Why ‘black box’ AI is failing FP&A and what to use instead

Over the past decade, we’ve watched artificial intelligence seep into every corner of our professional lives. But something feels different now. The buzz isn’t just about potential anymore, it’s about impact.

When ChatGPT reached 100 million users in just two months, it was a wake-up call that we’re in a new era of productivity and innovation. At the intersection of finance and technology, a transformation is underway, and financial planning and analysis (FP&A) professionals are uniquely positioned to lead it.

I want to share what I’ve learned about how we can leverage AI in financial forecasting, so in this article, I’ll cover:

Understanding AI and it’s role in FP&A
How AI is impacting FP&A
AI algorithms to use
Practical AI: From data to forecast
Explainable AI (moving away from 'black box' AI)
AI tools for forecasting
How to prepare for AI and ML in FP&A

Understanding the role of AI in FP&A

Let’s take a step back. What exactly is artificial intelligence?

At its core, AI is the science of creating machines that can perform tasks that normally require human intelligence. In finance, AI is used in three main ways:

Data-driven decision making
Automation and efficiency
Predictive analytics

In FP&A specifically, AI offers a wide range of opportunities. I like to categorize its use into three pillars:

Simplifying complexity
Generating new knowledge
Saving time

Let’s unpack these a bit more.

1. Simplifying complexity

AI tools like ChatGPT, Microsoft Copilot, and Google Bard/Gemini have made it easier than ever to demystify financial concepts. Imagine asking a chatbot to explain IFRS 10 to your marketing manager, and getting an answer that makes sense.

AI-powered reporting is another major benefit. It consolidates your financials and provides real-time reporting capabilities. You can even build your customized version of GPT, embedded with your company’s internal policies so that it becomes an expert in your context.

2. Generating new knowledge

This is where financial forecasting comes into play.

AI allows us to go beyond basic reporting into predictive and prescriptive analytics. It not only tells us what might happen (predictive) but also what we should do about it (prescriptive).

AI can help identify patterns, uncover hidden trends, and generate insights. Beyond forecasting, AI can be used for clustering, predictive insights, and scenario planning.

3. Saving time

AI can automate tedious tasks like file merging and data consolidation, freeing up FP&A teams to focus on strategic initiatives. I often tell my clients: "Let the bots crunch the numbers, so you can focus on the business."

How AI works in FP&A

To grasp AI's potential in forecasting, it's important to understand the four stages of data analytics:

Descriptive Analytics - What happened?

Descriptive Analytics focuses on summarizing and interpreting historical data. It answers questions about what occurred in the past, such as sales trends customer behavior, patterns, and financial performance over time.

Diagnostic Analytics – Why did it happen?

Diagnostic Analytics digs deeper into data to understand the causes behind observed events. It shares insights on the reasons for past performance or trends, such as why sales dropped in a particular quarter or why a marketing campaign was successful.

Predictive Analytics – What could happen?

Predictive Analytics uses statistical models and forecast techniques to make educated guesses about future events. It reveals what might happen next based on patterns and trends identified in historical data, such as predicting future sales growth or market trends.

Prescriptive Analytics – How can we make it happen?

Prescriptive Analytics goes beyond predicting future outcomes by suggesting actions and strategies to benefit from these predictions. It answers questions about the best course of action, such as what strategies should be implemented to increase market share or how to optimize resource allocation for maximum efficiency.

Forecasting falls into the predictive analytics bucket, but its real power is unlocked when combined with prescriptive insights.

AI algorithms you can use

Various algorithms can support financial forecasting. Here’s a quick rundown:

Intuitive Forecasting (Human judgment)
Run Rate Analysis (Using historical trends to project future performance)
Linear/Logarithmic Regression (Modeling relationships between variables)
Time Series Models (ARIMA, SARIMA)
Machine Learning Models (Random Forest, Neural Networks)
Prophet (Developed by Meta for seasonal time-series forecasting)

I personally find Prophet to be highly effective for datasets with strong seasonality patterns.

