I built my first AI cash forecast back in 2023. It was one of the first use cases I tried, and it worked amazingly well. Then I tried building them for other companies, and my success rate dropped to 50/50.

A cash forecast is one of the most valuable deliverables a fractional CFO can provide, but it's not always easy to make it work. And AI tools aren't the biggest differentiator here - the tools evolved significantly since 2023, but it's still hard to create a meaningful, reliable cash forecasting process.

Like many things with AI, the specifics matter more than the tool.

In this edition, I'll walk through the three steps that turn unreliable forecasts into reliable ones. Then, in the paid section, I'll show you three ways to actually build your forecast: a simple spreadsheet for validation, an interactive dashboard for scenario planning, and a full app that stakeholders can use independently. You'll get the prompts, the frameworks, and working examples for each level.

I’m introducing a paid, on-demand 1:1 consultation for finance leaders and professional services founders working on custom, finance-related AI automations or AI strategies.

For moments when you’re unsure where to start, what to prioritize, or when a finance AI automation isn’t working as expected and needs to move from idea to a reliable, controlled setup.

As one of my clients put it:
“When I get stuck, Anna gets me unstuck.”

You can book a private 60-minute session here.

My first AI cash forecast was basically duplicating the work I was doing manually every week for years, but the LLM made it faster. I uploaded account statements, asked it to find dependencies and create a model, checked the calculation logic, and uploaded anonymized statements every week.

When I tried to replicate the same process with new clients, it worked only 50% of the time.

Here's what I learned: the forecast that worked was for a business I knew inside and out. I knew the business model. I knew what drove cash. I knew which numbers mattered and which ones were noise. I also had all the prep work done years before - we classified the transactions, my forecast was built in a way that reflected the most important drivers, and the scenarios we discussed actually made sense.

AI gave me a boost, but didn't change the core process.

Now, in 2026, the tools are better. We can build shareable apps, not just spreadsheets. We can create interactive dashboards. But the prep work still matters more than the tool.

If you want a cash forecast that actually helps you make decisions, you need three things in place before you touch an LLM.

Bank statements are a terrible data source for cash forecasting. "Payment received" tells you nothing about whether that's from your largest client or a one-time project. "Wire transfer sent" doesn't tell you if that's payroll, rent, or a vendor payment.

If you're working from bank statements, you need an extra step. Upload the statements (I would use at least 6 months of data), ask the LLM to classify the transactions, and then check the output. You'll get a better result if you provide the classes you want to use before running this exercise, but you can still get good results with a very plain prompt.

Better approach: use cash reports from your ERP or accounting system where transactions are already properly classified. If your classification isn't reliable, fix that first. A forecast built on bad data is worse than no forecast at all.

What actually drives cash in your business?

If you have a few large clients, payment timing is everything. One late payment from a major client can blow up your forecast.

If you have DSO-based revenue (think professional services, subscription businesses, anything with regular billing cycles), you're forecasting patterns, not individual transactions. What's your average collection period? How much variance do you see month to month? Are there seasonal patterns?

If you have many small clients with single payments (think e-commerce, transactional businesses), payment timing averages out. Individual client behavior doesn't move the needle. You're forecasting volume and conversion rates instead.

The same logic applies to expenses. Your biggest cost driver has the biggest impact on cash. If payroll is 70% of your costs, that's what matters. If you carry inventory, working capital swings drive your cash position. If you have variable COGS tied to revenue, that's the dependency you need to model.

Generic cash forecasts ignore all of this. They treat every dollar the same. They don't distinguish between businesses where timing matters and businesses where volume matters. That's why they fail.

Your forecast needs to reflect your business model, not some theoretical template.

Step 3: Deal with Data Security

Your ability to use AI for cash flow forecasting depends on your access to tools, your subscription tier, and your company/client policies around data security. In no scenario can you use a free or personal subscription to work with real cash data. If you have a free or personal tool, you can only work with anonymized data.

If you're using a paid version, disable data sharing in the settings. Most platforms use your data for model training unless you explicitly opt out.

Platform settings checklist:

If you only have a free or personal account, anonymize:

Data security isn't an afterthought. It's the first question you answer before you build anything.

So before you rush into creating an AI tool, spend time refining classification, identifying cost drivers, and validating the model. Only then proceed with the cash flow forecast generation.

In 2026, there are multiple tools and apps you could be using for that. There are special standalone apps, you can create custom GPTs, Projects, Agents, or Gems as a first automation step, but none of it will work unless the groundwork is done.

And this is what I'm always saying to my classes and clients: AI will not replace your expertise. It can only accelerate what already works.

We've covered the prep work: classification, drivers, and security. These are the steps that turn a 50/50 success rate into reliable forecasts.

Now, in the subscriber-only section, I'll show you three ways to actually build your cash forecast.

Level 1 is a simple spreadsheet any LLM can generate - use this to validate your model before automating.

Level 2 is an interactive dashboard for scenario planning.

Level 3 is a full app where stakeholders can upload their own data and run scenarios. Pick the level that matches your needs.

The difference between a cash forecast that works and one that doesn't isn't the sophistication of the tool. It's whether you've done the prep work to understand your business model, identify your drivers, and validate your assumptions.

As always, I'd love to hear how this goes for you. Hit reply and let me know what you're trying first.

—Anna

This newsletter is for you, and we want to make it as valuable as possible. Please reply to this email with your questions, comments, or topics you'd like to see covered in future issues. Your input shapes our content!

Want to dive deeper into balanced AI adoption for your finance team? Or do you want to hire an AI-powered CFO? Book a consultation!

Did you find this newsletter helpful? Forward it to a colleague who might benefit!

Until next Tuesday, keep balancing!

Anna Tiomina
AI-Powered CFO