefer to live with.
Recommendation by Scenario
No single answer fits every regulated team. The scenarios below map common situations to a starting point. Treat them as a shortlist signal, then validate with a real workflow.
Choose Lorikeet if
Your toughest stakeholder is your compliance or risk lead, and you need to prove the agent's behavior before go-live, not after.
Your hardest tickets are regulated workflows (KYC unlocks, card disputes, transfers, claims, responsible-gambling flags) where correctness on the edge cases is the whole point.
You want 100% automated QA with replayable reasoning on every ticket, not sampled review.
You need to mix deterministic, compliance-mandated steps with natural-language handling in one interaction, and want to own that configuration in-house.
You prefer per-resolution pricing where you define what counts as resolved and escalations are not billed.
Choose Sierra if
You are buying for a broad enterprise CX mandate that spans regulated and non-regulated journeys, and a polished general-purpose agent is the priority.
Your finance team strongly prefers pure outcome-based billing and the incentive alignment that comes with paying only on full resolution.
You value a high-touch deployment with an established enterprise vendor and a strong brand-led procurement story.
Your regulated workflows are a minority of total volume and your compliance requirements, while real, are not the dominant decision driver.
Either could work, so run a bake-off, if
You have a mix of regulated and general tickets and no single criterion dominates.
Both clear your security bar (both should be able to speak to SOC 2 and enterprise security review; confirm specifics like HIPAA readiness and data residency for your jurisdiction).
The decision comes down to deployment fit and commercial preference rather than capability.
How to Run a Fair Comparison
Demos are built to look good. The way to separate the two is to make both run your hardest reality, not their happy path.
Bring your 10 hardest regulated tickets and ask each vendor to resolve them in a sandbox against your guardrails.
Ask to see a full audit trail for a real decision, end to end, with every tool call and the reasoning between them.
Ask whether your compliance team can run the guardrail or simulation suite before go-live and read the pass or fail report.
Ask exactly how a resolution is defined and counted, and who holds the veto.
Ask what happens when a downstream system returns an error mid-workflow: retry, escalate, or roll back.
Confirm security and compliance specifics in writing (SOC 2 scope, HIPAA or BAA readiness if relevant, data residency).
Common Mistakes Regulated Buyers Make
Most regretted decisions in this category trace back to a handful of evaluation errors. Avoiding them matters more than picking the right vendor, because either platform can disappoint a team that bought it for the wrong reason.
Leading with deflection rate. A high resolution percentage can be earned on easy tickets while the regulated edge cases quietly fail. In a regulated business, correctness on the hard tickets is the number that protects you.
Treating a transcript as an audit trail. A conversation log is not the same as a replayable record of every tool call and reasoning step. Ask to replay a real decision, not read a chat.
Assuming voice is the same agent as chat. If voice runs on a separate stack and is bolted on with a handoff, customers repeat themselves and context is lost. Confirm shared context and shared workflow logic.
Letting pricing model stand in for total cost. Outcome-based and per-resolution pricing each look clean on a slide. The real comparison is total cost on your actual ticket mix, and who gets to define a resolution.
Skipping the compliance team until late. The fastest way to stall a deal is to involve risk and compliance only after a favorite has been chosen. Bring them into the guardrails and audit evaluation from the start.
Lorikeet's Take
We will be direct about our own bias: Lorikeet is built for the regulated case, so when compliance is the dominant decision driver, that is where we are strongest. Sierra is a genuinely good platform, and for a broad enterprise mandate with outcome-based billing as a priority, it deserves a serious look. The honest framing is that this is a fit decision, not a quality ranking. If your compliance team is the toughest stakeholder in the room and your hardest tickets are the regulated ones, weigh the guardrails, audit, and workflow criteria above heavily, and make both vendors prove them on your real workflows. If you want to see how that pre-launch validation and 100% QA loop works in practice, talk to Lorikeet and bring your hardest tickets.
Key Takeaways
For regulated support, evaluate in this order: guardrails and pre-launch validation, audit and observability, workflow flexibility, then deployment and pricing.
Lorikeet is purpose-built for regulated industries, with defense-in-depth guardrails, deterministic plus natural-language workflows, 100% automated QA, and per-resolution pricing where the customer defines a resolution.
Sierra is a strong general-purpose enterprise agent with outcome-based pricing and a high-touch deployment, well suited to broad CX mandates.
Neither is a universal winner. Match the platform to how regulated your workflows are, how much you want to own configuration, and how your team prefers to buy.
Decide it with a bake-off on your 10 hardest regulated tickets, not a demo script.
