They didn’t need AI to tell them revenue fell in a segment.

Their analysts already catch that. What they wanted was the next step:

• So what are we going to do about it?

• What are the opportunities?

• How do we correct or redirect?

Very little AI in finance is built for that question yet. Here’s what this team asked for and what worries them. The company and the people stay anonymous.

Read on.

1. They got the data ready first

This team didn’t start with a chatbot.

They started with the data.

They had already built an ontology, a data layer that lets language models sit on top of their financial data. It covers the P&L and every slice of profitability: by customer, industry, product, volume and revenue type, and cost.

Most teams skip this step.

FP&A is no different. F

P&A still spends 47% of its time on low-value work like collecting and validating data.

AI can only read your P&L by customer, industry, and product if someone has built that layer. If they haven’t, the model is guessing.

2. FP&A wants AI at three levels

The team described what they want as an agent with their own FP&A discipline, working at three levels:

Below, we break down what makes AI output trustworthy enough for a board pack, the worries that stop FP&A teams from using it and what to do next. Paid readers also get an early invite to see finstory live before launch on Nov 18.

Read more

He started out consulting for Fortune 500 finance teams.

Then he co-founded CXO Software, a reporting platform for CFOs. Over 11 years he grew it to serve companies like Johnson Controls, Honeywell, Experian, and Facebook, and insightsoftware acquired it.

He stayed on at insightsoftware to lead M&A.

He helped some of the best CFO software companies grow. After that, he started Born Capital to invest in CFOTech founders working on FP&A, the close, and finance automation.

He sat in hundreds of demos. Almost every tool made the numbers faster. But very few helped a CFO explain what the numbers meant.

Storytelling is the most important skill in finance.

But financial storytelling is not easy.

So together with his co-founders, he started finstory.ai

Inside finstory lives Reeve, and it works differently.

Every figure comes straight from your connected systems, and you can click the basis button next to it to see exactly where it came from. The narrative is traceable too.

The part you will find most interesting is memory.

finstory is where your FP&A team’s memory lives, including past decisions and the context from your meetings. Each story builds on the last one, so Reeve gets better at telling yours and at helping you see what to do next.

Join Wouter Born and experience how AI is happening inside finance live on Wednesday, November 18, at 11 AM ET.

He will show how you can build financial stories in 1 minutes instead of 10 days and take your questions. AI in Finance readers who join get early access before the public launch on November 19.

To keep it practical, the event is open to paid subscribers only.

The link to join the live AI session is below:

Read more

Imagine it’s 10:52 pm on the night before the board meeting, and the deck goes to your CEO at seven. Your CEO has asked twice this quarter why AI isn’t writing the commentary yet, so tonight you let it.

You hand the assistant the opex file your analyst exported on day three of close, and a minute later it gives you clean, well-sourced commentary: opex is over budget, almost entirely because of a big marketing campaign accrual.

It even adds a polite caveat that someone should confirm the timing with the budget owner. It’s late, the sentence reads well, and you paste it in.

At 11:40 you open the ledger to grab one last number, and it doesn’t tie. The agency’s revised invoice arrived on day four. The campaign slipped to next quarter, almost nothing was billed this month, and on day five accounting reversed the accrual.

The overrun your AI explained so carefully no longer exists. Nothing in the paragraph was false about the file. Everything in it was false about your business.

AI didn’t get your number wrong.

It read an old Excel export, and that file went stale the moment someone hit export.

The Excel file did nothing wrong.

It was accurate at 9:14 on day three, and it stayed a picture of day three forever.

The ledger kept moving, the Excel file didn’t, and the AI read the Excel file. Every finance team has a version of this file: the CSV in the shared drive, the “final” tab in the forecast model, the numbers pasted into slide 41 of the board deck. Each one is a photograph of the business taken at a moment nobody wrote down.

None of this is new.

JPMorgan’s London Whale trade lost at least $6.2 billion, while the model meant to measure its risk ran on spreadsheets filled in “by a process of copying and pasting data from one spreadsheet to another.”

What’s new is that AI turns a stale Excel file into fluent, confident prose in seconds, and fluent prose is much harder to doubt than a cell in a spreadsheet.

