Adam Tzagournis, CPA

• Founder of a new AI stealth startup

• I'm a CPA (Certified Public Accountant) and a full-stack software developer

• I built AI financial projections software

• I've worked with companies for over 10 years to help them raise money (and not run out of it)

• I write about business finance, cash runway, modeling, metrics, FP&A, equity ownership, programming, and more.

Founder & CEO

Business finance simplified. FlowCog Canvas is a free tool to help you understand your company's past and future financial health.

I lead the AI practice at VAi Consulting, where I guide CFOs and finance executives on GenAI strategy, automation, data analytics, and the full spectrum of AI implementation. My work spans AI readiness assessments, workflow automation, and driving change across FP&A, audit, and risk functions. I help shape our service offerings and collaborate with a national network of consultants to deliver impactful results for clients in healthcare, manufacturing, technology, and business services.

Prior to consulting, I spent 20 years as a CFO in private equity-backed companies, along with interim and advisory roles supporting public companies. I’ve built finance functions from the ground up, led M&A transactions on both the buy and sell side, and navigated the audit and controls complexities of institutional ownership. This hands-on experience shapes how I approach AI in finance. CFOs need more than just the potential of new technology. They need clear answers on how it will impact their control environment and the messaging to audit committees.

I’m currently working on my third book on AI in Finance, The AI-Ready CFO, which is due out in 2026 from Wiley Finance. My previous books include Deep Finance and AI Mastery for Finance Professionals.

I teach in Duke University’s CFO Program and for the AICPA and Corporate Finance Institute. I also sit on advisory boards for Preql, an AI data platform, the AI Leaders Council, and the Crews School of Accountancy at the University of Memphis. 

I’m the host of the FP&A Today podcast and a U.S. Navy veteran.

Glenn Hopper

Datarails just released FinanceOS, and the thesis behind it is worth paying attention to. Gartner's latest survey puts AI adoption in finance at 59%, up one point from last year. 91% of teams report low impact. The models aren't the problem. The data is.

Your analysts can draft a forecast in Claude or Copilot in seconds. The output only ties out if the data feeding it is consolidated, governed, and refreshable. Most of the time it isn't. Numbers lose their audit trail the moment they leave the source system.

FinanceOS sits underneath those AI tools as a governed execution layer. It connects to 400+ sources, refreshes in seconds, routes data through role-based permissions, and logs everything for SOC 2/GDPR/ISO 27001. MCP handles the connection to Claude, ChatGPT, Gamma, Cursor, and others. Finance teams can be live in days with usage-based pricing.

The interesting move is the open ecosystem play. Datarails is building the infrastructure layer and letting teams plug in whatever AI tools fit. That accelerates adoption faster than any closed platform can.

Building Practical AI Solutions for Finance | Head of AI at VAi

The same week Anthropic  released Claude Sonnet 4.6, Fortune ran a story about HPE's CFO Marie Myers  deploying "Alfred," an agentic AI that eliminated her team's 100-slide Monday meeting.

Myers framed it as "CFO as steward of AI." Finance transformation as the proof point that opens every other door in the enterprise.

That framing is right. Finance has clean data governance, established controls, and clear metrics for success. When you automate the monthly close or replace a 100-slide deck with an agent that answers questions in real time, the results are measurable. You can show the board exactly what happened and what it saved. No ambiguity. No six-month wait for ROI.

That makes finance the best proving ground for enterprise AI adoption. Every other department benefits when the CFO goes first.

Meanwhile, the models keep getting better. Sonnet 4.6 is faster, cheaper, and more accurate on structured data tasks than anything available six months ago. The tools to do what Myers did at HPE are more accessible than they've ever been.

What you won't find in the release notes: knowing where to start, what controls to build around the output, and how to get a finance team to trust it. That part takes deployment experience.

At VAI Consulting  we are building custom agentic workflows for close automation, variance commentary, and FP&A.

What Myers is describing at HPE scale is achievable at mid-market too. The architecture patterns are similar. The change management is harder. Getting people to stop building slides and start asking questions of a live system requires more than a software rollout.

The tooling is ready. The hard part is the work between the demo and the first month someone trusts the output enough to skip the deck.

