AI & Finance Weekly wk 28  ·  Jul 13 2026  ·  sources verified Jul 2026
AI & Finance
Weekly
Issue 28 Monday, Jul 13 2026

Get fluent in AI — and in what it means for your numbers.

Research-grade stories from the close, the audit room, and the finance-software stack. No hype, no filler — just what a sharp controller needs to know.

Models & Policy The Finance Stack Money & Rules AI Demystified Hype Check Personality

3 things to know this week

1
The SEC's quarterly-reporting era may be ending. The comment period on the SEC's proposed optional semiannual reporting framework (Form 10-S in lieu of 10-Q) closed July 6. If adopted, companies could begin filing twice a year as early as 2028 — cutting interim reporting cycles but raising real questions about ICFR frequency and audit-committee oversight.
2
Microsoft rewired Copilot in Excel for finance. A June 26 update added finance-specific skills, live ERP connectors (SAP, Dynamics 365, Workday, NetSuite), and a partner data marketplace. Controllers can now pull AP aging from SAP in natural language — without leaving the spreadsheet.
3
The White House is weeks away from a formal AI release framework. EO 14409 (June 2) established the skeleton; the Financial Times reports labs and the administration are in advanced talks to finalize voluntary pre-release testing standards — announcement expected this week. The first government-managed AI release cycle is already operational.

By the numbers

74%
of finance frontline employees use AI at least several times per week — yet BCG says the time savings is "not necessarily yielding value"
BCG via CFO Dive · Jul 9, 2026
8%
of CFOs have deployed AI-enabled solutions or agents at scale in key finance activities. Seventy percent are still in planning or pilot stage
Oliver Wyman / NYSE CFO Survey · 2026
36%
of organizations cite data quality as their greatest AI vulnerability — and their most cited opportunity
KPMG 2026 Global AI in Finance
34%
of finance organizations have already cut headcount, even as more than two-thirds of customers are paying slower than six months ago
Billtrust via CFO Dive · Jun 5, 2026

Lead story · Money & Rules

SEC.gov · Reporting Reform · Jul 6

The SEC Just Closed Comments on Ending Quarterly Reporting. What Controllers Need to Know.

On May 5, SEC Chairman Paul Atkins proposed optional semiannual reporting: companies could file one Form 10-S covering their first six months plus one annual 10-K, replacing three 10-Qs. The comment period closed July 6 after 60 days. SIFMA, Grant Thornton, and dozens of others responded — and the concerns weren't about the calendar. They were about the close.

Grant Thornton's comment letter flagged the absence of an assurance framework for quarterly earnings releases that semiannual filers would still be expected to provide voluntarily. That gap matters: if a company drops quarterly filings but still runs earnings calls and publishes releases, investors get numbers that are not subject to auditor review — and the auditor's involvement timing gets scrambled. PwC's audit committee memo raised a parallel issue: how DC&P certifications work for information that is furnished but not filed.

A final rule is unlikely before mid-2027. Calendar-year companies could first elect semiannual reporting for fiscal year 2028. If adopted, the compliance cost reduction is real — two fewer interim filings per year — but ICFR testing frequency, audit committee charter language, and trading-window policies all need rethinking.

The concept Disclosure cadence is the rhythm that forces your systems to close, reconcile, and certify on a schedule. Quarterly reporting is why your close is monthly-adjacent. Half-year reporting could stretch that rhythm — and expose process debt that the quarterly cycle was quietly disciplining.
For your desk Even if your company never elects semiannual filing, this proposal changes the conversation with your audit committee and external auditors about what "adequate interim controls" means. Watch the final rule for ICFR frequency guidance — that's the part that touches your work directly.
Read the SEC press release
The Finance Stack
CFO Dive · Finance Stack · Jun 26

Microsoft Rewired Copilot in Excel — Live ERP Data, Finance Skills, and a Partner Marketplace

Microsoft's June 26 update gave Copilot in Excel three new capabilities it previously lacked. First, reusable finance "skills": a controller can show Copilot a workflow once — say, the steps for a month-end variance commentary — save it as a skill, and have the same structured output on demand every close. Second, live ERP connectors to SAP, Dynamics 365 Finance, Workday Financial Management, and QuickBooks Online, letting a user pull AP aging or cash-flow data with a plain-language prompt without building Power Query connections. Third, a partner data marketplace: CB Insights, FactSet, Morningstar, PitchBook, and S&P Global are now callable from inside the spreadsheet.