The real cost of disconnected data

Time. Your best analysts spend more time proving the Excel export matches the ledger than explaining what the numbers mean, as the FP&A Trends Survey 2026 shows year after year.

Trust. Only one CFO in ten fully believes the data, and in RGP’s research that doubt is the single biggest thing standing between AI and a return on it. In BlackLine’s survey, the people closest to the numbers trusted them least.

Errors and decisions. Two-thirds of finance leaders in the same BlackLine survey said manual work exposes them to errors that could undermine decisions. That is the board-night scene, measured across an entire survey of finance leaders.

AI impact. Finance is not short of AI, but Gartner’s data shows only a sliver of finance teams see real impact from it. When the data underneath is scattered and stale, the AI on top inherits every one of those costs and adds speed and confidence to them.

Why AI fails on Excel data

Every AI answer in finance sits on a stack of layers.

At the bottom is the live ledger, with every transaction as it posts. Above it sit your roles and permissions, then your definitions (the chart of accounts, dimensions, and owners), then the question the CFO actually asks.

The AI’s answer sits at the very top, and most AI pilots start there.

An export cuts the pyramid just above the ledger. Everything above that line freezes at the moment of the export, and your permissions, dimensions, and late entries fall off with it. That’s why better prompts never fix a stale file: you’re polishing the top of a pyramid whose foundation stopped updating on day three.

Make NetSuite the live ledger your AI reads

The fix is not a better model.

It’s a shorter road between the question and the ledger, and that’s the gap NetSuite, which sponsors this series, is built to close.

The Excel chain is the road most teams drive today: ledger, export, spreadsheet, deck, and AI. Every arrow is a place where a late entry, a permission, or a dimension falls off. NetSuite removes the arrows. Financials and operations run on one unified data model, so the ledger the AI reads is the same ledger your team posts to, updated as each transaction lands. NetSuite’s founder, Evan Goldberg, put it plainly at SuiteWorld 2025:

“AI doesn’t just guess. It starts from the source because your data is unified.”

That unified source now has three AI doors, and each one changes something specific in a CFO’s week.

Ask Oracle came into the current NetSuite experience with release 2026.2, which began rolling out to US and Canadian customers on July 15, 2026. It answers questions inside NetSuite within each user’s existing roles and permissions, and its Show explanation button shows how it built the answer. The variance question on board night gets asked where the invoice actually landed.

The NetSuite AI Connector Service brings the same live data to the assistants your team already uses, such as Claude or ChatGPT, through the Model Context Protocol. It runs through a non-administrator role, so the AI sees only what that role can see, and every call is recorded in an execution log your auditor can review. Nobody exports a CSV to feed the chat window.

Intelligent Flux Analysis, also new in 2026.2, uses AI to draft explanations of why account balances changed between periods, right from the close. The first draft of the flux commentary starts from posted entries rather than from the day-three file.

Go back to board night with that setup. The question is the same, and so is the model. But the answer now includes the day-four invoice and the day-five reversal, with a link to the entry that changed the story, and you go to bed at eleven.

A live connection makes your AI exactly as good as your ledger. If your chart of accounts has three overlapping travel accounts, or subsidiaries were mapped in a hurry after an acquisition, the AI will read that mess accurately and in real time. Budget time for that cleanup before you roll AI out across the close.

Live Ledger or Excel Export?

Screenshot this and take it into your next close review.

Answer it for every AI workflow your team runs, from variance commentary to board-pack drafts, and ask your team rather than your vendor.

If three or more land in the Excel column, the problem sits upstream of the model, and that’s the cheaper fix.

Want to go further?

NetSuite's 2026 AI-Powered Financial Analysis Handbook walks through how finance teams are setting up AI on live data.

The Bottom Line

AI keeps failing in finance because we hand it numbers that went stale the moment someone hit export, and then we blame the reader for what was on the page.

A smarter model can’t read an invoice that isn’t in the file.

Before you buy another AI tool, find out what your current one is actually reading. Most of you will find it’s an Excel export, and the fix is a live ledger, not a new model.

Team at finstory

finstory.ai builds reporting tools that connect to ERPs, including NetSuite.