No alternative text description for this image

I dove deep this week with a head-to-head comparison of the two most powerful AI models in the world for corporate finance workflows.

🥊 ChatGPT 5.4. Claude Opus 4.6. Head-to-head with pubco financial data (10-K and historical financials). Chat interface and Excel plug-ins. Scored across seven dimensions.

(For those wondering, no Microsoft Copilot for Finance didn't make the cut.)

OpenAI shipped a ChatGPT for Excel add-in with live data connections to Moody's Corporation's, S&P Global, and FactSet. Anthropic launched pre-built finance agents with PwC as an implementation partner. Both moved into Excel in the same 30-day window. On February 3, traders wiped out $285 billion in SaaS market cap in a single session.

Meanwhile, 72% of CFOs say they use AI tools. Only 14% can prove ROI. The barrier isn't model capability. It's workflow fit, token economics, and governance.

This week's Deep Finance Dispatch covers the full frontier landscape: what shipped, what the benchmarks actually mean, and why model selection for finance teams is now a governance decision.

The Pro edition has the full bake-off. Both models. Two rounds. Real scores.
One of them surprised me.

Link in comments.

I’ll admit it. I’ve been the guy railing against all these so-called AI agents flooding the marketplace. Repeatedly ranting about how *they’re coming* for finance workflows … but they’re not here … yet.

That changed this week.

Goldman Sachs put Claude agents on live accounting tasks. Ramp shipped an agent that codes transactions and syncs to your GL. NetSuite launched a full autonomous close suite. Hewlett Packard Enterprise’s CFO publicly staked her 2026 priorities on agentic AI.

Four big shifts in one week. The gap between “coming soon” and “shipping now” just closed.

I’ve been writing about this shift for two years. Telling finance leaders to prepare. Build the governance. Get the data ready. Learn the mechanics.
The preparation phase is over.

This week’s Deep Finance Dispatch breaks down what shipped, what it actually does, and what finance leaders should do about it right now. Three takeaways and a decision framework for choosing your path.

Free to read. Link in comments.

🔊 Upcoming webinar:

The AI-Ready Finance Model

Here's a pattern I keep seeing in finance orgs: Leadership greenlights an AI initiative in Q1. By Q3, adoption is stuck at 20%. By Q4, someone suggests "maybe we need a different tool."

The tool was fine. The adoption plan was missing.

On March 26 I'm running a free masterclass with Ramp where I'll break down a 90-day playbook for getting AI tools actually embedded in finance workflows. We'll cover sequencing, ownership, the trust problem with skeptical team members, and what a realistic timeline looks like when you're also running month-end close.

One hour. Tactical. Worth your lunch break.

https://lnkd.in/gJqcqpDh

Most finance teams don't need a giant AI transformation plan this quarter.

They need one clean workflow where AI does real work and controls stay intact.

Start with a month-end lane that is painful and repetitive (a common one is variance commentary prep):
• Agent pulls trial balance deltas and budget comparisons.
• Agent drafts first-pass explanations by account and cost center.
• Rules engine flags items above threshold for controller review.
• Reviewer edits, approves, and routes final commentary to the reporting pack.
• System logs source data, prompt version, edits, and approval timestamp.

Now you have speed and evidence.
The workflow ties out. Ownership is clear. Review is explicit.

This is how agentic AI earns trust in finance teams.
Small lane. Tight controls. Clear handoff points.

After that, expand to reconciliations, flux analysis, and forecast narrative drafts using the same approval architecture.

Want the exact swimlane template? Reply ‘SWIMLANE’ and I’ll share it.

AI in Finance Workflow

If your AI ROI case is still based on generic productivity estimates, expect pushback.

Finance leaders sign off when the math ties to operating metrics they already trust.

For AI in finance, I track four measures first:

• Cycle time reduction for close tasks and reporting prep
• Error/rework rate before and after AI-assisted steps
• Escalation volume to controller or accounting policy teams
• Reviewer load per output batch

Then I map those to economic value:

• Hours returned to higher-value analysis
• Lower rework cost
• Fewer late-close risks
• Better capacity during forecast and board cycles

This gives you a grounded ROI conversation with CFO, CAO, and audit stakeholders.

