Hello Reader,

Let me ask you something: How many hours did your team spend on last month's forecast?

If you're like most of the finance teams I’ve coached, the answer is between "too many" and "way too many." You chase down departments. You reconcile spreadsheets that do not match. You ask the same questions you asked last month. And by the time you're done, you're already behind on this month's close.

But here is the good news. Companies like Coca-Cola and Microsoft completely flipped this process. And today you can learn from them.

They went from spending one full week every month on forecasting to just 2-3 hours per team.

And no, they didn't hire 50 more analysts.

They did something smarter. Something you can steal for your team, whether you are at a $50M company or a $5B one.

The Forecasting Treadmill Nobody Talks About

Here's how it works in most companies right now:

End of the month arrives. The Finance Group Team sends out ‘the email’. You know this. The subject is: "Q4 Forecast Due Friday."

Your team panics. They message department heads:

"Will you hit your target on revenue? What's the status on that big contract? What's the status on that big CAPEX project?"

Some teams respond quickly with detailed updates. Some send back details not discussed with the business due to a lack of time. The junior analyst on your team spends three days just trying to reconcile why the production forecast shows 100 employees, but payroll only has the budget for 80.

This is bottom-up forecasting. And it is exhausting.

Every month is like this.

The problem is not just the time. The issue is the quality.

Your experienced ops manager gives you a good projection based on pipeline data and historical trends. Meanwhile, the new marketing person just guesses. Because nobody trained them on what "granularity" actually means in the forecast.

Should they break it down by product, by region, or just give you one big number? They just don’t know.

What happens with this is you have disparity everywhere.

Some departments are strategic.

Others are making it up as they go.

And here is the bigger problem: by the time you've compiled everything, validated it, and presented it to leadership, you are halfway through the next month.

The forecast is outdated already!

I see this all the time in the AI Finance Club. CFOs are telling me they spend 40+ hours a month on this process. This is a full work week, just to answer the question: "Where do we think we will finish?"

This is not strategic. This is not scalable. And you did not sign up for this when you became a finance pro.

The Flip: Let AI Propose, Let Humans Refine

So here's what companies like Microsoft and Coca-Cola figured out:

Stop asking your teams to build the forecast from scratch.

Instead, flip the entire process.

Here's how it works:

Your central finance team uses AI to generate a baseline forecast automatically every month. The model pulls from:

• Actuals (what actually happened last month)
• CRM data (pipeline, deal stages, conversion rates)
• Demand planning (production schedules, inventory levels)
• Headcount data (who's actually hired vs. what was planned)
• External macro data (market trends, economic indicators)

AI runs the numbers. It builds the model. It produces a forecast for every team, every region, every business unit, all at the same time.

After this, and this is the most important bit, you send it to your teams and say:

"Here's what we think you're going to do next month based on the data. If you think you will do differently, just tell us what and why."

That's it.

You don't build spreadsheets from zero. No more chasing people for estimates. No more reconciling 47 different versions of the truth.

Your teams go from spending a full week on forecasting to spending 2-3 hours just reviewing and flagging exceptions.

Why This Actually Works

This is not just about speed. Though saving 30+ hours per month per team is really cool ;)

The real advantage is consistency and quality.

Coca-Cola improved its forecast accuracy from 70% to 90% using this approach. This is a 20% increase that translates directly to better inventory management and cash flow planning.

When you centralize the forecasting model, you get the benefit of your best people's expertise included in the baseline for everyone. Those two experienced analysts who really understand your business? Their logic now powers the forecast globally.

Junior teams aren't guessing anymore. They're reacting to a smart baseline.

And guess what. AI is really good at pattern recognition. It spots trends in your actuals that humans miss. It knows that Q3 always dips 8% in EMEA. It remembers that marketing overspent by 15% last time they launched a campaign like this. This pattern recognition alone meant a 15% increase in operational efficiency at Coca-Cola.

It's like having a very fast, very capable junior team member who never forgets anything and works 24/7.

But, and this is another thing that is really important to understand…

This only works if you are willing to change your culture.

You have to let go a bit. You have to trust that the central team can build a model that makes sense. Some finance leaders struggle with this because it feels like they are losing control.

But what you're actually gaining is time to be strategic instead of operational.

Instead of spending all month building the forecast, you spend your time analyzing it, challenging assumptions, and helping leadership make better decisions.

This is the job you really want to be doing.

You have to let go a bit. You have to trust that the central team can build a model that makes sense. Some finance leaders struggle with this because it feels like they are losing control.

But what you're actually gaining is time to be strategic instead of operational.

Instead of spending all month building the forecast, you spend your time analyzing it, challenging assumptions, and helping leadership make better decisions.

That's the job you really want to be doing.

Your Centralized Forecasting Roadmap in 5 Steps

Okay, so you are thinking this question: "Great, Nicolas, but I'm not Coca-Cola. I do not have a team of data scientists and a year to build this."

I get this.

This is why I'm going to give you the framework that works whether you're a $50M company with three people in finance or a $500M company with a full FP&A team.

Here are the five steps to flip your forecasting process:

Step 1: Audit Your Current Baseline Data

Before AI can do anything useful, you need to know what data you actually have.

What you're looking for:

• Where do your actuals live? (ERP, accounting system, spreadsheets?)
• What's your source of truth for the pipeline? (CRM, deal trackers?)
• How do you track headcount? (HRIS, manual counts?)
• What external data might be relevant? (Industry benchmarks, market data?)

Pro tip: Start with just actuals and one or two forward-looking inputs (like CRM pipeline or headcount plans). You do not need to do everything in one go. The goal is to have something AI can pattern-match against.

For smaller teams: If you're working primarily in Excel or Google Sheets, that's fine. Just make sure your data is clean and consistently formatted. AI can work with that.

