How Accountants Use Discrepancy AI for Financial Document Review

Double entry bookkeeping goes back as least as far as the Medici family in Renaissance Italy, and remains relatively the same today. Despite innovations in fintech and banking, debits and credits are still the bedrock of accounting. And in daily practice, accounting still involves hourly billing and the same core question: ‘can I trust this number?’

Needless to say, accountants have historically been some of the most reluctant supporters of AI. As of 2024, 73% of firms are not using AI tools, and over half of them would not trust AI for even internal reports. In the same survey, though, 70% of accountants believe AI will have a positive impact on their job. 

AI Should Complement, Not Replace, Accountants’ Professional Judgement

AI is here to stay in accounting, but manual review is part of the job. How do you trust a number? Maybe you know what subledger it comes from and that it matches the line item in the GL. Or you test a sample of receipts or payroll runs. Or you compare a customer’s income or mileage to last year’s filing. And so on. 

Manual work becomes a challenge when it's time to scale. Client paperwork, filing deadlines, and quarterly cycles mean accountants work around “seasons” of heavy workload. Tax preparers, for example, will work consistently for only 4 months of the year, and with 31% of Americans procrastinating on their taxes, these seasonal swings are expected to continue for the foreseeable future. 

Short earning seasons mean accountants need to make the most of their time. At the same time, accountants' professional judgement is often the final checkpoint for fair reporting…which inherently involves some manual workload.

Did you know double-entry bookkeeping was actually built for scale? The system of debits and credits helped the Medici bank to keep track of more complex transitions across different currencies, which allowed the bank to grow beyond traditional activities and expand. 

Artificial intelligence in accounting operates in much the same way. The modern accountant does the same debits and credits as ledger clerks in the Medici-era, on a much greater scale. The speed of accounting today is made possible by tech like document portals, accounting software, and Microsoft Excel – but accountants of all kinds are likely to get stalled by piles of paperwork, longform written documents, or unscannable pdfs. Tax files can be hundreds of pages long. Supporting documents are often unstructured (like gas station receipts for business travel) and sensitive. Even scanned bank statements can be difficult to take straight into Excel, and accountants usually can’t request clients connect the bank themselves. It’s easy for a human to interpret a bar chart, but harder for tools like OCR. 

Scaling Firm Workflows with Assisted Document Review 

‍Natural language processing (NLP) has a much more accurate capacity for financial document review because it can understand content and context. A user can name an entity, organization, or specific data (vendor names, payment terms, line item codes, the total from the gas station receipt…), and the NLP will pull it from an unstructured doc. 

Here the accountant will still physically review each document, they’ll just do so in a fraction of the time. For example, expense reports are uploaded into a staff portal. The firm’s CFO uses AI to scan through documents in this portal, flag unusual or vague items, identify inconsistencies across prior periods, or compare them against a policy mandate from the company. 

AI tools understand phrases like “we’ll make an exception this time” or “I used my frequent flyer points, submitting for cash equivalent” and can flag them for the CFO to physically look at. It can also flag phrases like “rolls over unless cancelled” in renewal documents or “let’s fast track this.” 

Tools targeted specifically at accountants can be customized to provide a content log for how, what, and where the model decided to flag text, which can make a big difference in ensuring the completeness and reliability of the data. 

Efficient Document Review Sets Priorities and Saves Time

Human oversight is a part of the accountants’ role, but AI can help prioritize documents, surface key data, and speed up manual document review. AI can act as an extension of your own professional judgement and complement your workflow. The end result is capturing more client opportunities (or billable time) during busy periods, quarters, or year ends – all without sacrificing the accuracy of your results.