The finance systems & AI brief for controllers and CFOs

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.

PayPal board rejects Stripe and Advent's $53 billion bid (11 minute read)

PayPal's board formally rejected a $53 billion buyout offer from Stripe and private equity firm Advent International on July 20, arguing the $60.50 per share price undervalues the company and pushing for closer to $70. The two sides are expected to keep negotiating, with PayPal's July 28 earnings report likely to shape the next move.

Augustus raises $180M to build AI-powered banking infrastructure abroad (4 minute read)

Augustus, operating as Global Dollar Bank, raised a $180 million Series B at a $1 billion valuation to expand its API-first banking platform, which pairs a US national bank charter with an AI-powered core banking system called Marble. The round was led by Tiger Global with participation from fintech veterans behind Nubank, Ramp, Circle, and Deel, and will fund expansion into Latin America, Southeast Asia, the Middle East, and Africa. What this means for your stack: if your company operates internationally, API-first banking platforms like Augustus are starting to offer direct dollar-denominated accounts and multi-rail money movement (Swift, ACH, SEPA, stablecoin) as an alternative to stitching together separate local banking relationships in each country.

Natural raises $30M to build payment rails for autonomous AI agents (3 minute read)

Natural raised a $30 million Series A led by Forerunner's Kirsten Green to build what it calls an agent orchestration layer, payment infrastructure that lets AI agents autonomously pay vendors, collect funds, and transact with humans and other agents without a human approving each transaction. The one-year-old startup is positioning itself as a direct challenger to Stripe, betting that payment volume could grow by several orders of magnitude once transactions happen at computer speed instead of human speed. What this means for your stack: as agent-to-agent payment infrastructure matures, expect vendors to start asking whether your AP (accounts payable) systems can authorize and audit payments initiated by AI agents rather than people, well before that becomes a standard procurement question.

OpenAI's CFO proposes 'useful intelligence per dollar' to measure AI ROI (6 minute read)

OpenAI CFO Sarah Friar proposed a framework for measuring AI return on investment called useful intelligence per dollar, based on how much real work AI completes, what each successful task actually costs after retries and human review, and how dependable the results are. The framework argues the cheapest model per token is not always the cheapest per finished task, since a pricier model that gets a task right the first time can beat a cheaper one that needs several attempts. What this means for your stack: when comparing AI tools or models, track cost per successfully completed task rather than just price per token, since retries and manual review can make a cheap model more expensive in practice.

OpenAI's Codex and ChatGPT Work agents pass 10 million weekly users (3 minute read)

OpenAI says its two agent products, the Codex coding assistant and the newer ChatGPT Work, passed 10 million combined weekly users less than two weeks after ChatGPT Work launched on July 9. The figure counts weekly active users rather than paying subscribers, and finance teams are among the early adopters using it to speed up month end reporting, reconciliations, and forecast decks. What this means for your stack: if you're evaluating ChatGPT Work or similar agent tools for close or FP&A (financial planning and analysis) tasks, remember that OpenAI's 10 million figure measures weekly logins, not proof the tool reliably finishes finance tasks unsupervised, so pilot before you rely on it.

88% of London finance staff use AI weekly, Bloomberg survey finds (3 minute read)

A Bloomberg survey of 500 London financial services staff found 88% use AI tools daily or weekly and 90% say it has boosted their productivity, but nearly three quarters say it has not reduced their overall workload. Just 1% think human oversight is unnecessary, with risk, compliance, investment, and trading decisions cited most often as the areas needing a human final check. What this means for your stack: expect AI adoption to add new, higher-value work to people's plates rather than cut headcount need, and build human review checkpoints into any AI workflow touching risk, compliance, or trading decisions specifically.

EY faces class action after tax data breach exposed client files (2 minute read)

EY, one of the Big Four accounting firms, is facing a proposed class action lawsuit after an unauthorized party accessed an internal IT help desk platform between March 28 and April 12 and downloaded tax and financial documents belonging to a number of clients. The lawsuit, filed in New York, argues EY had the resources to protect the data and failed to do so, and seeks class certification plus monetary damages. What this means for your stack: if any advisor or vendor has access to your tax and financial documents through a support ticketing platform, ask what data that platform retains and for how long, since support tools are increasingly the weak link rather than the core system.

