Use Case Intelligence RSS Search use cases Search
Anthropic Software / SaaS North America

Rocket Money

Customer Service Automation

Rocket Money wanted to put a trustworthy financial expert in every customer's pocket, but a personal finance agent that reads account data, answers questions, and takes actions has to be highly accurate, observable, and cheap enough to scale.

Read the original story Explore all use cases
11x increase in code commits from 11 in December to 128 in July
9th of 76 PR authors across the entire repository within three months of her first commit
Near-zero hallucinations across Rowan's beta

Reported impact

Business outcome

Rocket Money reached near-zero hallucinations in Rowan's beta, increased monthly code commits 11x as it built the product, widened code contribution beyond engineering, and launched a specialized-agent architecture that can expand Rowan's financial assistant capabilities without sacrificing traceability.

  • Customer Experience
  • Productivity Improvement
  • Speed to Market

Solution

Rocket Money built Rowan as an agent-workflow hybrid on the Claude Platform. User requests arrive by text message and first pass through a Haiku 4.5 classifier that routes each job to a single-responsibility subagent. Most subagents run on Sonnet 5, while Opus 5 handles the financial tasks that need deeper reasoning, and a final formatting agent keeps Rowan's voice consistent. Routine tasks stay deterministic code; when a website changes, a repair agent updates the automation instead of leaving the flow broken. The full system runs through Amazon Bedrock on top of Rocket Money's existing app and data services, with every failure traceable to the exact classifier or downstream call that failed. Claude Code and Fable also accelerated development, letting a small team ship and maintain many specialized agents in parallel.

Data flow

Customers text Rowan, a Haiku 4.5 classifier routes each request, Sonnet 5 or Opus 5 subagents execute the task against Rocket Money's finance workflows and app data, a formatter standardizes the reply, and deterministic code paths or repair agents handle external-site automations. Everything runs on Claude through Amazon Bedrock with traceable failures.

Solution architecture

6 components · 4 layers
  1. Compute
    • Sonnet 5 Runs most of Rowan's single-responsibility subagents for day-to-day financial tasks.
    • Opus 5 Handles the highest-stakes financial tasks that need deeper reasoning.
    • Claude Code Helps Rocket Money's small team build and maintain Rowan's specialized-agent codebase.
  2. Serving
    • Claude Platform Supplies the core model platform behind Rowan's customer-facing financial assistant.
  3. Orchestration
    • Haiku 4.5 Classifies each incoming request and routes it to the right specialist agent.
  4. Uncategorized
    • Amazon Bedrock Provides the production runtime for Rowan on top of Rocket Money's existing systems.

Architecture clues

  • A Haiku 4.5 classifier routes each request to single-responsibility subagents.
  • Most subagents run on Sonnet 5, with Opus 5 reserved for financial tasks that need deeper reasoning.
  • Routine tasks remain deterministic code and a repair agent fixes flows when websites change.

Evidence from the source

A classifier running on Haiku 4.5 reads each incoming request and routes it to a subagent built to do exactly one thing.
Across Rowan's beta, the team has seen near-zero hallucinations.
All of this runs in production on Claude through Amazon Bedrock.
Most subagents run on Sonnet 5; for a financial task that needs real reasoning, the system reaches for Opus 5
the team's monthly code commits climbed from 11 in December 2025 to 128 in July 2026
Transparency Evidence & methodology 90/100 reference value · 87% confidence · 1 source

Review how this reference was scored and trace each normalized claim back to its documented source.

01

Scoring methodology

How scores are calculated

Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.

Customer reference value90/100
  • Problem clarity20/20
    Measures whether the customer's operational problem is specific enough to compare.
  • Implementation depth25/25
    Rewards named components, data flow, integrations, and architecture detail.
  • Outcome specificity20/20
    Measures whether the reported benefit is concrete rather than a generic claim.
  • Quantified impact15/25
    1 metric describe measurable impact.
  • Deployment scale10/10
    Rewards evidence of a live or scaled deployment and meaningful integration scope.
Evidence confidence87%
  • Source authority20/25
    The main evidence is an official vendor-published customer story.
  • Claim traceability21/25
    10 of 12 key fields link to a named source.
  • Direct evidence20/20
    5 direct source quotes retained.
  • Date provenance8/10
    The publication date comes from publisher-controlled metadata.
  • Source transparency8/10
    1 source reference document publisher, URL, authority, retrieval, and supported fields.
  • Implementation specificity10/10
    Rewards source-backed components, data flow, quantified impact, and a specific outcome.
02

Traceability

Sources & provenance

Field coverage shows which facts each source supports. Dates are shown only when documented by the source.

  1. Anthropic

    Rocket Money customer story

    First-party
    https://claude.com/customers/rocket-money
    Source type
    Vendor-published customer story
    Published
    Aug 26, 2026
    Supports
    • Customer
    • Business problem
    • Solution
    • Products
    • Solution components
    • Data flow
    • Integrations
    • technical area
    • use case category
    • Business outcome
    • outcome category
    • Metrics
    • Architecture
    • evidence quotes
    • Publication date

Date evidence

  • Publisher catalog date Aug 26, 2026 · Anthropic · First-party
    Open date source
Not separately sourced Industry, Region, Company size. Treat these normalized attributes as lower-confidence until a dedicated source is added.

Related

Similar use cases

See all like this
Anthropic Archived · Oct 8, 2026 Zendesk Software / SaaS North America Zendesk wanted customer-specific service agents to handle the complex work between a request and its resolution, but one-off automations were expensive to build, hard to localize, and difficult to govern across millions… 1 million custom agent executions in the first seven weeks of early access Customer Service Automation Amazon Bedrock · Claude Agent SDK · Claude Code Anthropic Aug 26, 2026 Rocket Money Software / SaaS North America Rocket Money wanted Rowan to deliver financial help by text message without making users open dashboards, while still keeping the assistant accurate enough to act on real money decisions. Customer Service Automation Claude Platform Google Cloud Modified · Oct 9, 2026 Tally Group Software / SaaS Australia This rapid processing surfaces targeted insights in under 20 seconds, eliminating manual checklists. At utility billing platform Tally Group , call center agents previously navigated up to six applications to… Customer Service Automation Apigee API Management · BigQuery · Gemini platform