Practical AI: From data to forecast

There are three main stages in financial forecasting where AI plays a role. The first is data collection and consolidation. FP&A teams spend a massive chunk of time gathering data from different departments. AI and automation tools can reduce this effort by up to 60%. Imagine pressing a button and having all your data consolidated and cleaned in minutes.

With data ready, the next step is building forecasts. Using AI models, you can generate more accurate predictions. Better yet, AI allows you to test different assumptions and scenarios quickly.

Finally, forecast optimization is where explainable AI becomes crucial. Rather than being a black box, explainable AI allows users to see why a forecast is what it is. You can understand the key drivers and make informed decisions.

The shift toward explainable AI

One of the biggest barriers to adopting AI in FP&A is trust.

Traditional AI models often operate like black boxes: you feed in data and get an answer, but you don't know how the system arrived at that answer.

Explainable AI, or "glass box" AI, changes the game. It offers transparency by providing the reasons behind its predictions. This makes it easier for FP&A professionals and stakeholders to understand, evaluate, and trust model recommendations.

So, what does that look like?

Well, with the black box approach, we begin with traditional data and the model will start learning from that training data. The key output will be a decision or recommendation. Now the problem with this is that the only output is that learned function and that decision without actually explaining the drivers or the root causes of that forecast.

It's very similar to if you need to create a sales forecast. You just say that the number of sales that you are going to sell the next year is going to be one million units and then you stop there. You don't say anything else.

With that, the actual humans consuming those models will have a lot of questions. Like, why did AI do that? Why not something else? How should we interpret those results? And even more importantly, how do we trust that model?

With the second example, we have a different approach - the Glassbox AI approach.

We still have the training data, but then we have this explainable AI and machine learning algorithm. With that, the key output is not just the decision or recommendation, but also an explainable model.

So something for the actual human to understand why the sales forecast was one million, why not two million, or why not 500,000. With this approach, the humans in FP&A can actually interpret and improve that forecast. Most importantly, they can trust that forecast.

Glass-box AI enables:

Transparency
Auditability
Trust in model outputs

AI tools for forecasting

I get asked all the time, "Which tools should I be using?"

Here’s a quick guide to some of the most useful ones for financial forecasting:

ChatGPT + GPT Builder: Build finance-specific GPTs that understand your policies and workflows.
Microsoft Copilot Pro: Integrated with Excel, Power BI, and Fabric. You can generate reports, analyze trends, and ask financial questions using natural language.
Azure: Offers robust cloud-based AI capabilities for scalable models.
Power BI: Leverage Copilot in Power BI for dynamic, interactive dashboards.
Excel + Python: With Python integration in Excel (rolling out now), you can run models like Prophet or ARIMA directly in your spreadsheets.
Google Colab: Great for building and running Python-based models in a browser-based environment.

I often teach a four-step process for using ChatGPT for forecasting:

How to prepare for AI and ML in FP&A

To stay competitive, finance teams need to prepare. Here are five steps I recommend:

1. Data quality and integration

Data is the lifeblood of AI and ML. Organizations must ensure data quality, consistency, and integration across various sources. This includes cleaning and structuring data to make it machine-readable.

2. Talent acquisition and training

Hiring data scientists, analysts, and AI/ML experts is essential. Training existing FP&A teams in AI and ML techniques can also bridge the skills gap and empower them to leverage these technologies effectively.

3. Selecting the right tools

Choosing the right AI and ML tools and platforms that align with FP&A goals is crucial. Options range from predictive analytics software to AI-driven financial planning platforms.

4. Change management

The integration of AI and ML will bring organizational change. It’s important to communicate the benefits and challenges to all stakeholders and facilitate a smooth transition.

5. Continuous learning

AI and ML are ever-evolving fields. Continuous learning and staying updated on the latest advancements are essential to maximize their potential in FP&A.

Attend one of our finance summits to learn more about AI in finance

Join our global summit series in a city near you to discover how to unleash the full strategic potential of your finance function.

Get in a room with industry leaders as they share best practices and common challenges to help you build the skills you need to increase accuracy and predictability, integrate AI effectively, and foster a high-performing team.

Find an event near you