The finance systems & AI brief for controllers and CFOs
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.
🏛️ Systems & Stack
Numeric, best known for close management software, launched a Financial Data Platform that keeps every business event as a full record with its source document and metadata, then generates the accounting entries from that data instead of starting from the ledger. It is pitched as a replacement for the ERP (the core accounting system) that NetSuite customers can adopt one module at a time rather than in a single cutover. What this means for your stack: before you take the pitch at face value, ask how audit trail, period locks and manual adjustments work when journal entries are generated from event data, because that is where an event-first platform differs most from the ledger your auditors already know. Discuss with ChatGPT →
Ramp moves into accounts receivable, from contract to cash (4 minute read)
Ramp launched Ramp AR, which reads a signed contract or purchase order to draft the invoice, prepares collections follow-ups for a person to approve before sending, and matches incoming payments to open invoices before syncing them to the ERP. It takes Ramp, known for cards and bill pay, onto the revenue side of the ledger. What this means for your stack: if Ramp already runs your payables, ask for a trial on one customer segment and check how its revenue recognition timing lines up with your ASC 606 (revenue recognition standard) policy before moving invoicing off your ERP. Discuss with ChatGPT →
Also: CFO Dive coverage
Numeral raises $100 million to automate sales tax end to end (3 minute read)
Numeral, a sales tax compliance startup founded in 2023, raised a $100 million Series C led by Insight Partners, bringing total funding to $157 million, and reported 327 percent year-over-year growth in the transactions it processes. It covers nexus monitoring (tracking where you owe tax), registration, calculation, filing and exemption certificates, and supports VAT in more than 90 countries across 40-plus billing and ERP integrations. Discuss with ChatGPT →
Ema raises $77 million on the bet that agents replace SaaS apps (2 minute read)
Ema, which builds AI agents that run multi-step processes across HR, IT and finance inside a company's existing software, raised a $77 million Series B led by Creaegis, citing 50-fold revenue growth over two years, about 80 percent gross margins and roughly 180 percent net dollar retention. Its founder argues legacy business applications are "mostly becoming like a database." Discuss with ChatGPT →
🤖 AI in Finance
Why AI spending is booming but productivity is not (4 minute read)
Chamath Palihapitiya argues that AI costs the median US company about $12.50 per employee a month against roughly $8,500 in monthly employee cost, so it needs only a 0.15 percent productivity gain, about three extra productive minutes a week, to pay for itself, and that gains are missing because most firms automate broken processes instead of redesigning them first. He points to computers, where US productivity growth only rose from 1.5 to 3.3 percent a year once work was reorganized around them. What this means for your stack: before you point AI at a close or AP step, map it and cut the steps you would not design today, or you will pay to run a bad process faster. Discuss with ChatGPT →
Gartner: even simple finance AI takes 9 to 10 months to pay off (3 minute read)
A Gartner survey of 160 senior finance leaders found that simpler AI uses such as data extraction, AP and AR automation and report creation take about 9 to 10 months to deliver value, forecasting and scenario planning take longer, and AI literacy has replaced hiring as the top barrier. CFO Dive's write-up adds that 55 percent of CFOs saw positive returns overall, but 57 percent could not tell for individual use cases. What this means for your stack: give every AI project a named metric and a 9 to 10 month review date, so the decision to scale or cut is made on evidence. Discuss with ChatGPT →
Also: CFO Dive write-up
Esker spent roughly four times its AI budget this year, so CFO Scott McDermott now plans AI token use (the unit AI providers bill by) per employee next to salary, bonus and payroll tax, using actual provider usage by function to set a monthly run rate. He says finance and R&D "run up overages like crazy." What this means for your stack: pull usage exports from your AI vendors by department now, so your 2027 plan carries an AI cost per role instead of one lump sum in the software line. Discuss with ChatGPT →
One in five finance staff now fear AI will take their job (2 minute read)
An insightsoftware survey of 854 finance professionals at companies with more than 500 employees found 20 percent fear AI will replace their jobs, up 7 points from last year, while 18 percent cite accuracy concerns, including hallucinations (confident but wrong AI answers). Discuss with ChatGPT →
⚖️ Regulation & Reporting
AICPA and a former IRS chief launch a council on AI risk in tax (2 minute read)
The AICPA (the national CPA professional body) and former IRS Commissioner Danny Werfel formed the Council on AI Risk in Tax, which will test an AI Risk Framework for Tax covering 20 risk areas in four groups: information integrity, fairness and legitimacy, security and data, and institutional capacity. Werfel's line: "we cannot assume that emerging AI tools will police themselves." What this means for your stack: if your tax team or provider uses AI in return prep, use the framework's four risk groups as the agenda for your next review of their controls. Discuss with ChatGPT →
25 state attorneys general ask Congress to regulate frontier AI (3 minute read)
A bipartisan group of 25 state attorneys general asked Congress for a federal framework for frontier AI (the most capable models from the largest labs), including mandatory safety testing, investigator access to company records after incidents, and explicit room for states to set stricter rules. Discuss with ChatGPT →
⚡ Quick Links
Anthropic's IPO filing shows $11.5 billion in quarterly revenue (4 min read) An $8 billion loss in 2025; two customers drove nearly a quarter of revenue.
