AI-Powered Accounting Automation Using Claude and Amazon Bedrock

floqast · Finance · 2025 · source Agentic: yes · promotional bias: high · source-backed · extracted with prompt 0.5.4 (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: Accounting reconciliation and audit workflows

FloQast wanted to automate the bespoke final portion of accounting work, where organization-specific reconciliation rules and audit annotations often require manual intervention.

Constraints:

Architecture

FloQast combines Claude 3.5 Sonnet, Bedrock Agents, RAG/few-shot classification, audit-trail storage, and accountant review to automate custom reconciliation and compliance annotations while preserving human accountability.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  accounting-workflow-intake["Accounting workflow intake"]
  bedrock-agents["Bedrock Agents"]
  claude-3-5-sonnet["Claude 3.5 Sonnet"]
  rag-classifier["RAG classifier"]
  audit-trail-storage["Audit trail storage"]
  exception-handling["Exception handling"]
  accountant-review["Accountant review"]
  accounting-workflow-intake --> bedrock-agents
  bedrock-agents --> claude-3-5-sonnet
  bedrock-agents --> rag-classifier
  rag-classifier --> claude-3-5-sonnet
  bedrock-agents --> audit-trail-storage
  bedrock-agents --> exception-handling
  claude-3-5-sonnet --> accountant-review
  audit-trail-storage --> accountant-review
  exception-handling --> accountant-review
Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component Role Interacts with
Accounting workflow intake Receives reconciliation, transaction matching, and annotation tasks. Bedrock Agents
Bedrock Agents Orchestrate multi-step tasks across company data and accounting systems. Claude 3.5 Sonnet, RAG classifier, Audit trail storage, Exception handling
Claude 3.5 Sonnet Generates matching rules, annotations, and accounting workflow outputs. Accountant review
RAG classifier Uses few-shot examples and collected user data instead of fine-tuning for classification. Claude 3.5 Sonnet
Audit trail storage Records extracted data, annotation responses, and status updates for compliance. Accountant review
Exception handling Routes custom or low-confidence accounting cases for human attention. Accountant review
Accountant review Accepts or edits generated text and maintains professional accountability. Accounting workflow intake

Data flow

Bedrock Agents coordinate multi-step accounting tasks, RAG/few-shot classification supplies user-specific examples, and Claude generates matching or annotation outputs for accountant review.

3 supporting quotes
“Amazon Bedrock Agents enables generative AI applications to run multi-step tasks across company systems and data sources.”article, stated
“they use Retrieval Augmented Generation (RAG) with few-shot classification on data collected on the user’s behalf, removing the overhead of fine-tuning an LLM.”article, stated
“At this point, the accountant has the option to either accept the generated text or edit the text.”article, stated

Why this architecture

The architecture focuses on organization-specific final-mile accounting work and auditability rather than generic transaction automation.

“FloQast addresses the most complex and custom aspects of financial processes (the final 20%)—those intricate, bespoke aspects of accounting that are highly specific to each organization and often require manual intervention.”article, stated
“Structured storage and audit trail – You can maintain a structured record of each annotated document, capturing all extracted data, annotation responses, and status updates for streamlined compliance and audit trails”article, stated

Models used: Claude 3.5 Sonnet

Scale

Alternatives considered (roads not taken)

Fine-tuning a custom accounting model

Rejected because: FloQast uses RAG with few-shot classification because it avoids model fine-tuning while preserving security and developer ease of use.

“FloQast doesn’t need to fine-tune the model as a model consumer, so they use Retrieval Augmented Generation (RAG) with few-shot classification”article, stated

Owning generative AI infrastructure directly

Rejected because: Bedrock serverless deployment handles secure integration and spiky traffic without FloQast managing infrastructure.

“Because Amazon Bedrock is serverless, you don’t have to manage infrastructure to securely integrate and deploy generative AI capabilities into your application, handle spiky traffic patterns”article, implied

Accounting automation without structured audit storage

Rejected because: The system stores extracted data, annotation responses, and status updates for compliance and audit trails.

“Structured storage and audit trail – You can maintain a structured record of each annotated document, capturing all extracted data, annotation responses, and status updates for streamlined compliance and audit trails”article, stated

Generated annotations posted without accountant review

Rejected because: Accountants can accept or edit generated text before it becomes part of the workflow.

“At this point, the accountant has the option to either accept the generated text or edit the text.”article, stated

Patterns

tool-use-orchestration (orchestration)

Bedrock Agents provides the orchestration layer for multi-step accounting workflows across company systems and data sources.

“Amazon Bedrock Agents enables generative AI applications to run multi-step tasks across company systems and data sources.”article, stated

rag-retrieval (retrieval)

FloQast uses RAG with few-shot classification on customer-collected data instead of fine-tuning Claude for the accounting task.

“they use Retrieval Augmented Generation (RAG) with few-shot classification on data collected on the user’s behalf, removing the overhead of fine-tuning an LLM.”article, stated

human-in-the-loop-gate (workflow)

Accountants review generated rule text before saving and applying it as executable matching logic.

“At this point, the accountant has the option to either accept the generated text or edit the text.”article, stated

privacy-preserving-processing (safety)

The design emphasizes customer-understandable security and avoids fine-tuning on customer-specific accounting data.

“For this use case, this design mechanism produces a higher level of accuracy, a better security model that is understood by FloQast’s customers, and ease of use as a developer.”article, stated

Tradeoffs

Whether to fine-tune a model for accounting classification.

Chose RAG plus few-shot classification over fine-tuning the LLM.

Rationale: The article says RAG/few-shot removes fine-tuning overhead while improving accuracy, security understanding, and developer ease.

“FloQast doesn’t need to fine-tune the model as a model consumer, so they use Retrieval Augmented Generation (RAG) with few-shot classification”article, stated

How much infrastructure to own for generative AI deployment.

Chose serverless Amazon Bedrock over self-hosting and managing model infrastructure.

Rationale: Bedrock reduces infrastructure management while supporting spiky traffic and cross-region reliability.

“Because Amazon Bedrock is serverless, you don’t have to manage infrastructure to securely integrate and deploy generative AI capabilities into your application, handle spiky traffic patterns”article, stated

Gotchas (what bit them)

The hardest accounting work is organization-specific and resists generic automation.

Root cause: The final 20% includes bespoke processes, custom rules, and manual judgment that differ by company.

Resolution: Let accountants select example transactions and generate natural-language matching rules tailored to the specific process.

“FloQast addresses the most complex and custom aspects of financial processes (the final 20%)—those intricate, bespoke aspects of accounting that are highly specific to each organization and often require manual intervention.”article, stated

Accounting automation needs auditability, not just successful matching.

Root cause: Compliance processes require traceable records of matching, annotation, extracted data, and status updates.

Resolution: Maintain structured audit trails for matching activities and annotated documents.

“Structured storage and audit trail – You can maintain a structured record of each annotated document, capturing all extracted data, annotation responses, and status updates for streamlined compliance and audit trails”article, stated

Evaluation

FloQast reports customer impact metrics for reconciliation, audit process duration/discrepancies, and workload management after deploying the solution.

Metrics: 38% reduction in reconciliation time · 23% decrease in audit process duration and discrepancies · 44% improvement in workload management

“delivering measurable improvements, including a 38% reduction in reconciliation time, a 23% decrease in audit process duration and discrepancies, and a 44% improvement in workload management.”article, stated

Outcomes

AWS partner/customer post; results are vendor-reported and promotional.

“the solution has saved accounting teams countless hours, enabling them to shift away from manual spreadsheet work and focus on higher-value activities”article, stated