Parties to litigation proceedings are often required to disclose their personal spending. The information provided to the Court may require certification that the information is complete, true, and accurate under penalty of perjury, yet often based, at least in part, on estimates. This article shows how a few simple steps can achieve the desired results without the costs, time, or guesswork.
Parties to litigation proceedings are often required to disclose their personal spending. The information provided to the Court may require certification that the information is complete, true, and accurate under penalty of perjury, yet often based, at least in part, on estimates. In some cases, the party may defer to a professional for its preparation despite the cost. Fortunately, rapid improvements in artificial intelligence (AI) platforms have created a viable alternative to this burdensome activity. This article shows how a few simple steps can achieve the desired results without the costs, time, or guesswork.
Consumers in the U.S. had an average of 5.3 different accounts across financial institutions in 2019.[1] To establish a pattern of spending, at least 12 months of activity would capture infrequent payments (such as insurance, vacations, holidays, etc.). By some analysts’ estimates, two years or more would be required to support a reliable forecast.[2] Using the 12-month minimum, a litigant with the average of 5.3 accounts would need to examine approximately 64 monthly statements to adequately support a personal spending analysis; under the two-year standard, that figure roughly doubles to 128. This becomes more complicated when crypto or digital wallets (like Venmo or Zelle) are considered.
Manual data entry is time-consuming, tedious, and prone to errors. Even spreadsheets have a typical error rate of 1% to 10% for nontrivial but seemingly simple cognitive tasks.[3] Fortunately, with accurate prompting, AI models can reduce these errors. This will be demonstrated below.
Example Divorce Engagement
The most important document that a party to divorce will complete is a financial statement.[4] For our hypothetical example, an individual has 76 account statements for the period May 2025 through June 2026 and wants to disclose the spending in an efficient, cost-effective manner. Luckily, they have access to a large language model (LLM) and typed in the following prompt:
Using the attached statements, please make an Excel spreadsheet of all the inflows and outflows. Have the columns for each month and the rows categorized into activities (such as utilities, food, housing, travel, etc.). Exclude any duplicate statements and add a separate column to track the source data.
Despite the respectful entreaty above, most AI models limit the number of statements that can be utilized at any given time. This can be remedied through repeated searches as discussed below. See example output from this procedure at Table 1.
Additional wording can include the following:
- Format the spreadsheet in a professional manner where all columns will fit on one page and use Calibri font.
- Show list of sources for easy reference.[5]
- Add a column that creates a unique identifier for each transaction.
- Maintain separate tabs in the spreadsheet for each account. (Note: This can also be modified to segregate bank, credit cards, investment accounts or other ways data can be segregated such as transfers in/out.)
Crafting an effective prompt may take some repetition to get the output to match the user’s intent. For example, the output may contain too many or too few categories for a meaningful purpose. Consider uploading the personal financial statement used in the divorce to ensure comparability.[6] The AI will also flag items for review (such as recurring payments or new accounts) or may add a column to the source data highlighting questionable transactions.
This procedure may only take a few minutes or it may require some back and forth with the platform to ensure there are no duplicate transactions, the transfers are segregated, and the data is properly aggregated. See example output from this procedure at Table 2. Note: The data contains excluded categories which may require additional formatting/investigation before finalizing.
Advantages
The positive aspects of using AI platforms for personal spending analyses include the following:
- User Friendly—In the past, software such as IDEA or ACL Analytics required users “teach” the software the location of important fields, type of data, and other information to capture. That meant a custom import template for each financial institution and knowledge of specialized software. Now, AI models can recognize the relevant information based on the prompt without the tedious data cleaning.
- Faster—The time needed to perform the analysis has been reduced from days/hours to minutes. This does not eliminate the need to verify the output, but the tedious data entry aspect is minimized so users can focus on interpreting the output.
- Smarter—AI platforms can also reduce the diligence required of the practitioner, as the models are capable of self-checking their work. While this does not eliminate the need for human oversight, it is always easier to reconcile the output with the source data than manually create from scratch. The following prompt can be added to the request to improve accuracy of the output:
Add a tab that ensures the source data reconciles to the statement summary. For example, the beginning balance of the statement plus the inflows, minus the outflows, should equal the ending balance. The “rollforward” of the monthly account should be shown on the new tab.
