TOGETHER WITH CFOLAYER

Bespoke consulting for finance tech stack strategy, selection, & implementation

Get a personalized roadmap for your finance tech stack. CFOLAYER partners with CFOs and controllers to design, select, and implement the right systems for your scale and business model - eliminating vendor confusion and slow time-to-value.

Microsoft adds finance skills and data connectors to Copilot in Excel (6 min read)
Microsoft added finance features to Copilot in Excel, including skills that let teams define repeatable workflows such as a discounted cash flow model or the monthly close, new connectors to market and private company data, and clearer traceability that attributes each change to Copilot. What this means for your stack: the new skills let you standardize how Copilot builds recurring models like a variance pack, and the Show Changes attribution gives you an audit trail, but several data connectors require separate paid subscriptions.

Mistral releases OCR 4 with confidence scores for document extraction (7 min read)
Mistral released OCR 4, a document reading model that returns not just text but the location of each block, a label for what it is, and a confidence score for each word, priced at $4 per 1,000 pages. It can run inside a single container so teams can keep documents on their own systems, and Mistral lists invoice processing and compliance checks as core uses. What this means for your stack: OCR (optical character recognition, turning document images into data) that returns a confidence score for each field lets you auto-post the high-confidence invoice data and send only the uncertain fields to a person, the practical path to nearly touchless accounts payable. On clean digital invoices, AI OCR now runs about 95 to 99 percent accurate, and confidence-based review can push effective accuracy past 99 percent.

Baidu open-sources Unlimited-OCR for long documents (2 min read)
Baidu released Unlimited-OCR, a free and open-source document parsing model that reads long, multi-page documents in a single pass and builds on the earlier DeepSeek-OCR project. Teams can run it on their own hardware rather than paying a per-page fee. What this means for your stack: a self-hostable, open-source parser is worth watching if data privacy or per-page fees have blocked you from automating invoice and document capture, though open models can be non-deterministic, meaning the same invoice may not extract identically twice, which matters for accounting controls.

Analysis questions whether AI-native ERP will unseat incumbents (5 min read)
The Financial Revolutionist questions whether AI-native challengers can actually displace established ERP (core accounting system) vendors, using Brazil's market as a test of the thesis.

Anthropic releases Claude Sonnet 5 at a lower price point (6 min read)
Anthropic released Claude Sonnet 5, its most capable mid-tier model, at introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, after which it rises to $3 and $15. The company positions it close to its top Opus model on agentic tasks but at a lower price. What this means for your stack: cheaper capable models can lower the cost of finance AI work, but Sonnet 5 uses a new tokenizer that can turn the same input into up to 35 percent more tokens, so check real usage rather than assuming a flat price cut.

Anthropic launches Claude Tag, an AI teammate you tag into Slack (4 min read)
Anthropic launched Claude Tag, a way for teams to add its AI assistant to a Slack channel and hand it work by tagging it, with the assistant planning and completing tasks using the tools it is granted. Administrators control which tools and data it can touch, set limits on token spend, and can review a log of everything it did. What this means for your stack: if your finance team pilots AI agents in chat, the controls here, scoped tool access, per-channel spend caps, and a full action log, are the kind of governance your controllers and auditors will expect.

PLUS Mythos and Fable 5 progress through governmental review roadblocks

How companies are budgeting and capping AI token spend (4 min read)
The Wall Street Journal reports on how companies are trying to control artificial intelligence token spend, the per-use cost of running AI models, as it becomes one of the fastest-growing lines in the technology budget. Approaches include per-employee caps, charging costs back to the teams that use them, and matching cheaper models to routine tasks. What this means for your stack: plan to own AI token spend as a real budget line, with per-team caps, chargebacks, and model tiering in place before usage runs ahead of you.

Companies scramble to stop staff from maxing out AI budgets (4 min read)
TechCrunch reports that many companies blew past their 2026 AI budgets early in the year, often because employees run many small, casual queries, and are responding with per-person spending caps and by tying AI budgets to specific projects. What this means for your stack: the small, everyday queries are what drain AI budgets, so meter usage at the team level instead of watching only the large projects.

Deloitte guide on budgeting for AI token spend (5 min read)
Deloitte offers a reference on how AI token pricing works and how finance teams can budget and forecast it, suggesting they treat it like cloud spend with usage monitoring and tiered model use. What this means for your stack: treat AI tokens like cloud spend, with usage monitoring and model tiering, premium models for hard reasoning and small or open models for routine tasks.

Bain: CFOs move from funding AI to using it (7 min read)
Bain argues that finance chiefs who spent the past two years funding AI across the company are now adopting it inside their own finance teams, and lays out where it is delivering results and where it lags.

Four questions to ask before trusting AI with tax work (3 min read) — Thomson Reuters' checklist for vetting AI before it touches tax work.

Payments opinion: infrastructure is not a strategy (4 min read) — Why committing to payments plumbing before strategy costs you later.

Worth a second look

AICPA outlines 2026 state policy shifts for accounting (4 min read)
The AICPA and CIMA outline the state policy changes reshaping accounting in 2026, including new pathways to CPA licensure and a shift toward individual-based mobility, as roughly 43 states act to ease the accountant shortage. What this means for your stack: if you hire or serve clients across state lines, the move to individual-based CPA mobility and new licensure paths changes how you verify credentials and staff engagements.

BCG says vibe coding is coming to finance, and to set guardrails first — As finance teams let AI write more of their own tooling, the guardrails case is worth a second look.

Using AI agents for QA, and what it means for testing finance implementations — A practical take on AI agents for quality assurance, applied to testing finance system changes.

Why governance clearinghouses may win the AI agent era (9 min read)
Investor Jamin Ball argues that the durable advantage in the AI era will not be owning the system of record but becoming the clearinghouse that governs what agents can see, do, and touch, and that keeps the audit trail. He frames governance and permissions as the new lock-in, harder to leave than data was. What this means for your stack: as you buy agent tools, weigh who holds the governance and audit layer, since that layer becomes the hard-to-replace system over time.