Find us on LinkedIn | finstory.ai

Someone pulls an Excel export from the ERP, pastes it into a model, and asks AI to explain why opex is over budget. The answer sounds confident. Nobody in the room trusts it, so your team spends the evening rebuilding the numbers by hand before the board deck goes out.

That’s where most finance teams are with AI right now.

The tools are everywhere, but the results aren’t.

Plenty of large companies say they are scaling AI agents, yet only 6% get at least 5% of their profit from AI.

The model isn’t the problem.

The setup around it is.

The finstory.ai team looks at it as an operating system with 5 layers, each one built on the layer below it, with checks wrapped around all of them. You don’t need to build it all at once.

You need to build it in the right order, and this edition shows you how.

Expect AI to get half the job right

Before you build anything, be honest about what today’s tools can do.

On a benchmark of 927 entry-level finance analyst tasks, the best model scores 65.4% with partial credit and 50.9% when every part of the answer has to be right.

It handles about 85% of earnings questions well.

But it scores only 34.5% on financial modeling and 36.4% on precedent transactions.

Finance doesn’t give partial credit.

A close that is 95% right is wrong, and a board deck with one broken link is a broken board deck. Your people aren’t perfect either, since 91% of real spreadsheets checked in field audits had material errors.

So the goal isn’t to trust AI or to ban it.

The goal is a system where every number has a source, a reviewer, and an owner.

Keep your data where it lives

Read more

Anthropic released Claude Dashboards in beta on October 8, 2026, alongside Claude Motion.

You connect a data platform such as Amazon Redshift, BigQuery, ClickHouse, Databricks, or Snowflake, or another connected app such as Salesforce, and ask a question in plain language. Claude writes an SQL query for each chart, runs it against your data source, and builds the dashboard.

For finance teams, the key detail is that every chart shows the query behind it and when its data was last refreshed. Anthropic positions it for quick, exploratory questions and says it does not replace your BI tools; for deeper analysis, you send the dashboard on to tools such as Sigma, Omni, or Grafana.

It is in beta on Pro, Max, Team, and Enterprise plans.

On Enterprise it is off until an owner turns it on in Organization settings > Artifacts, and dashboards start private until shared.

Read on.

I built a finance control tower from one prompt

I used Google’s public e-commerce dataset in BigQuery, thelook_ecommerce, which holds more than 3.3 million synthetic records covering orders, products, inventory, customers, and returns.

With one prompt, asking Claude to act as the CFO of this business, I got this dashboard below.

The TheLook Finance Operations dashboard built from Prompt 1

At first glance, the business looked great.

Over the last 12 months, net merchandise sales were $2.14M, up 77.3%, with a merchandise margin of $1.11M at 51.9% on 44,861 orders. Claude built five tabs with filters by period, department, and category, and it added a Method & Controls tab that runs 11 data checks every time the dashboard loads.

Then I noticed how much inventory sat on the same page, and the growth story started to feel too easy.

Push Claude to find where the cash is stuck

So I asked the question I would ask any finance team after a year like that: sales are growing, but is the business actually getting healthier?

I half expected Claude to cheer the 77%. Instead, its answer started with one word:

“Partly. Turnover is improving because the cost of goods sold grew faster than stock. But most of the stock isn’t moving, and that looks more like a problem with the stock records than a buying problem.”

The rebuilt Inventory & working capital tab after Prompt 2

What surprised me most was what Claude did next. It didn’t hand me an $8.62M write-down. It tagged the finding “Data check first,” which means it believes this is more likely a records problem than a real loss and that it should be verified before anything gets booked.

This is exactly what I would want from a good analyst, and it was the moment I started seeing this as a tool a controller can actually use.

Open the LIVE Dashboard

Below, I walk you through exactly how I built it, with both prompts word for word.

Build the Finance Operations dashboard in 10 minutes

Connect Claude to your data and COPY/PASTE PROMPTS

Read more

Here is the Born2Cycle September 2026 Board Story.

Born2Cycle is a fictional bicycle retailer, finstory’s demo company, but the story runs on a full set of books: P&L, channels, cash, inventory, KPIs, and forecast.

Look at how it opens.

