# Machine Learning

Apply machine learning to automate financial categorization and analysis

14 posts [View all tags](https://beancount.io/blog/tags) [Back to All Posts](https://beancount.io/blog)

November 11, 2026 12 min read

## [Feature Views, Not Seats: Booking Tecton's Consumption-Credit Billing as ASC 606 COGS for an ML Platform](https://beancount.io/blog/2026/11/11/feature-views-not-seats-tecton-consumption-billing-cogs-guide)

Tecton bills consumption credits plus cloud spend in your own account: book production serving to COGS, experimentation to R\&D, and accrue the arrears.

Mike Thrift

cost-of-goods-sold machine-learning saas

November 11, 2026 12 min read

## [Qdrant and Milvus at Scale: When Self-Hosted Vector Search Crosses Over From CapEx to Cheaper Than Pinecone's Per-Query Bill](https://beancount.io/blog/2026/11/11/qdrant-milvus-self-hosted-vs-pinecone-crossover-breakeven-guide)

Self-hosted Qdrant or Milvus beats Pinecone once query volume pushes per-unit metering past fixed infrastructure cost — often at 10 to 50M vectors.

Mike Thrift

cost-of-goods-sold machine-learning cloud-services

October 10, 2026 10 min read

## [Feast vs. Tecton: Capitalizing In-House Feature-Store Infrastructure Under ASC 350-40 When You Build Instead of Buy](https://beancount.io/blog/2026/10/10/feast-vs-tecton-feature-store-asc-350-40-capitalization-guide)

Building on Feast turns engineering payroll into a capital asset under ASC 350-40; buying Tecton keeps it operating expense. Document the stages.

Mike Thrift

software-capitalization saas machine-learning

July 26, 2026 7 min read

## [AI-Powered Fraud Detection for Small Businesses: Real-Time Auditing Without a Big-Four Budget](https://beancount.io/blog/2026/07/26/ai-fraud-detection-small-business-real-time-continuous-auditing-guide)

Business email compromise cost U.S. companies over $3 billion in 2025, averaging $137,000 per incident — and 45% of small businesses hit by BEC close within six months. AI-driven continuous auditing now starts around $20–$70/month; here's how anomaly detection, vendor account validation, and free controls like payment-change callbacks cut the risk without an enterprise fraud stack.

Mike Thrift

fraud-detection fraud-prevention small-business

July 20, 2026 6 min read

## [AI Underwriting in 2026: How Small Business Loans Went From Weeks to Hours](https://beancount.io/blog/2026/07/20/ai-underwriting-small-business-loans-funding-speed-2026-guide)

Non-bank lenders now fund small business loans in an average of 1.8 days versus 4–8 weeks at traditional banks, driven by AI underwriting that reads live bank feeds instead of tax returns. Here's how the models judge your business, why a 1.4 factor rate can hide a 150%+ APR, and how clean, reconciled books improve both approval odds and pricing.

Mike Thrift

ai machine-learning small-business-loans

July 13, 2026 8 min read

## [Embedded AI in Accounting: Why Explainable AI and Real-Time Oversight Are Becoming Non-Negotiable](https://beancount.io/blog/2026/07/13/embedded-ai-explainable-ai-real-time-controller-oversight-accounting-guide)

Embedded AI now auto-categorizes transactions and flags anomalies in real time inside tools like QuickBooks and Xero, but industry estimates put manual bookkeeping error rates at 1-3% versus below 0.5% for well-monitored AI systems — the tradeoff is that AI errors compound systematically until a human catches the pattern, which is why explainable AI and real-time controller oversight are becoming standard practice.

Mike Thrift

ai machine-learning automation

July 11, 2026 8 min read

## [Audited by an Algorithm: What the IRS's New AI Governance Policy Means for Your Small Business](https://beancount.io/blog/2026/07/11/audited-by-algorithm-irs-ai-audit-selection-irm-10-24-1-guide)

The IRS's February 2026 policy, IRM 10.24.1, formalizes how AI models — now running 126 active projects across six annual cycles — flag small business and self-employed returns for audit, requiring mandatory human review before any AI referral becomes an examination.

