Solutions · Financial Services
Sonzai Labs builds and operates AI Employees for banks, wealth managers, and insurers — for lead qualification, advisor enablement, customer support, and compliance triage. Forward-deployed engineers from the team behind $70M+ AI value at DBS Bank and OCBC's first gen-AI rollout. Production-grade in 4–6 weeks; gets sharper every month.
What we build
Score and route inbound leads across digital channels in seconds. Pull product fit from your catalog, KYC posture from internal systems, and historical signals from your CRM. Hand a fully-briefed lead to the right RM or sales team — already enriched.
An AI Employee that sits behind every RM, not in front of every customer. Pre-meeting briefs, portfolio commentary, regulatory-safe answers, and follow-up drafting — built on the customer context your RMs already have access to.
Before an agent picks up, the AI Employee has already pulled every relevant case, identified the product, surfaced the regulatory edge cases, and drafted three response paths. Faster handle time, fewer escalations, full audit trail of why each decision was made.
Inbound documents — KYC packs, dispute letters, regulatory queries — get parsed, classified, and routed with a structured summary. Humans approve every decision; the AI Employee just handles the reading and the first draft.
Reconciliation exceptions, payment investigations, internal queue triage. The AI Employee runs the boring parts of operations 24/7 and only escalates when a human is genuinely required.
How we build
We map the workflow, the systems, the failure modes, and the business outcome worth automating. We ask which decisions the regulator audits and which ones don't matter.
We design the agent loop, the state model, the integrations, the human escalation paths, and the production guardrails. Output: a build plan and a model risk doc.
Focused first version, validated on real production-shaped data behind a feature flag. We expand only where the data supports it — never speculative scope.
Hosted on Sonzai or your VPC. Weekly tuning windows. Memory and personality compound, so the AI Employee gets sharper with every conversation, not staler.
Why financial services is different
Every decision is logged with the inputs it saw, the prompt assembly, the model used, and the policy that constrained it.
Run inside your tenancy or ours. Memory and audit data can stay on your side; only inference traffic crosses the boundary.
Configurable approval gates on irreversible actions — you decide which thresholds escalate to a human.
Deployments in Singapore, Indonesia, Malaysia, Philippines, Thailand, and Vietnam are in scope. We've shipped under MAS and OJK regimes.
Wealth advisor, KYC analyst, fraud-triage specialist — every AI Employee on the project shares a typed knowledge graph. Decisions, account context, MAS-relevant policy, and customer history all live in the same store, with audit trail and source anchoring. Newer information supersedes older facts; nothing the desk learns goes back to zero.
Every conversation ends with sessions.End() — kicking off fact extraction, dedup, personality drift, mood update, diary, and quality scoring. Per-(agent, customer) SGD and bandits tune retrieval to the way that specific account thinks. Day 1 is competent; Year 1 is a relationship the customer doesn't want to leave.
Frequently asked
Yes. Sonzai's runtime supports the audit trails, human-in-the-loop escalation, configurable PII redaction, and on-prem / VPC deployment options that MAS-regulated workflows require. The team has shipped production AI inside DBS Bank and OCBC under exactly these constraints — we know what survives a model risk review and what does not.
Free 30-minute consultation with the founders. We'll scope the workflow, the systems, and the regulatory edge — and tell you honestly whether managed build is the right path.