Source: finstory.ai Born2Cycle September story

“Sales ahead. Profit is not.”

Year…

Read more

If you run finance, you know the night.

The close tied out two days ago. It’s now 11:48 pm, you’re on slide 41, and you’re still writing commentary from memory. Your CEO wants the narrative.

Your board will flip through variance tables and ask the one question the pack never answers: so what?

The numbers are right.

The story arrives la…

Read more

You can close the books. The hard part is turning the numbers into a story the board actually understands, with every figure traceable back to a source.

That is what the session was about, and we built it live in Claude.

How to build financial decks with Claude (LIVE)

Paulo Tabarelli Batista runs finance at RAR Holding in Dubai today, with the close and the board pack and the follow-ups that come with it. Over 15 years across multinationals, private equity-backed businesses, and family-owned groups.

And Wouter Born is the author of the top finance newsletter AI CFO Office. He has spent the last 30 years at the intersection of finance, tech, and AI, helping CFOs move beyond spreadsheets to become AI-native key persons of influence. He’s also the founder of finstory.ai

Watch the full recording below

Read more

The idea came from yesterday’s live session with Wouter Born and Paulo Tabarelli Batista.

Paulo Tabarelli Batista runs finance at RAR Holding in Dubai today, with the close and the board pack and the follow-ups that come with it. Over 15 years across multinationals, private equity-backed businesses, and family-owned groups. Transformations in FMCG, retail, e-commerce, telecom, and petrochemicals. Operations across five continents. Portfolios above $1B.

And Wouter Born is the author of the top finance newsletter AI CFO Office. He has spent the last 30 years at the intersection of finance, tech, and AI, helping CFOs move beyond spreadsheets to become AI-native key persons of influence. He’s also the founder of finstory.ai

You’ll get the full setup, the board pack (.skill), and all the prompts with files you can upload to Claude and build your own financial deck and impress your CEO.

Let’s dive in.

Building a storytelling financial deck with Claude

Step 1: Install the board pack skill

This skill is really advanced.

It has templates inside the skill with Python code files that know how to build the reporting model.

Download the skill below, then copy/paste the prompt to build a storytelling financial deck.

Read more

You can close the books. The hard part is turning the numbers into a story the board actually understands, with every figure traceable back to a source.

That is what this session is about, and we are building it live in Claude.

How to build financial decks with Claude (LIVE)

Read more

You can close the books. The hard part is turning the numbers into a story the board actually understands, with every figure traceable back to a source.

That is what this session is about, and we are building it live in Claude.

How to build financial decks with Claude (LIVE)

Paulo Tabarelli Batista runs finance at RAR Holding in Dubai today, with the close and the board pack and the follow-ups that come with it. Over 15 years across multinationals, private equity-backed businesses, and family-owned groups. Transformations in FMCG, retail, e-commerce, telecom, and petrochemicals. Operations across five continents. Portfolios above $1B.

He is not demonstrating a product.

He is showing you the environment he actually works in, including the parts that break.

Joost Vogelezang is a COO who has built, scaled, and exited operations across SaaS, finance, and tech for 30 years. He was a VP at insightsoftware, running a 250-person team across 30,000 customers and 20-plus acquisitions, and co-founded two SaaS companies, both of which exited.

What Paulo and Joost will cover

• Building financial decks with Claude.

• How to make your finance function AI-native

• What to deploy this quarter and what to wait on

• Where Claude saves real time and where it does not

• Skills for the workflows that compound every month

• Projects for persistent work with the right files in context

• Connectors for live data from your ERP, BI, and source systems

Who this is for

CFOs, VPs of Finance, FP&A leaders, controllers, and the operators who sit next to them. Anyone responsible for turning finance data into a story a board will act on.

What to expect

A live, hands-on session with Joost and Paulo building financial storytelling workflows in Claude in real time. Open Q&A throughout. Honest discussion of what works and what breaks.

Read more

In nearly every conversation with finance teams, we heard the same sentence:

“If only this could plug directly into NetSuite and stay up to date.”

We agreed.

finstory now integrates directly with NetSuite, bringing your ERP data into finstory in near real time and turning it into structured financial…

Read more

You’re in NetSuite, late again, exporting reports into Excel.

