For accountants and client advisory firms interested in using AI to build financial models, here's a practical framework you can follow to deliver more value to clients while streamlining your workflow:
Step 1: Define objectives and financial modeling goals
Start by clearly articulating what you're trying to achieve for your client. Are you forecasting cash flow for the next 12 months? Modeling different tax scenarios? Projecting growth after a potential acquisition?
A clearly defined objective might look something like: "Create a 3-year revenue forecast model for our manufacturing client that accounts for seasonal fluctuations, projected material cost increases, and three different market growth scenarios." The more specific your objective, the more accurate and useful your AI-assisted model will be.
Step 2: Collect and preprocess data
Next, it’s time to gather the data and get it ready for the model. Leverage digital tools you already have access to through your client’s accounting system such as QuickBooks. This might include historical P&Ls, balance sheets, cash flow statements, and general ledger data. You can supplement this with industry benchmarks, economic indicators, or alternative data sets with, for example, social sentiment.
Before inputting your data into a generative AI (GenAI), you’ll want to clean it. This involves removing outliers such as, for example, equipment purchases of $80,000 in a quarter where capital expenditures are normally under $5,000. You’ll also want to ensure consistent categorization, and double-checking the set to verify that everything looks accurate. Always keep in mind that AI isn’t error-proof, and human oversight will be key at every stage of this process.
Step 3: Identify key financial drivers and inputs
After you've collected and cleaned your data, you need to determine which variables actually matter before selecting a modeling approach. Without this step, you’d run the risk of creating models that include too many irrelevant variables and create noise for your client rather than valuable insights.
Depending on the nature of your client's business and their industry, certain data points might be more meaningful than others. For example, service-based businesses might be heavily influenced by employee utilization rates and hourly billing efficiency, while product-based companies might be more sensitive to inventory turnover and supply chain metrics.
To leverage AI in determining which metrics and relationships will be critical for building your model, you can upload your prepared financial statements into a GenAI tool and use specific prompts such as, "Analyze these financial statements and identify which factors show the strongest correlation with profitability," or "Based on this data, which metrics best predict cash flow constraints for this business?"
Note that as you begin this process, there are a few ways you can put your financial data into the system. You can:
- Upload the file directly when the AI platform supports file uploads.
- Copy and paste tabular data from spreadsheets into the chat interface.
- Convert spreadsheet data into a structured text format—for example, CSV content as plain text.
As patterns start to emerge, your role will be to layer in your accounting expertise to evaluate whether different correlations make practical business sense for your client. Your knowledge of their operations, industry trends, and business realities will help you discern meaningful relationships such as a correlation between marketing spend and revenue, from statistical coincidences; for example, a correlation between office supply expenses and gross margin. This ensures you're building models based on factors that genuinely impact your client's financial performance.
Step 4: Select and train your model
This stage involves understanding the various AI approaches available for financial modeling and determining which will work best for your client's specific needs, as well as your objective outlined in step 1.
Here, there are two main things to consider:
- Traditional machine learning: Traditional machine learning refers to algorithms that learn patterns from data by identifying relationships between variables. Regression, classification, and clustering models work well with smaller, structured datasets. They produce more interpretable results that you can clearly explain to clients and are effective for many standard financial tasks. When using regression models, for example, you receive clear coefficients that show exactly how much each input variable influences the outcome.
- Deep learning: Deep learning refers to advanced neural network algorithms that process data through multiple layers, mimicking how the human brain identifies complex patterns. Models such as neural networks, recurrent neural networks, and convolutional neural networks excel with extremely large datasets containing complex, non-linear relationships and time-dependent financial data with intricate sequential patterns. The downside here is that the end result tends to be a bit of a black box, making it harder to explain and justify the logic of the output.
Again, your selection between these approaches will be largely dependent on your objective and use case.
Traditional machine learning works best for situations where transparency and the ability to explain information in depth to your clients is crucial. This includes:
- Explaining key profit drivers.
- Forecasting based on clear patterns.
- Demonstrating ROI of specific activities.
- Analyzing why profit margins are changing.
Deep learning is more appropriate for cases where predictive power might outweigh the need for explanation, including:
- Predicting market movements from vast amounts of information.
- Forecasting with hundreds of interdependent variables.
- Identifying subtle market signals.
- Detecting sophisticated fraud patterns.
Step 5: Generate and validate projections
With your model selected, it's time to generate the projections you'll use in your financial model. You can start by prompting the AI based on your chosen model. Here are a few examples:
- "Use regression analysis to identify the top factors influencing our client's profitability.”
- “Analyze these datasets to detect patterns in cash flow fluctuations."
Once the data has been processed and key relationships have been identified, prompt your GenAI to generate projections based on these insights.
For example:
- "Using this analysis, what would be the projected monthly revenue figures for the next 12 months?"
- Consider challenging the results directly with a prompt like this: “Our client has never achieved margins above 35%. Can you elaborate on this projection or revise to provide more realistic figures.” This is where your knowledge will need to come into play, again. If projections show figures or growth rates that seem unrealistic given your client's history or industry benchmarks, challenge these results directly.
The goal is to ensure your final projections are accurate and mathematically sound, and practical in the context of your specific client's circumstances.
Step 6: Scenario planning and sensitivity analysis
At this point, you can leverage AI to model various "what-if" scenarios that can help your clients prepare for different potential outcomes.
If we call back to the example of a manufacturing client, you might use the following prompt: “Create three scenarios for our 12-month forecast:
- Base case using current trends.
- Conservative case assuming raw material costs increase 15% and orders decrease 10%.
- Growth case assuming we secure a new contract we're bidding on, increasing production volume by 25%.”
Sensitivity analysis helps identify which variables most dramatically impact outcomes. Ask the AI to test variable ranges:
- "Show how our manufacturing client's profitability changes when raw material costs fluctuate between +5% and +20%, holding all other factors constant."
This step transforms static models and forecasts into dynamic planning tools that allow your clients to prepare contingency plans, set triggers for action, and make more informed strategic decisions. It also elevates your role from a reporter of historical performance to a strategic advisor with customized business intelligence for future outcomes.