Microsoft said its own finance organization uses Copilot across FP&A, accounting, tax, compliance, and treasury — and the features were pressure-tested internally before release. The move comes as Workday and other ERP vendors are building similar natural-language layers. The question for finance teams is not whether this category of tooling wins; it is which data flows to trust it with first.

The concept Skills in this context are saved prompt-plus-workflow templates — closer to a macro than a chat. The model still generates the output, but the structure, format, and steps are locked in. Think of it as a new-hire that has your SOPs memorized and applies them consistently — until the SOP changes.
For your desk The ERP connectors are where the real risk lives: live data going through an AI layer needs the same governance you would apply to any new reporting tool. Audit the permissions before you flip the connector on. Which accounts can Copilot see? Which can it write to? Get those answers in writing before your first earnings close with the feature enabled.
Read on CFO Dive
Models & Policy
WhiteHouse.gov / Financial Times · AI Governance · Jul 7

The White House Is About to Formalize How Frontier AI Gets Released — Here's What That Means

Executive Order 14409, signed June 2, established the legal skeleton: a voluntary framework under which developers of "covered frontier models" give the federal government up to 30 days of access before release to other trusted partners, while the NSA develops classified benchmarks to determine which models cross the threshold. By August 1, agencies must have that framework in place.

The Financial Times reported last week that the administration is in advanced talks with OpenAI, Anthropic, and Google on formal standards — an announcement could come any day. The practical effect is already visible: GPT-5.6 launched to roughly 20 government-vetted organizations before broader access, and Anthropic's Fable 5 spent 19 days under Commerce Department export controls after a jailbreak was reported before access was restored July 1. Pre-release government review is operational; the framework formalizes what is already happening on a deal-by-deal basis.

The concept Covered frontier model is the policy category that will determine which AI tools face government review before your vendor ships them. The threshold is classified — which means your AI vendors will know whether their next release is "covered" before you do. Add "model release delays due to government review" to your vendor-risk checklist alongside uptime SLAs.
For your desk If you rely on a frontier-model API for a close-critical workflow — a reconciliation, a variance commentary, a disclosure draft — a 30-day government review window on a major model update could delay access to new capabilities at the worst time. Build model-version pinning into your finance AI governance policy now.
Read EO 14409 on WhiteHouse.gov
Enterprise
CFO Dive · AI Adoption · Jul 9

CFOs Are Spending on AI. The Returns Are Not Keeping Up.

CFO Dive's July 9 deep-dive — the first of a two-part series — captures the paradox cleanly: 74% of finance frontline employees use AI at least several times a week, and 42% report saving a full workday or more per week. But BCG's finding is that the time savings is not necessarily yielding value. The productivity surplus is real; the business impact is not. KPMG's 2026 Global AI in Finance report found active AI use in finance more than doubled from 2024 to 2026, rising from 30% to approximately 75%, while data quality remains the most cited barrier and the most cited opportunity.

One CFO offered a counter-example. Pipedrive's CFO Regi Vengalil told CFO Dive his head of accounting built a close-management tool from scratch using AI. "My head of tax, my head of treasury can start taking it into their own hands," Vengalil said. The pattern in the data: organizations redesigning workflows before selecting technology report the strongest returns. Bolt-ons report the weakest.

For your desk The distinction Vengalil draws — team members building tools for their own workflow, not adopting a vendor's vision of what finance should look like — is worth pressure-testing on your own team. What would your close look like if your senior accountant had the same latitude?
Read on CFO Dive

AI, demystified

What Is RAG — and Why Does It Matter for Finance AI?

The Microsoft Copilot-in-Excel update — pulling live data from SAP or Workday at query time — is a form of what AI practitioners call RAG: Retrieval-Augmented Generation. The model doesn't know your ledger. It retrieves what it needs, then generates an answer. Here's the three-step mechanic.

1
Retrieve
The system fetches relevant data from a source — a database, an ERP, a document — at query time. The model sees only what it retrieved, not the whole dataset.
2
Augment
The retrieved data gets packaged with the user's prompt and handed to the language model. The model is grounded in your actual figures, not training data.
3
Generate
The model writes a response — a variance comment, a reconciliation summary, a question answer — using the live data as its source of truth.

The finance analogy: RAG is how you'd brief a smart new analyst. You don't hand them a copy of your brain; you hand them the specific report, the prior-period actuals, and the budget file. They work from those documents and answer your question. The difference from a standard LLM: without retrieval, the analyst is working from memory — and memory goes stale. With retrieval, they're reading the live file.