The numbers are correct, but the story isn’t there yet. So you rebuild logic, check formulas, reconcile versions, and write commentary by hand.

If you’ve run finance on NetSuite, this feeling is familiar.

NetSuite records financial reality for 40,000+ companies.

But the p…

Read more

This week we asked a harder question.

Can the same bot build the financial model the board deck sits on?

A working 3-statement model. Assumptions. Scenarios.

Then a board pack that traces every claim back to those numbers.

We ran it on Born2Cycle again. Same company. Same May emails. Same inventory spike. By morning we had an Excel model, a six-slide board deck, and a short list of decisions the board still owns.

ChatGPT and Claude are still tools you operate.

Grok Bot is still a worker you delegate to.

The difference this week is the artifact. The bot did not only refresh commentary. It built the engine under the story.

Let’s dive in.

How to build a board model and deck as a CFO with Grok Bot

You do not prompt your way through a close.

You hand the bot a job with rules, source files, and a folder structure. You tell it what it must never invent. You review the output the way you review an FP&A analyst’s first draft.

The CFO moves from builder to approver.

From assembling the model to stress-testing the story.

Download the Skill files (try this yourself)

Download the zip. Open the Excel model. Open the HTML deck in your browser. Then run the steps below with Grok Bot. Everything you need is here:

Read more

FP&A and Controllers can produce reports. You can close the books. The hard part is turning the numbers into a story the board actually understands, with every figure traceable back to a source.

That is what this session is about, and we are building it live in Claude.

How to build financial decks with Claude (LIVE)

Paulo Tabarelli Batista runs finance at RAR Holding in Dubai today, with the close and the board pack and the follow-ups that come with it. Over 15 years across multinationals, private equity-backed businesses, and family-owned groups. Transformations in FMCG, retail, e-commerce, telecom, and petrochemicals. Operations across five continents. Portfolios above $1B.

He is not demonstrating a product.

He is showing you the environment he actually works in, including the parts that break.

Joost Vogelezang is a COO who has built, scaled, and exited operations across SaaS, finance, and tech for 30 years. He was a VP at insightsoftware, running a 250-person team across 30,000 customers and 20-plus acquisitions, and co-founded two SaaS companies, both of which exited.

What Paulo and Joost will cover

• Building financial decks with Claude.

• How to make your finance function AI-native

• What to deploy this quarter and what to wait on

• Where Claude saves real time and where it does not

• Skills for the workflows that compound every month

• Projects for persistent work with the right files in context

• Connectors for live data from your ERP, BI, and source systems

Who this is for

CFOs, VPs of Finance, FP&A leaders, controllers, and the operators who sit next to them. Anyone responsible for turning finance data into a story a board will act on.

What to expect

A live, hands-on session with Joost and Paulo building financial storytelling workflows in Claude in real time. Open Q&A throughout. Honest discussion of what works and what breaks.

Paid monthly and annual members get the recap through the newsletter and the live recording. The live session and the recording are exclusive to Premium members.

If you want the live experience, upgrade to Premium Membership and register below.

Read more

Grok Bot is not a chatbot. Each bot gets a cloud computer.

It can sign into your favorite tools and do whatever a normal human does. It can log in to your Salesforce, click, navigate, and fill out forms. But it doesn’t shut down when you turn off your laptop. Instead, it keeps working for you.

SpaceX and xAI use Grok Bot internally.

You can give the CFO any finance tasks and personal tasks.

All you have to do is just describe the task.

Prompting is the worst way to use AI.

The best way to use AI is to delegate your work.

Elon has been tweeting about it nonstop. He told one user to connect it to his bank account and promised xAI would cover any mistakes.

That is a fundamentally different working model than anything we have tested before. ChatGPT and Claude are tools you operate. Grok Bot is a worker you delegate to.

Grok Bot can do something no AI tool has done for finance before.

You give it a job, connect it to your financial data and your email, and it runs on a schedule without you starting it.

Every night it reads new emails, pulls the latest numbers from your ERP, cross-references the two, updates your reporting deck where the evidence supports a change, and sends you questions where it does not have enough information to conclude on its own.

Let’s dive in.

