We’re Ellen and Simone. After 36 years in finance, we’re ready to share what textbooks won’t tell you.

💛 Welcome to CFO Playbook – your practical finance insights delivered bi-weekly. The full read will take approximately 5 minutes. Like what you see? Share it! Use the button below.

READ OF THE WEEK

Dozens of projects. Some shipped. Some got stuck for a year. The difference was rarely the software.

We interviewed two people who have done this many times.

Fabienne Doerig, founded carousel, an agency that builds finance functions giving PE-backed companies real-time control of their margins. Previously she served as  acting COO and led the finance transformation at wefox. Before that she was CFO and COO at Starmind, after seven years in audit and M&A at EY.

Sebastian Walther co-founded ValueWorks.ai (a platform that builds driver trees and brings structure to messy finance data). Before that he was an Associate Partner at McKinsey and researched corporate data quality at the University of St. Gallen.

Both advise companies on this exact problem today. Mostly PE-backed mid-market firms and family-owned companies, the kind with legacy structures. The same lessons hold for VC-backed companies too.

One pattern shows up every time. The tech is usually the easy part. The data and the people decide if it works.

1. The bottleneck is rarely the tech

Pulling data out is never the hard part. There is always at least an Excel export. With MCP, you can now plug almost any system into Claude, or the LLM of your choice, and pull the data straight out. Sebastian has never met a system he couldn’t get data from.

The real blocker is usually the data itself. A project can lose a whole year because the data just isn’t there inside the company, and no partner can fix that from outside. Say your cost data has no shared ID to link each cost to a project. Then you can’t see profit per project, and AI can’t fill that hole.

There is one case where the tech really does bite: old, fragmented systems. Five legacy ERPs that don’t talk to each other. The bigger the group, the worse it gets. Then you have a real data project, and it takes time.

AI adds real value here. But it can’t rescue bad foundations. The golden rule: if the data structure isn’t there, or there is no common definition across the business, AI won’t help either.

2. Agree on a fixed set of KPIs first.

In many companies, every manager counts margin their own way. One team Fabienne worked with went from 15 agreed KPIs to 30. All after one management change. This KPI chaos is the top reason projects fall apart.

The fix starts at the top. The bosses must agree on what matters. The CFO must own the logic. How do we define a customer? How do we count margin? Who is allowed to change it? Lock this down before anyone builds a dashboard.

3.  Bring your people along

This is where most projects really break. With the people.

Most teams underestimate change management. They plan the systems and forget the humans. That is why projects fail.

Fabienne sees three feelings underneath:

First, the sense that nothing can change. „The numbers can't be fixed anyway, it always runs the same way." Old structures that no longer fit, and the belief they're stuck with them.

Second, fear that the job itself is changing. A controller who suddenly has to work with AI and doesn't know where to start.

Third, AI making it harder, not easier. It speeds everything up, and that amplifies the uncertainty people already feel.

You can't skip these people. You have to win them over. Start small. Don't drop 60 new fields on them. Pick one. Show a quick win. Tools like Claude help here. They bring structure and take away the fear.

And watch for management changes. A new CEO or CFO can throw out everything you built. The KPIs, the definitions, the discipline, gone. Then you start over. Months of work, undone by one new hire at the top.

4. Let AI help, but keep your numbers traceable for years

AI can do a lot here today. It reads your data, writes commentary, flags odd numbers, and answers questions in plain language. Use it.

But there is a catch. AI is probabilistic. It predicts the most likely answer, so ask the same question twice and you can get two different answers. A calculator is the opposite: deterministic, same input, same result, every time. That difference is obvious to finance people, less so to everyone else. Two answers are fine for writing text. They are a problem for a number that goes to the board, and one you still need to explain in two or three years.

A PE investor will ask: were these values changed after the fact? For a due diligence or an audit, you must show the raw data, the calculation logic, and every change since. An LLM that „just answers“ can’t do that. It has no record of how it got there.

So the setup that works keeps two jobs apart. The AI (probabilistic) writes the story. A fixed engine (deterministic) does the math. A driver tree shows where each number comes from. Store the raw data and the logic for years, not just the final report.

One caveat. This space moves fast. What AI can’t do safely today, it may handle next month. Treat all of this as a snapshot.

🔎 CFO Watchlist

The Big Four bet on Claude
PwC massively expanded its alliance with Anthropic and is building its new "Office of the CFO" finance practice entirely on Claude, training and certifying 30,000 staff first before a wider rollout to its ~364,000 people. PwC is positioning itself as the first professional-services firm to build an at-scale finance function around a single foundation model. That puts three of the four Big Four on Claude (PwC, Deloitte, KPMG); only EY went the other way, committing over $1 billion to roll out Microsoft's Copilot across its workforce.

 👉 The Big Four AI line-up · Fortune: inside the race · Anthropic: PwC partnership

Making legacy systems AI-ready
The ex-Palantir team behind Conduct raised €51M to make complex enterprise systems (SAP, Salesforce, Oracle and more) AI-ready. SAP itself joined the round as a strategic investor and named Conduct a transformation partner for SAP Cloud ERP, a strong signal that the incumbent sees "AI-ready legacy" as the next real battleground. 

👉 EU-Startups: Conduct raises €51M

Funny LinkedIn meme

🌐 Finance Collective DACH

Your go-to CFO Network (by Simone)

CLOSING REMARKS

Thanks for reading 💛

Send your feedback, suggestions, or requests to feature something in future editions to [email protected]. We’d love to include your input.

Follow us on LinkedIn for more updates!

CFO Playbook reflects our personal opinions, not professional advice.