Mike Thrift

tax audit ai

July 10, 2026 8 min read

## [AI Receipt Scanning Apps in 2026: The OCR-to-LLM Shift and the Hallucination Problem](https://beancount.io/blog/2026/07/10/ai-receipt-scanning-expense-apps-ocr-hallucination-guide)

LLM-based receipt scanners read crumpled and faded receipts far better than template OCR, but when a field is ambiguous they can hallucinate a plausible wrong total instead of leaving it blank. What accuracy benchmarks actually show, the IRS's $75 receipt threshold and lodging exception, and a weekly review workflow that catches confident errors before they hit your books.

Mike Thrift

ai machine-learning expense-management

July 8, 2026 8 min read

## [Will AI Replace Your Bookkeeper? What the BLS Data Actually Shows](https://beancount.io/blog/2026/07/08/will-ai-replace-your-bookkeeper)

BLS data projects a 6% decline in bookkeeping clerk employment through 2034 while accountant and auditor employment grows 5%, a split driven by AI automating transaction coding and data entry while leaving judgment-based work to humans.

Mike Thrift

ai bookkeeping automation

May 13, 2026 12 min read

## [AI Bookkeeping for Small Business in 2026: Where Generative AI Wins and Where It Fails](https://beancount.io/blog/2026/05/13/ai-powered-bookkeeping-automation-generative-ai-transaction-categorization-reconciliation-real-time-reporting-small-business-guide)

AI bookkeeping tools now hit 85–95% categorization accuracy and shrink month-end close from days to under two hours, but the 2–5% confidently wrong residual is where audit risk hides. A practical workflow for small businesses covering daily, weekly, monthly, and quarterly review steps.

Mike Thrift

ai bookkeeping automation

May 22, 2025 4 min read

## [AI Fraud Detection in Plain-Text Accounting](https://beancount.io/blog/2025/05/22/how-ai-powered-fraud-detection-in-plain-text-accounting-protects-financial-records)

Learn how integrating artificial intelligence with plain-text accounting systems like Beancount.io enhances financial security by identifying subtle fraud patterns that traditional methods often miss.

Mike Thrift

ai fraud-detection accounting

May 21, 2025 5 min read

## [Beyond Human Error: AI Anomaly Detection in Plain-Text Accounting](https://beancount.io/blog/2025/05/21/ai-driven-anomaly-detection-in-financial-records-how-machine-learning-enhances-plain-text-accounting-accuracy)

Learn how AI-powered anomaly detection transforms plain-text accounting by enhancing error detection and maintaining transparency, addressing the critical vulnerabilities in financial systems.

Mike Thrift

ai fraud-detection accounting

[Start Free→](https://beancount.io/sign-up/?next=%2Fledger%2F) [Documentation](https://beancount.io/docs)

## Where to look next

- [Sitemap](https://beancount.io/sitemap.xml) - all pages with hreflang alternates
- [LLMs index](https://beancount.io/llms.txt) - curated links for AI agents (also /llms-full.txt for full docs)
- [Documentation](https://beancount.io/docs) - every doc is also available as text/markdown via .md suffix or Accept negotiation
- [REST API reference](https://beancount.io/docs/api-reference) - interactive endpoints, schemas, authentication, and code samples
- [OpenAPI spec](https://beancount.io/openapi.json) - OpenAPI 3.1 spec for this website's public content endpoints; the product contract for ledger operations is at https://beancount.io/api-gateway/v1/openapi.json
- [GraphQL](https://beancount.io/graphql) - public, read-only content API over the blog and research logs (no credentials, no ledger data): 12 query fields over 23 types
- [MCP server](https://beancount.io/mcp) - Model Context Protocol at https://beancount.io/api-gateway/mcp
