Gradient Labs

What Is Gradient Labs?

Most teams evaluating an AI customer-service agent start with Sierra or Decagon. Both are strong horizontal platforms, built to handle support conversations across retail, travel, and SaaS. Neither was built for a bank that needs to prove, to a regulator, exactly why an AI agent approved a refund or flagged a KYC exception.

That gap is where Gradient Labs sits. The London startup, founded in May 2023 by three early Monzo engineers — Dimitri Masin, Neal Lathia, and Danai Antoniou — builds an AI agent called Otto for financial services customer operations. Otto handles disputes, collections, KYC and KYB onboarding checks, and insurance claims intake, in addition to standard customer service across email, text, and voice.

The founding team's Monzo background matters here. They spent years building compliance-heavy banking infrastructure before starting the company, and it shows in the product. Gradient Labs states Otto runs more than 20 guardrails on every conversation turn, mapped to FCA Consumer Duty, CONC, PSD2, and Reg E/Z requirements. Production customers include Wise, Monzo, Zego, Plum, and Pockit, with Gradient Labs reporting Otto's agent serves over 32 million end users. That is a meaningfully different customer profile than a horizontal agent platform running pilots across a dozen unrelated verticals.

The company has raised a total of $26 million, after a $3.8 million seed round and a Series A that later doubled, led by Octopus Ventures and CommerzVentures with Redpoint Ventures joining. It is UK-incorporated (Gradient Labs Limited, Companies House number 14864752), which places it under the ai-assistants category as a European rather than an EU-member business.

That positioning is deliberate. General-purpose agent platforms sell breadth: dozens of pre-built connectors, templates for a hundred industries, and a promise that the same underlying model can learn any business given enough documents. Gradient Labs sells depth in one direction only. It does not compete for a retailer's returns desk or a SaaS company's onboarding flow, and the product roadmap reflects that focus rather than chasing every category a horizontal vendor might enter next.

Key Features

Procedural learning, not a generic knowledge base

Otto is trained on a company's actual operating procedures — its escalation flows, its policy documents, its edge cases — rather than a generic FAQ dump. Gradient Labs calls this procedural learning, and the practical effect is an agent that can reason through a multi-step dispute investigation the way a trained operations analyst would, instead of pattern-matching against a support article.

Financial-services guardrails

The 20-plus guardrails per turn are the product's central claim to differentiation. They cover consumer-protection frameworks specific to regulated finance: FCA Consumer Duty in the UK, CONC for consumer credit, PSD2 for payments, and Reg E/Z equivalents for US-facing fintechs. Gradient Labs also states its guardrails track EU AI Act requirements, relevant for any fintech serving customers across the continent even though the company itself is UK-based.

Back-office automation, not just front-line chat

Beyond conversational replies, Otto automates the back-office work that normally follows a customer message: dispute intake, evidence gathering, case file preparation, and KYC/KYB verification. A published case study with digital bank Pockit reports the agent resolves more than 70% of inbound queries end to end, not just the easy ones.

Voice agent for regulated call handling

Gradient Labs launched a dedicated Voice AI Agent in December 2025, built for the latency and compliance demands of phone-based financial services support. This puts it in direct competition with voice-capable rivals at Parloa and PolyAI, though Gradient Labs' pitch remains narrower and finance-specific rather than a general enterprise contact-centre platform.

Fast integration, deeper API path

For teams already running Intercom, Zendesk, or Freshdesk, Gradient Labs advertises day-one deployment with an estimated 20-50% resolution rate on simple queries out of the box. Reaching the higher 80-90% resolution figures the company cites for mature deployments requires a custom integration through its public API, which is a meaningfully bigger implementation project.

Multi-model infrastructure with financial-services controls

Otto is not tied to a single model provider. Gradient Labs orchestrates across OpenAI, Anthropic, and Google models, choosing the model per task rather than committing an entire deployment to one vendor's roadmap. That matters in a sector where model providers change pricing and capability quickly, and where a bank's risk team may want visibility into which model handled a given decision. Combined with SSO and audit logging, it gives compliance teams a paper trail that a single-model black box would not.