Global regulators handed out $931M in Q2 fines for ignored controls, not missing ones (3 minute read)

Global regulators issued more than $931 million in penalties above $1 million during Q2 2026, the largest quarterly total in a year, according to enforcement data from risk-intelligence firm Corlytics. Nearly a third of the 37 major penalties involved firms that already had the right controls in place but failed to escalate alerts or update oversight, including a $100 million SEC penalty against Western Asset Management and a first-of-its-kind $24.5 million Australian scam-controls fine against HSBC. What this means for your stack: this quarter's fines mostly targeted firms that had controls on paper but failed to escalate or act on the alerts those controls generated, so it's worth auditing whether your team actually reviews and closes out exceptions, not just whether the controls exist.

UK brings Buy Now, Pay Later fully under FCA regulation (3 minute read)

Starting July 15, Buy Now Pay Later lenders in the UK must be authorized by the Financial Conduct Authority (FCA, the UK's financial regulator) and run affordability checks on every purchase, even ones under fifty pounds. The rules also require clearer upfront disclosure of repayment dates and the consequences of a missed payment, which adds steps to the checkout flow. What this means for your stack: if you sell into the UK and offer Buy Now Pay Later at checkout, confirm your provider is FCA-authorized and test how the new affordability checks affect checkout conversion before assuming performance holds steady.

Finance hiring is shifting from junior analysts to AI-literate mid-level roles (7 minute read)

FP&A (financial planning and analysis) newsletter Mostly Metrics reports that hiring managers are increasingly redirecting junior analyst budgets toward manager and director-level hires who can pair AI tools with cross-functional judgment. One hiring manager at a pre-IPO company cited in the piece said her team expects AI to absorb much of the reporting and analysis work traditionally done by junior, investment-banking-style analysts, shifting hiring focus toward candidates with deeper data skills instead. What this means for your stack: before backfilling a junior analyst role, consider whether the position's reporting and analysis work could shift to AI tools, and whether that budget is better spent on a mid-level hire who can direct that work instead.

Ant International raises $1.2B to expand cross-border payments and agentic commerce — Ant Group's international payments arm closed a $1.2 billion round backed by Alibaba to fund global expansion and new agentic commerce tools across its 150 million-plus merchant network.

CSI acquires Qolo to expand commercial banking and embedded finance tools — Core banking provider CSI picked up payments infrastructure company Qolo to add multi-rail money movement and embedded ledgering to its commercial banking platform; deal terms undisclosed.

Worth a second look

Building AI governance frameworks for finance — A four-part framework (policy, model risk controls, change management, monitoring) that's aged well now that this week's AI-ROI and AI-trust stories all point back to the same root issue: governance before scale.

Square lets ChatGPT and Claude take orders for small sellers — Worth another look now that OpenAI's agents have crossed 10 million users and Stripe is fighting for control of the next commerce layer; agentic checkout is no longer a side experiment.

Texas TRAIGA is live: what finance teams deploying AI need to know — A practical compliance checklist for a law that's still ramping up enforcement, useful context alongside this week's EY breach and BNPL stories.

a16z: AI agent workforces fail the same way human ones do (8 minute read)

Argues that AI agents fail for the same reasons mismanaged employees do: without clear instructions, they burn effort in unproductive 'loops' instead of doing useful work. The piece frames evals (tests that define what a good AI result looks like) as the new equivalent of OKRs (a common goal-setting framework), calling them the most valuable resource a company can build for scaling AI reliably. What this means for your stack: before rolling out more AI agents, write down specific evals, clear pass or fail criteria for what a correct result looks like, since vague instructions are what create expensive, unproductive AI loops.

Two terms that come up a lot in this issue:

Context window: the amount of text and information an AI model can actively hold and reference at one time while working on something. Once a task or conversation grows past it, older details start falling out of view, which is a big part of why AI agents lose the thread on long or complex jobs (see the Worth The Read piece below).

Orchestration: the software layer that coordinates a multi-step AI job, calling tools, checking intermediate work, retrying failures, rather than a single question-and-answer exchange. It's what lets something like ChatGPT Work or Codex plan a task, work through several steps, and stay on it for hours instead of stopping after one response. It's also, this week, a literal product category: see Natural's agent payment rails below.