Access to Grok Bot

Download the desktop app here: x.ai/bot

You’ll see 3 tiers.

Sign in with your account.

Setup takes less than 5 minutes.

On your top right you’d see an actual PC desktop you can open.

That’s where the bot lives, operates, and stores everything.

That’s how the bot works 24/7.

Let’s move on and build our first Autonomous Financial Reporting Analyst.

How to Build an Autonomous Financial Reporting Analyst (Bot)

Read more

But do they trust it? Nope, sorry.

Only your finance team trusts the forecast.

So why your CEO and the Board are skeptical.

The reason is simple.

Your data is a mess.

You don’t have a single source of truth.

It’s more like garbage in, garbage out.

We are live in uncertain times. Revenue is not cash.

The assumptions are lost. Scenarios are terrible. Your trail is weak. You can’t connect back to the real truth. When a number moves, you don’t know why it moved.

But you’re lucky.

We have Claude that we can use to figure it all out faster.

I used Claude to build a full 13-week cash flow forecast.

Do you know that you can keep the Excel model as the source of truth and yet build an interactive storytelling dashboard.

Every finance person must use Claude and make your life easier.

The great thing is you can stress-test.

It’s just not a pretty dashboard.

This is the dashboard that tells you the truth about your numbers.

While building this, I had a lot of fun… The dashboard gives you a complete overview of the weak spots. Exactly what a CFO need to see before making a hard call.

Okay, enough talking.

In today’s letter I’ll show you how to build the cash flow forecast with Claude.

I used the cash flow model builder skill (.md) and Cash flow scenario dashboard skill (.md)

You can download both of these skill files and all the prompts.

This is the real AI in finance work.

Let’s dive in.

How to build a full 13-week cash flow forecast with Claude

Step 1: Install the cash flow model builder skill

You’ll get 2 skills and test files.

1. Cash flow model builder skill

2. Cash flow scenario dashboard skill

Download the skills below, then copy/paste the prompt to build a cash flow model.

Read more

But it still runs on a spreadsheet someone built in 2019.

• Your planning tool exports to Excel.

• Your team edits the Excel.

• Someone uploads it back into the tool.

• Nobody knows which version is current.

• The assumptions live in someone’s head and the audit trail lives nowhere.

You are paying enterprise prices for a spreadsheet with a login screen.

So we asked ourselves a question.

What if we could build a working FP&A system using nothing but Claude, three downloadable Skills, and a simple Excel workbook?

Not a demo or a concept.

Something you can actually run on your own Marketing OPEX data this afternoon and walk away with a versioned, assumption-documented budget.

The interesting part is not the budget itself.

It is how Claude builds it. Instead of extrapolating last year’s numbers or blindly trending forward, Claude actually reasons about each account…

• is this headcount-driven, contractual, seasonal, discretionary? and asks you the planning questions a strong FP&A analyst would ask on their first day before it calculates a single number.

Read on.

Doing FP&A work with Claude

Step 1: Download the 3 FP&A skill files below

Read more

Her team has access to the most advanced AI tools on the planet. They build them. They ship them. They sit in the same building as the engineers who make ChatGPT, Codex, and every model the rest of us pay to use.

Read more

The answer is yes. And I’m going to show you exactly how.

I built a full monthly planning system for a real SaaS company inside Claude. A planning model with customer metrics, headcount planning, full financials, and a balance sheet.

Wait until you see the dashboard at the end. Because Claude handled the entire reforecast in the same Project, it had full context for every chart, every narrative, and every scenario. It’s the most comprehensive planning dashboard I’ve built so far

I took CoolDashboards, LLC monthly planning model, uploaded it to a Claude Project, and ran the full FP&A cycle. Actuals loaded. Forecast updated. Assumptions pressure-tested. Scenarios built. Dashboard created.

In today’s edition, you’ll see how to set up a Claude Project as your persistent FP&A workspace, load a real planning model and new actuals, spot anomalies against the plan, update the forecast, run scenarios with sensitivity analysis, and save the entire workflow so next month takes half the time.

No code, no API or any technical skills needed

Let’s dive in.

How to do financial modeling with Claude and build a storytelling board deck

Read more