Step 2: Define Your Baseline Logic

This is where you document the rules your business actually follows, not the ones you wish it followed.

Questions to answer:

• What drives your revenue? (Sales cycles? Seasonality? Contract renewals?)
• What are your biggest cost drivers? (Headcount? Materials? SaaS tools?)
• What patterns repeat every year? (Q4 spike? Summer slowdown?)

Why this matters: You're teaching AI the fundamentals of your business model. Think of it like onboarding a new junior analyst. What would you tell them about how your company actually works?

For smaller teams: Write this down as a simple document. Doesn't need to be fancy. "Our revenue is 70% subscription, 30% one-time projects. Subscriptions are predictable; projects spike in Q1 and Q4." This is enough to start.

Microsoft consolidated its field forecasts down to ~300 key data points (yours might be 30 or fewer). Much simpler than before, but with better accuracy because the logic was clearer.

Step 3: Build (or Adapt) Your Forecasting Model

Here's where it gets practical. You have 2 paths depending on your resources:

Path A - The easiest way:

We are going to use common available tools like LLM Chatbots (ChatGPT, Copilot, Gemini) and AI in Excel but also an additional tool which you might not use yet, but is open to everybody: Python (don't be afraid, I will explain to you later how).

But, before we start, here's what’s super important: you're not asking AI to magically generate projections for you. This is a ‘black box’ where the calculations are kept secret.

You can't audit it.

You can't explain it to the business.

And when it's wrong, you have no idea why.

Instead, you need to use AI like an external consultant to help you build a transparent model that you understand, can audit, and can change.

Here's how this actually works:

1. Start by identifying one homogenous item you want to forecast (sales of one product, maintenance expenses, travel expenses…)
2. Go into your favorite LLM Chatbot and explain that you want to create a simple model to forecast this line item
3. Brainstorm together on the best strategy to do it. Really important is to define:
   1. Tool stack: Do you want to do it in Excel or already use Python (which has more capabilities, and by the way, you can already use Python in Excel)
   2. What you are going to forecast: give a clear definition
   3. Which other factors impact the item you forecast (travel expenses are probably influenced by headcount in some departments and by seasonality, for example)
   4. Historical data you need to build the model
4. Then, together, review the different models you can use and spot patterns or trends you have missed

This way, you avoid the ‘black box’.

You're in control.

You decide the logic.

AI just speeds up the building process and helps you see blind spots.

Path B - For more ambitious teams:

Maybe you should start using Python?

Write a simple machine learning algorithm using linear regression or a seasonality model like Prophet (these are easily produced using tool like Google Colab).

The below is a screenshot from my training showing a forecast generated from 11 lines of AI generated code (even me who doesn’t code I can understand it and use it)?

My beginner advice? Don't try to forecast the full P&L at once. Start with something that has enough consistency, like one revenue stream or one cost category.

Pro tip: Ask your AI chatbot “Any questions you have for me before we start?”. This will help improve the accuracy and the output quality.

Step 4: Shift Your Team's Role from "Builders" to "Reviewers"

This is the culture change.

Old way: "Tell me your forecast for next month."

New way: "Here's what the model thinks you'll do next month based on actuals and trends. Does this look right? If not, what's different and why?"

Your teams now spend their time on exceptions and insights, not data entry.

What to tell them:

• "Review the baseline. If it looks accurate, approve it."
• "If something's off, flag it and tell us why. New customer signed? Big project delayed? Headcount hire fell through?"
• "Focus on what the model can't see: strategic decisions, one-off events, things that just changed this week."

Time saved: Teams go from 20-40 hours building forecasts to 2-4 hours reviewing and refining them.

Step 5: Iterate and Improve Monthly

Your first forecast will not be perfect. That's okay.

Here's what I recommend: run the model parallel to your traditional forecast for 3 to 6 months. Review every month. Adjust. And when the model will become more accurate than your traditional forecast, everybody (including and especially your management) will want to switch.

And thanks to the 3-6 months period where you ran both in parallel, you would have had enough time to learn from it but also educate your business partners and management.

Every month, look at:

• Where was the model accurate? (Reinforce that logic)
• Where was it wrong? (Update your baseline assumptions)
• What did your teams flag as exceptions? (That's a signal for what the model should learn)

Over time, your baseline gets smarter. The model learns your business better. Your teams trust it more.

And you? You get your time back to do actual strategic finance work instead of being a glorified spreadsheet compiler.

Resources to Get You Started

If you want to go deeper on this:

• If you’re a member of the AI Finance Club, watch our masterclass from the team from Coca-Cola here (if you are not yet a member, you can join us here)
• My free video tutorial on the new Excel AI agent to build financial models in a few minutes
• Check out the Microsoft and Coca-Cola case studies

The Bottom Line

I remember a CFO asking me for the 'silver bullet' for AI in finance.

And what I said to him is exactly what I've just said to you.

Centralized forecasting powered by AI.

1. Audit your baseline data
2. Define your baseline logic
3. Build (or adapt) your model
4. Shift the team from builders to reviewers
5. Iterate and improve over time

Your Move

So, take 30 minutes this week and audit your current forecasting process. Then test the approach on one small piece of your forecast.

Pick something low-risk, maybe one revenue stream. Build a baseline with AI. See if your team can review it in a fraction of the time.

You don't need to be Coca-Cola to make this work. You just need to be willing to try something smarter.

And remember… You became a finance leader to drive strategy, not compile spreadsheets. This is your chance to finally do that work.

Best,

Your AI Finance Expert,

Nicolas
​
​P.S. - What did you think of this new format? Did you find this useful and practical? Hit reply, and let me know (I read all replies)

P.P.S. - If you missed it, check out my latest YouTube video, “How This CFO Uses AI to Save 10+ Hours Per Week” here.

​