How will we use AI to build spreadsheet forecast models?

Can artificial intelligence (AI) be used to build more accurate and efficient spreadsheet forecast models? As we've started to see in ...

• AI-powered image editing and generation (e.g. Dall-E ^[1], Facet),
• AI chatbots for writing, research and coding (ChatGPT, Jasper, CopyAI, Moonbeam, and scores of others),
• AI-powered video generation and editing (e.g. Synthesia),
• AI-enabled programming assistants (Github's Copilot),
• AI-powered legal document creation and reviews (Tome),

... and many other areas, the idea of using AI to help people do work is on the cusp of mass experimentation. Prompting AI engines (the idea of "prompt battles" is fire) might soon be the first thing we do to create a paragraph or a block of code. Instead of using Google search to dig through endless half-answers around the web and figure out how to apply them to our needs, why not just ask an AI chatbot to do it for us? ^[2]

A decade ago I wrote a post about personal APIs, the idea of creating a way for someone to access people's knowledge and insights without direct interaction. Way too early then, but it's becoming possible today. It's possible today to use machine learning to train an AI model on a set of things you're previously written, and then create a prompt input box on a website for people to ask your AI model questions and get advice.

The days of using AI in financial reporting are coming soon. Examples of AI around financial forecasting and spreadsheets are starting to bubble up:

• Microsoft Excel, Zoho, and Google Sheets are all adding AI-powered features to their platforms to help make it easier for users to use spreadsheets to accomplish common tasks. Function suggestions with pre-filled guesses based on highlighting a section of data, cleaning messy data, and creating reports are early examples of their efforts.
• Microsoft has released a preview of Copilot for Finance, to help automate the processes to get, aggregate, and reconcile data, as well as recommend insights, build charts and displays of data, and speed up the process of analyzing financial data.
• Formulabot translates text prompts into formulas, using AI to write Excel and Google Sheets spreadsheet formulas. Spreadsheets involve their own coding "language" and thought process, and making it easier to write spreadsheet formulas is analagous to making it easier to write programming code.
• Prophet is a forecasting procedure implemented in R and Python, released as open source software by Facebook's data science team. It's built to use historical time series data to create forecasts of operational and financial data.

And there's likely countless others I'm missing. It makes sense that incorporating AI into the process of building a spreadsheet forecast model can enable companies to benefit from the enhanced data analysis and predictive capabilities of AI. Gather financial and operational data - historical financial information such as income statements, balance sheets, and cash flow statements, as well as data on expected changes in the company's operations, such as new products or services, changes in pricing, or expansion into new markets - and train an AI model to use for forecasting and data analysis. Take public market data to create a tool to query for benchmarks and comparisons for types of businesses at specific stages. Train a model on your own company's data to create a tool to help forecast future sales, inventories, cash position.

Once we have an AI model, the results can then be incorporated into a spreadsheet forecast model, using cells as prompts to generate forward-looking projections or generate insights from historical data to inform the forecast assumptions and projections.

In addition to improving the accuracy and efficiency of the forecast model, AI could be used to automate certain aspects of the model-building process. Automatically update the model with new data as it becomes available. Automatically generate reports and presentations based on AI-powered inferences on the actual and forecasted data in the model. Dump data on a company into an AI model for it to figure out which statistical and forecasting methods to use, and why.

The days of using AI to build spreadsheet forecast models to make more accurate and informed financial projections is coming closer by the day. Does it replace the role of an analyst or a CFO? No, but that's not the point.

Joel Shapiro at 4 Phases of Analytics Evolution: From Spreadsheets to AI Workbenches:

... if an AI model indicates that a certain customer segment is likely to increase in value, humans still need to decide what to do about it. Should they engage more with these customers to ensure a good outcome occurs or keep the status quo because these customers are likely to lead to a good outcome on their own? In other words, it takes human expertise, and AI provides valuable inputs to supplement that expertise.

Models are about the analysis, not the artifact, and the more we can leverage technology to create the artifacts, the more time and energy we leave for people for analyses and decision-making. The evolution from spreadsheet jockey to prompt CFO is upon us.