See example output from this procedure at Table 3. The prompt can also be modified to include hash totals, pivot tables, etc. to verify spreadsheet accuracy.
- Supporting Documentation—The platforms will typically have a tab with notes and methodology to show compliance with the search criteria. If not initially provided, the prompt can be modified to add this parameter. When in doubt, users can select platforms that include citations, reasoning chains, or confidence scores.
- Broad Application—While the focus of this article is on personal spending, the techniques can extend to litigations regarding self-employment activities, business income determination, or other financial disclosures.
Once in Excel, the data can be easily manipulated for additional formatting (ex. Currency), converted to weekly/annual amounts, or used creatively via graphs and highlighted trends.
Disadvantages
The negative aspects of using AI for personal spending analyses include the following:
- Data Privacy—The primary concern for anyone using an AI model is data privacy. For individuals with access to a paid subscription or enterprise platform, limiting the search to “work” rather than “web” (i.e., disabling any web-search or internet-connected feature) reduces the risk that the statements will be exposed outside the platform. Several platforms offer enterprise-grade privacy protections and there are resources to help select the platform with secure document handling.[7]
The privacy policy and account settings should also be checked before conducting any search using sensitive data. For example, Anthropic will use inputs and outputs to train and improve Anthropic AI models, unless users opt out through their account settings.[8] Knowing and using the available safeguards (such as incognito mode) will reduce the likelihood that financial data is retained or used by the models.
- Non-Expense Transactions—Spending reported on bank and credit card statements are not limited to expenses. For example, there could be gifts, asset purchases, loans, or other items that affect the outflows of funds from the accounts. These should be segregated so they are not double counted or misinterpreted (e.g., a transfer from one account treated as an expense for one account and income on another). In addition, savings may need to be addressed separately, as some jurisdictions consider savings part of a supported spouse’s need when modeling the analysis.[9]
- AI Platform Subscriptions—There are limitations on the number of statements that an AI tool will accept for an individual search. The following chart highlights the options available at the time of this writing:
|
Platform |
Plan/Tier |
Files per Chat |
Max File Size |
|
Claude (Anthropic) |
All |
20 |
500 MB |
|
ChatGPT (OpenAI) |
Free |
3/Day |
512 MB |
|
ChatGPT (OpenAI) |
Plus |
10-20 |
512 MB |
|
ChatGPT (OpenAI) |
Pro |
Unlimited |
512 MB |
|
Gemini (Google) |
Free/Pro/Ultra |
10 |
100 MB |
|
Microsoft Copilot |
Free/365 Personal |
20 |
50 MB |
|
Microsoft Copilot |
365 Enterprise |
Varies |
512MB |
Individuals with few accounts may be able to use free or lower-cost tiers. However, parties with numerous accounts or transactions will likely need a paid subscription to accommodate the volume of statements.
Using our sample prompt for divorce spending, the multi-step process resulted in four Excel spreadsheets which can be consolidated into one file with the following prompt:
Combine the attached spreadsheets into one professionally formatted Excel document to be used in connection with a divorce filing in (county), (state). Include all four files’ unique transactions combined, clearly labeled by source.
- Hallucinations and Formatting Inconsistencies—Asking the model to demonstrate that it has self-checked its work may not be enough if it is hallucinating. Users can overcome hallucinations by forcing the model to show its step-by-step reasoning (chain-of-thought), asking for a probability of falsehoods, or other methods. Also keep in mind that multi-step scans may result in dissimilar layouts. Formatting issues can be remedied in the consolidated Excel file, though unsettling when four searches yield four different formats.
- Technical Requirements—The platform needs to fit with the user’s computer system. Even the best AI tools may not be compatible with the user’s tech stack or technical abilities. Further, users should be wary of prompt interjection attacks which can process external, untrusted files (like a downloaded Word document or web page) containing hidden malicious instructions that can trick the AI.
