Multi-Agent Customer Support Automation Platform for Fintech

gradient_labs · Finance · 2025 · source Agentic: yes · promotional bias: med · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Fintech customer support automation

Gradient Labs needed to automate not only frontline support answers, but the back-office fraud, KYC, dispute, and outbound follow-up work that consumes much of fintech support time.

Constraints:

Fintech support is regulated, so agents need guardrails and deterministic orchestration around actions.
Back-office workflows can span days or weeks and involve internal systems, merchants, and customers.
Subject matter experts own support procedures, so workflow logic must be expressible without heavy engineering translation.
The platform needed to work across chat, email, and voice channels.

Architecture

Gradient Labs built a three-agent customer-support platform: an Inbound Agent handles customer requests and actions, a Back Office Agent runs internal processes such as disputes and KYC checks, and an Outbound Agent proactively contacts customers for missing information. Natural-language procedures encode business logic, deterministic skill orchestration controls tool use, and guardrails/state management support regulated multi-turn workflows.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  customer-channels["Customer Channels"] --> inbound-agent["Inbound Agent"]
  inbound-agent --> backend-apis["Backend APIs"]
  inbound-agent --> back-office-agent["Back Office Agent"]
  back-office-agent --> backend-apis
  back-office-agent --> outbound-agent["Outbound Agent"]
  outbound-agent --> customer-channels
  natural-language-procedures["Natural Language Procedures"] --> inbound-agent
  natural-language-procedures --> back-office-agent
  natural-language-procedures --> outbound-agent

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Inbound Agent`	Handles customer-initiated support over chat and email, answers questions, and calls APIs.	`Back Office Agent`, `Customer Channels`, `Backend APIs`
`Back Office Agent`	Processes internal workflows such as disputes, fraud investigations, and KYC compliance.	`Inbound Agent`, `Outbound Agent`, `Backend APIs`
`Outbound Agent`	Proactively contacts customers when information or action is needed.	`Back Office Agent`, `Customer Channels`
`Natural Language Procedures`	Plain-language business instructions that tell agents what steps and tools to use.	`Inbound Agent`, `Back Office Agent`, `Outbound Agent`
`Backend APIs`	Account, card, dispute, KYC, and operational systems used by support agents.	—
`Customer Channels`	Email, chat, and voice interfaces for inbound and outbound support.	`Inbound Agent`, `Outbound Agent`

Data flow

A customer may contact the inbound agent about a fraudulent transaction. The inbound agent freezes the card by API, triggers the back-office agent to start a dispute, and later the outbound agent contacts the customer for merchant-requested information. Procedures describe the steps in natural language while tools execute backend actions.

“a customer contacts support reporting a fraudulent transaction. The inbound agent receives the complaint and immediately freezes the card via API call.”summary, stated

“It then triggers the back office agent to initiate a dispute with the merchant. Days later, when the merchant responds requesting additional information”summary, stated

Why this architecture

Gradient Labs designed separate agents because the visible support interaction is only part of fintech support work; back-office and outbound processes need different triggers, timescales, and systems. Natural-language procedures make the platform deployable by subject matter experts who already write human-agent training documents.

“existing customer support automation solutions only address what their CEO calls the "tip of the iceberg" - the visible frontline customer support interactions”summary, stated

“By allowing procedures to be written in plain natural language, Gradient Labs enables fintech companies to essentially train their AI agents the same way they would train human agents”summary, stated

Models used: unspecified LLMs

Scale

The company raised a Series A in 2025.
“The company operates in the fintech customer support space and raised their Series A in 2025.”summary, stated
The architecture coordinates three agent types.
“To address this comprehensive challenge, Gradient Labs built three distinct but coordinated agent types:”summary, stated
Back-office processes may span days or weeks.
“These processes often involve interactions with internal systems and can span days or weeks.”summary, stated

Alternatives considered (roads not taken)

Automate only frontline customer queries

Rejected because: Gradient Labs argues the bulk of fintech support work is hidden back-office work.

“the bulk of actual work in fintech customer support occurs below the surface: back-office tasks such as fraud dispute management”summary, stated

Encode procedures as state machines or code

Rejected because: Plain-language procedures let non-technical SMEs transfer knowledge directly.

“Traditional approaches might use rigid state machines, decision trees, or complex code. Gradient Labs chose a different path: natural language procedures”summary, stated

Make support purely reactive

Rejected because: The outbound agent proactively reaches out for KYC updates, dispute information, and mass communication campaigns.

“The Outbound Agent proactively reaches out to customers when information is needed or actions are required.”summary, stated

Patterns

`multi-agent-orchestration` (architecture)

Gradient Labs coordinates inbound, back-office, and outbound agents across complex support workflows.

“featuring three coordinated AI agents: inbound, outbound, and back office.”summary, stated

`tool-use-orchestration` (execution)

Agents use backend APIs to freeze cards, update accounts, trigger disputes, and perform support actions.

“This agent can answer questions but, critically, can also take actions by calling APIs to freeze cards, update account information”summary, stated

`prompt-versioning` (operations)

Natural-language procedures serve as editable operational instructions for agents.

“Procedures consist of step-by-step instructions written in natural language.”summary, stated

`safety-policy-enforcement` (safety)

The platform uses guardrails and deterministic skill orchestration for regulated fintech actions.

“Their solution uses a modular agent architecture with natural language procedures, deterministic skill-based orchestration, multi-layer guardrails for regulatory compliance”summary, stated

Tradeoffs

Use natural-language procedures

Chose Plain-language procedure documents with embedded tool calls over Rigid state machines, decision trees, or complex code.

Rationale: SMEs could transfer support knowledge directly and train agents like human agents.

“Traditional approaches might use rigid state machines, decision trees, or complex code. Gradient Labs chose a different path: natural language procedures”summary, stated

Build three coordinated agent types

Chose Inbound, back-office, and outbound agents over A single customer-support chatbot.

Rationale: The full workflow includes customer-facing, internal, and proactive follow-up stages.

“The power of this three-agent system emerges when they coordinate on complex workflows.”summary, stated

Gotchas (what bit them)

Frontline-only automation misses most fintech support labor.

Root cause: Fraud disputes, KYC, investigations, and compliance work happen behind the customer conversation.

Resolution: Gradient Labs added back-office and outbound agents rather than only an inbound chatbot.

“while missing the complex back-office tasks like fraud disputes and KYC compliance that consume most human agent time.”summary, stated

Procedure knowledge can bottleneck if only engineers can encode it.

Root cause: Support procedures are owned by non-technical subject matter experts.

Resolution: The system uses plain natural-language procedures similar to human-agent training documents.

“the knowledge of how to handle customer support scenarios resides in the heads of subject matter experts who are typically not technical.”summary, stated

Evaluation

The summary describes architectural behavior and example workflows, but does not provide a full public benchmark in the excerpted evidence.

“The company addresses the challenge that traditional customer support automation only handles the "tip of the iceberg"”summary, implied

Outcomes

Gradient Labs built a coordinated fintech-support platform spanning inbound, back-office, and outbound workflows.
The platform can automate multi-day dispute workflows from intake through follow-up.
Natural-language procedures lowered the barrier for customers to transfer human support knowledge into agent behavior.

The source describes three agent types and workflows spanning days or weeks; it does not provide live volume metrics in the summary excerpt.

“This represents true end-to-end automation of a complex, multi-day, multi-party process.”summary, stated