Ramp is staging a one-night Broadway musical about bill pay (2 min read) A show about bills, starring Bills, at the Broadhurst Theatre.
🔁 ICYMI
Primary's Emily Man, Cassie Young and Nick Daley argue AI applications can deliver what SaaS promised by turning every deployment into better judgment, through what they call a continual learning loop: a scorebook of decisions, expert grades and real outcomes that improves the system across every customer. They name Tabs (billing and collections) and Casap (card disputes) as examples. Why now: Published September 14 and never sent in LEDGER. It is the investor case for the same bet Numeric and Ramp made this week, that the vendor who does the work and learns from it beats the one that only stores the data. Discuss with ChatGPT →
Ramp CEO Eric Glyman announced Ramp's UK launch in one line: "UK finance teams have waited long enough. Ramp is now live in the UK. Sorry we're late." Why now: Read it beside this week's Ramp AR launch. In two weeks Ramp added a new country and the revenue side of the ledger, which matters if you run US and UK entities on separate spend tools. Discuss with ChatGPT →
📖 Worth The Read
Simon Taylor argues Amazon's block of Meta's Muse shopping agent is a fight over who owns the customer, who collects the fees and who vouches that an agent is legitimate, and that merchants, unlike banks facing screen scraping (logging in with a customer's stored password), have the tools and motive to keep blocking. He sees the September 10 Know Your Agent framework from Visa, Mastercard and Ant as the first serious attempt at a shared standard. What this means for your stack: add rules for AI agents that buy or pay on the company's behalf, including who can authorize one, which cards it can use and what limits apply, to your card and procurement policies before the first one shows up in your expense data. Discuss with ChatGPT →
📚 Definitions
Know Your Agent (KYA). A way to verify an AI agent before letting it transact: who operates it, who certified it, what it is allowed to do and who is liable if it goes wrong. Visa, Mastercard and Ant announced a shared KYA framework on September 10. Think of it as KYC (know your customer) for software that spends money.
Agentic commerce. AI agents shopping and paying on a person's or company's behalf, from finding the item to completing checkout. It is why Amazon blocked Meta's Muse agent and why Shopify opened its checkout to agents this week.
Agent payment protocols. Competing rulebooks for how agents identify themselves and pay: Visa's Trusted Agent Protocol, Mastercard's Verifiable Intent (proof a person actually authorized the purchase), Google's Universal Commerce Protocol and OpenAI and Stripe's Agentic Commerce Protocol. Until one wins, expect vendors to support several.
WebMCP. A browser-based version of MCP (Model Context Protocol, the standard way to connect AI tools to software) that lets an agent inside a web browser use a site's functions directly instead of clicking through pages. Shopify's new checkout tools use it.
Event-based accounting (financial data platform). Storing each business event, such as an order, a refund or a usage record, as a complete record with its source document, then generating journal entries from it. Numeric's new platform works this way. A traditional ERP keeps mostly the summarized entry, so the detail behind a number lives elsewhere.
AI time-to-value. How long an AI project takes to deliver measurable benefit after you start it. Gartner puts simple finance uses at 9 to 10 months. It is the AI version of a payback period, and a useful checkpoint for deciding to scale or cut.
AI literacy. Enough working knowledge to use AI tools well: writing clear instructions, checking outputs, knowing where they fail. Gartner now ranks the lack of it as the top barrier to finance AI, ahead of hiring.
AI cost per employee. Monthly AI spend, mostly tokens, divided across the people or roles who use it. Esker now plans it alongside salary and payroll tax. Social Capital puts the median at about $12.50 per employee a month.
Agent harness. Everything around an AI model that lets it do a real job: integrations with your systems, playbooks, permissions and past records. Primary VC splits an agent into context, harness and model, and argues vendors win on the harness.
Continual learning loop and scorebook. A continual learning loop feeds an agent's results back into the system so it improves with use. A scorebook is the record that makes that possible: each decision, an expert's grade on it and the business outcome. For finance, it is also the audit trail you want anyway.
Productivity paradox. When heavy investment in a technology does not show up in productivity statistics for years. It happened with computers until work was reorganized around them, and Social Capital argues it is happening with AI now.