Professional Standards for Financial Forensic Experts
Financial experts who incorporate AI into their methodologies are subject to professional standards and both authoritative and non-authoritative principles-based guidance. For example, the National Association of Certified Valuators and Analysts (NACVA) Artificial Intelligence and Machine Learning Commission (AIMLC) advisory brief said constituents should:
- Consider disclosing the use of any automated data output utilized in the analytical process
- Ensure such technologies were applied in an ethical manner while applying professional judgment and proper due diligence
- Consider utilizing non-confidential information or ensure the AI tool is designed to comply with confidentiality standards
- Commit to continuous learning to stay abreast of the latest technologies and methodologies
- Not let AI supersede the valuator’s discerning analysis
- Actively participate in professional development opportunities
- Engage in opportunities that discuss AI innovations, and
- Contribute to the body of knowledge on AI in valuation[10]
Hiring a financial expert to prepare a personal spending analysis may be cost prohibitive, even if the professional is using AI themselves. Any expert opinion of lifestyle or spending will be subject to professional standards and include disclosures of AI usage. This extends to the use of AI to critique opposing experts’ analyses.
Conclusion
The need for personal spending analyses will continue for the foreseeable future. Parties with busy schedules can efficiently self-report their lifestyle spending rather than hire professionals to input, analyze, and quantify their sensitive transactional data. The decision to use AI should weigh the cost of hiring qualified professionals against the downside risk of cyberattacks and unintended consequences. Expert opinions are rendered through deliberate processes and subject to professional standards whereas AI can produce slop if unverified.
AI platforms can be utilized to accurately report financial activity though vulnerable as they are powerful. Until AI can testify, there is no substitute for independent, objective experts; however, this article showed a viable alternative in the right circumstances.
—
Note: AI is rapidly changing the digital analysis landscape. Techniques suggested in this article are as of its publication date and are not proposed as long-term solutions.
[1] Reville, Peter. ATM Banking: It’s not just about Cash Withdrawal Anymore, Mercator Advisory Group – North American Payments Insights, June 6, 2019.
[2] Anamind Blog, How Much Data is Required for Forecasting, https://www.anamind.com/anamind-blog/how-much-data-is-required-for-forecasting/.
[3] Panko, Raymond, “The Cognitive Science of Spreadsheet Errors: Why Thinking is Bad”, 46th Hawaii International Conference on System Sciences, January 2013, Maui, Hawaii. The cell error rates (CER) range from 1%–3% and increase with spreadsheet formulas or chains of calculations.
[4] Bilodeau, David and Soilson, Jeffrey, Financial Aspects of Divorce in Massachusetts, MCLE, 2nd Edition, 2023.
[5] Note: Many AI models will automatically add a tab with notes to describe which statements used, the periods covered, and category definitions.
[6] For example, the following prompt could be used for Massachusetts: Using the expense categories from the Massachusetts Form CJD 301L long form, consolidate the spending activities into the expense categories listed in Section VI. Limit additional categories to 10, with the catch-all being miscellaneous.
[7] Hebbia Inc. Resource, 10 Best AI Tools for Financial Analysts, https://www.hebbia.com/resources/ai-tools-for-financial-analysis, last accessed July 4, 2026.
[8] Anthropic Privacy Policy, 1. Collection of Personal Data, https://www.anthropic.com/legal/privacy, Effective July 8, 2026.
[9] See Openshaw v. Openshaw, SJC-13473, 493 Mass. 599 (2024).
[10] NACVA Advisory Brief, The Use of Artificial Intelligence and Machine Learning, October 29, 2024.
Table 1: Cash Flow Summary by Month
Note: The above is based on a query using 20 PDFs of personal account statements
Table 2: Consolidated Cash Flow Summary
Note: The above is based on all 76 PDFs of personal account statements. Further analysis of excluded items may be necessary.
Table 3: Sample Account Rollforward
Note: The above is based on a query using 20 PDFs of personal account statements and the months were not continuous.
Jason Pierce, CPA, CMA, CFM, CVA, MAFF, has over 30 years of experience and serves as a financial expert in various accounting and finance-related disciplines. He is a Managing Director and leads Paradigm Forensics’ Anchorage and Boston offices, and works with clients on disputes, transactions, fraud, damages, family law, criminal, and estate planning engagements. He is a lead instructor for NACVA’s Master Analyst in Financial Forensics (MAFF) certification program and other courses including the Forensic Accounting Academy and the Commercial Damages and Lost Profits Workshop. He is a regular speaker at legal and professional organizations. He is an active member of the Massachusetts Society of CPAs (Business Valuation Committee) and the Boston Chapter of the Institute of Management Accountants (Vice President – Education).
Mr. Pierce can be contacted at (410) 609-4740 or by e-mail to jpierce@paradigmforensics.com.