The risk to remember: RAG is only as accurate as what you retrieve. If the ERP connection pulls from a table that was last refreshed at 11 p.m. and your close is running at 7 a.m., the AI is confidently working from yesterday's data. Freshness, permissions, and source-of-record clarity are the governance variables that determine whether RAG helps or misleads.

Hype check

Exaggerated

"AI can cut your month-end close in half."

The evidence base for dramatic close-cycle compression is real but narrow. Some organizations running AI-assisted reconciliation and automated journal entry posting have reduced specific task times significantly. The CPA Practice Advisor piece from July 1 makes the mechanism clear: AI deployed after a properly designed workflow can compress it. AI deployed into a messy workflow can replicate its errors across multiple periods before anyone notices — faster and at greater scale.

The KPMG data is instructive: 36% of organizations cite data quality as their greatest AI vulnerability. If your close relies on 50 entities with fragmented charts of accounts — like the controller profiled in the Journal of Accountancy in April — the AI's speed advantage is blocked by the mapping problem. The "half the time" claim is achievable for teams with clean, well-governed data and redesigned workflows. It is marketing for everyone else.

Sources: CPA Practice Advisor, Jul 1 · Journal of Accountancy, Apr 2026

Do this week · Prompt of the week

Do this week
  • 15 minRead the SEC's semiannual reporting fact sheet and flag one ICFR implication specific to your organization — even if you're private. The governance questions it raises apply broadly.
  • 20 minPull up Copilot in Excel (if your org has M365 Copilot) and try one natural-language finance query. Document what it got right, what it hallucinated, and what data it couldn't reach. That three-point log is your governance baseline.
  • 10 minAsk your largest AI software vendor: "Is your next major model update subject to government review under EO 14409?" If they don't know, that's your answer about their AI risk posture.
Prompt of the week — Variance commentary
You are a senior controller at [Company]. Revenue came in at $[Actual] vs. a budget of $[Budget], a [Fav/Unfav] variance of $[Amount] ([%]). Here is the breakdown by segment: [Paste segment actuals vs. budget table] Draft 3–4 sentences of variance commentary suitable for the CFO's management-reporting package. Use plain English. Name the two largest drivers, quantify each, and flag one risk or uncertainty that warrants follow-up. Do not use the word "significant."

Number of the week

26%

of organizations have real-time visibility into the cost of running AI at scale — meaning nearly three in four are flying blind on their AI spend, even as budgets accelerate

CFO Dive · Jun 25, 2026

Off the record

Off the record

The Finance AI Gap That Nobody Is Measuring: The Cost of What Didn't Get Fixed

Every AI ROI conversation in finance starts from the same place: how much time did we save? It's the wrong starting point. The BCG data — 42% of finance employees saving a full workday weekly — ought to translate into something measurable at the business level. It mostly isn't, and the reason is rarely the technology.

What gets measured is hours recovered. What rarely gets measured is what those hours would have found if they had gone into the analysis the team didn't have time for before. The late reconciling item that became a material adjustment in Q4. The budget assumption that nobody stress-tested because the close took too long. The vendor contract that renewed on auto-pilot for three years. AI is supposed to buy the team time to catch those things. Most teams spend the time on the next close.

Real talk The most honest CFO quote of the month came from Pipedrive's Regi Vengalil: his head of accounting built a close-management tool himself. Nobody waited for a vendor, a project, or a committee. The organizations seeing real AI returns share that pattern — someone close to the work built something specific to the work. The gap between "we use AI" and "AI changed our outcomes" is almost always a workflow-redesign gap, not a technology gap.

Personality of the week

KP
Kevina Purmanund
CFO, the 1826 Group

Kevina Purmanund joined the FEI podcast this month at the FEI + MFM 2026 Financial Leaders Forum in San Antonio to talk about something few finance leaders are willing to say directly: she was afraid AI would replace her job. That candor makes what she built afterward more interesting than any "we deployed AI" press release.

Purmanund describes moving from that early anxiety to treating AI as a competitive edge — for herself and for the team she leads. The 1826 Group is a boutique advisory firm, which means there is no IT department to wait for, no change-management budget, and no vendor contract to hide behind. The AI adoption she describes is practitioner-built, close to the work, and directly tied to client outcomes. That's the pattern the BCG data says produces the strongest returns.

Why this week The CFO Dive series on AI adoption paradox ran the same week Purmanund's podcast dropped — and she is the rare voice on the practitioner side of that paradox. Not a consultant explaining what finance teams should do. A CFO explaining what she did when she was afraid.
Listen on FEI.org

Until next Monday — keep a name next to every number.