No heroic assumptions required.

If your pilot can't produce these four measures by week two, fix instrumentation before expanding scope.

Finance doesn't need more AI slides.
It needs evidence that survives review.

OK, so this isn't my usual "AI in finance" kind of post ... but I've been running an experiment on robocfo.ai and found something worth sharing.

Over the past couple of weeks, I've been building a bunch of AI-friendly infrastructure under the hood. Inspired by (blatantly stealing from?) Moltbook, I built a guest book for bots, and added llms.txt and a few other modifications. 

I haven't seen any real change in traffic yet (still fine-tuning), but I ran an audit with Claude Code the other night and found that the boiler plate Terms of Service and privacy policy were telling the bots they're not welcome.

I'm sure plenty of sites owners don't want droids combing through their websites, stealing content, and rerouting web traffic, but since I'm in the space it seemed like a net positive to me.

I'd love to hear what everyone is doing with their own websites to prepare for the future of the internet.

Google search traffic to publisher websites dropped by a third in 2025, according to Chartbeat data. AI Overviews now appear in roughly a quarter of a

Open article: Your Website Has Two Front Doors. One of Them Is Probably Locked. by Glenn Hopper • 9 min read

Your Website Has Two Front Doors. One of Them Is Probably Locked.

Finance teams are asking the wrong first question about AI disclosure.

The first question isn't what to say.

It's what you can prove.

For any AI-related claim in your 10-Q, board deck, or investor narrative, I use a simple control stack:

1. Source: Which system produced the data point?
2. Transformation: What logic, prompt, or model touched it?
3. Reviewer: Who validated the output and when?
4. Materiality gate: What threshold routes the item to legal, controllership, or audit committee review?
5. Evidence: Where is the log, version history, and approval trail stored?

This is the difference between confident disclosure and cleanup work three quarters later.

If your team is piloting AI in planning, close, or narrative reporting, disclosure discipline starts now, inside workflow design. Tie each output to an owner, a control, and an audit trail. Then scale.

Comment ‘CONTROL MAP’ and I’ll send the 1-page checklist I use with finance teams.

If you're evaluating AI-native ERPs, the first question isn't which platform. It's whether you replace your current system or augment it.

That single decision changes everything: your migration risk, your audit timeline, your team's capacity for change, and what your board needs to approve. Seven startups have raised over $500M collectively to bet on different answers to that question. Four are building full-stack replacements. Three sit on top of what you already run.

I broke down all seven with funding data, compliance posture, migration reality checks, integration depth, and 10 due diligence questions every CFO should bring to a demo. It's the guide I wish existed when clients started asking me about these platforms six months ago.

Open article: AI-Native ERP Landscape for CFOs | 2026 Guide | RoboCFO.ai by robocfo.ai

RoboCFO.ai - AI Implementation for Finance Teams

AI-Native ERP Landscape for CFOs | 2026 Guide | RoboCFO.ai

The AI arms race continues with the latest behemoth models ChatGPT 5.4  Claude Opus 4.6 both changing the way we work.

We put them head-to-head with the kind of work your FP&A team does every week. Then we tested both Excel plug-ins.

▪️ One of them hallucinated a number that would have been embarrassing in front of a board.
▪️ One of them built a forecast you could trace cell by cell. The other pasted static values and hoped nobody would notice.
▪️ One understood the actual business drivers. The other gave generic assumptions you'd get from a first-year analyst who didn't read the filing.

The results weren't clean. Neither model ran the table ... and the tool that won Round 1 didn't necessarily win Round 2.

This week's Deep Finance Dispatch Pro includes results, screenshots, prompts, and everything you need to run the same test on your own data.

Open article: Which AI Copilot Actually Delivers? (Pro Version) by open.substack.com

Which AI Copilot Actually Delivers? (Pro Version)

taplio

The Battle

Adam Tzagournis, CPA

Glenn Hopper

Followers

Posts

Engagements

Likes

Performance ScoreGlenn Hopper's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Adam Tzagournis, CPA

Glenn Hopper

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Glenn Hopper's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreGlenn Hopper's LinkedIn Performance