Pricing

Gradient Labs does not publish a rate card. Every account is priced under an outcomes-based model: customers pay per successful resolution, with no platform fee and no per-seat charge. That structure is attractive on paper — you are not paying for an agent that sits idle — but it also means procurement teams cannot get a rough number without a sales conversation.

A bank or fintech with a clear sense of its monthly contact volume can align cost directly with value delivered under this model. Gradient Labs states it cuts support costs by roughly 75% versus human agents at scale. A smaller startup unsure how its query volume will grow faces a harder budgeting problem, since there is no published floor or ceiling to plan against. That's a real contrast with a flat per-seat model such as Crisp or a simpler assistant platform like Dust.

EU Compliance & Privacy

Gradient Labs is SOC 2 Type II certified and GDPR compliant, with SSO and full audit logging available for enterprise deployments. Its guardrail stack explicitly targets FCA Consumer Duty, CONC, PSD2, Reg E/Z, and EU AI Act alignment — a compliance-first design rare among general-purpose AI agent vendors.

The company is UK-incorporated, however, which places it outside direct EU jurisdiction even though its policies and infrastructure aim at EU-equivalent standards. For an EU-based bank or insurer, that distinction matters for data-transfer assessments and vendor risk reviews, even where the underlying compliance controls look equivalent on paper. Gradient Labs has not published a public statement confirming EU-region data hosting, so buyers with strict data-residency requirements should confirm hosting location directly during procurement.

Procurement teams at EU banks and insurers should treat Gradient Labs the way they would any third-country vendor. Run the standard contractual clauses assessment, confirm sub-processor locations, and ask directly whether conversation data ever transits outside the EEA during processing. None of that rules the vendor out — plenty of EU financial institutions already work with UK-based suppliers under existing adequacy arrangements. It is still a genuinely different conversation than onboarding a French or German vendor operating natively under EU law.

Who It's Best For

Regulated fintechs and digital banks handling disputes, KYC, and collections at scale are the clearest fit — the guardrail stack solves a real compliance problem that horizontal agents do not address out of the box.

Insurers automating claims intake get a purpose-built workflow rather than a generic chatbot repurposed for a use case it was never designed around.

Teams already on Intercom, Zendesk, or Freshdesk can pilot Otto quickly through the day-one integration path before committing to deeper API work.

General e-commerce, SaaS, or retail businesses should look elsewhere. Its guardrails and case-file workflows are built around financial-services regulation, and that specialisation adds no value — and real narrowness — outside the sector.

The Verdict

Gradient Labs makes a convincing case that regulated financial services need a different kind of AI agent than the one built for a retailer's returns queue. The founding team's banking background, the production deployments at Wise and Monzo, and the guardrail-per-turn architecture are genuine differentiators against Sierra and Decagon. Real trade-offs remain too: opaque pricing, a UK rather than EU legal base, and a scope narrow enough that it is the wrong tool for almost any business outside finance and insurance. For the fintechs it targets, though, it is one of the few AI agent vendors built around compliance risk rather than added as an afterthought.

Frequently Asked Questions

Is Gradient Labs GDPR compliant?

Yes. Gradient Labs is GDPR compliant and SOC 2 Type II certified, with guardrails built for FCA Consumer Duty, CONC, PSD2, and EU AI Act requirements. The company is UK-headquartered, so it sits outside EU jurisdiction even though its compliance posture is built to EU-equivalent standards.

How much does Gradient Labs cost?

Gradient Labs does not publish pricing. It uses an outcomes-based model where customers pay only for successful query resolutions, with no platform fee or per-seat charge. Every deal requires a sales conversation to scope pricing against your resolution volume.

What is Otto?

Otto is Gradient Labs' AI agent for financial services customer operations. It handles customer service conversations, disputes, KYC/KYB checks, collections, and insurance claims intake across email, text, and voice channels.

How does Gradient Labs compare to Sierra or Decagon?

Sierra and Decagon are horizontal AI agent platforms serving many industries. Gradient Labs is built specifically for regulated financial services, with compliance guardrails for FCA, PSD2, and Reg E/Z baked into every conversation turn rather than added as a configuration option.

Which companies use Gradient Labs?

Production customers include Wise, Monzo, Zego, Plum, Yonder, Pockit, and SteadyPay. The company reports its agent serves more than 32 million end users across these deployments.