As Ben Franklin famously said, there are only two certainties in life: death and taxes. We would argue AI should be added to that list. As LLMs continue to advance, AI is working its way into everything, from marketing to voice agents to professional services at large. Whether it’s PWC announcing a $1 billion investment into AI solutions, or Reuters setting aside $8 billion for AI dealmaking and development, firms are eager to establish themselves as market leaders in implementing this new technology — especially those that serve the accounting market.
This shouldn’t come as a surprise to anyone. Bookkeeping, accounting, tax preparation and auditing are fields full of largely formulaic and repetitive exercises that would immensely benefit from generative AI’s gift of efficiency and time savings. And there are real, quantitative tailwinds that make this a particularly critical moment for accounting firms to lean into AI and machine learning. For starters, 75% of CPAs could retire in the next 10 years. Simultaneously, the profession is attracting fewer job entrants, with the number of U.S. students who complete accounting degrees falling. This means in the coming years, far fewer professionals will be available to handle existing client demand — and firms are already struggling to keep up.
Luckily, genAI has hit the scene, and corporate finance and accounting workflows should benefit greatly from its capabilities. It is, however, important to be prescriptive about where specifically this new wave of LLM-based AI can help. At their core, LLMs are best at working with natural language. They are adept at summarizing research, answering questions, and delivering information that gets their prompter ~70% of the way to a definitive result. What they lack (for now!) is the ability to do complex calculations and quantitative analyses — two skills crucial to the accounting profession.
Despite this, accounting workflows are still ripe for disruption: the industry is filled with highly paid professionals who spend many hours per week ingesting data from disparate sources to make recommendations based on the recognition of patterns observed over countless hours of educational training and professional experience. While there are many nuanced subspecialties within the fields of finance and accounting (e.g., bookkeeping, tax preparation, auditing, outsourced CFO duties) as well as vertical-specific needs, we’ve identified several high-level jobs to be done that are prevalent throughout most, if not all, of the category. The key difference in many cases will be the relative level of emphasis on one aspect of the job vs. others (e.g., bookkeeping focuses more on data ingestion while giving tax advice requires complex analysis and output). Moreover, many of these workflows apply both to in-house accounting functions and to external accountants.
Here are what we view as the jobs to be done
Data collection and ingestion
Finance professionals and accountants must gather data from many disparate sources — bank statements, the general ledger, commerce enablement platforms, AP and AR tools, the business’s system of record, etc. — to consolidate performance metrics and resolve any contradictory entries. This manual comparison of entries across data sources is known as “reconciliation,” and it usually requires input from different teams at the company (e.g., the individual who approves invoices or manages a specific vendor relationship). Reconciliation is why small businesses spend on average 15 hours a week on accounting-related tasks and why larger companies pay for armies of support via business process outsourcing (BPO).
Prior to the introduction of LLMs, advancements in Open Banking (led by companies like Plaid) and universal APIs like Rutter (which normalizes transactions from commerce, accounting, and payments data) were already making it easier for accountants to automatically import much of the data they need for reconciliation. With LLMs, however, teams can use LLM-powered data extraction software to pull data from unstructured file formats like contracts, receipts and invoices. This has far-reaching implications in both enterprise and SMB settings. Within the enterprise, data relevant to the finance team can finally be more centralized, and the combination of powerful LLM search and matching functions makes reconciliation, error checking, and term identification far easier. In an SMB setting, instead of emailing Becky from Procurement or sending a Slack message to Ryan in FP&A to understand why the payable that got logged in Quickbooks doesn’t match the money that actually left the Mercury account, accountants could conceivably interact with an AI copilot that’s trained on the company’s electronic communications and contracts to resolve the issue more quickly. Moreover, the software could also generate an audit trail to trace its work — a fundamental part of accounting. Basis, for example, provides a copilot that enables this matching between payables and the cash that left the bank. Klarity helps enterprise customers automate the workflow around review and extraction across document types.
Research
Research is a natural use case for LLMs in accounting. A big part of a CPA’s job is to determine how certain line items of revenue and expense should be classified, reported, and ultimately taxed. Some line items of cost, such as charitable donations, could be eligible for deductions, while others, such as R&D expenses, could qualify for tax credits. Accounting and tax teams work with a variety of codes and standards to proceed in their work: tax codes (which can vary wildly by jurisdiction), accounting standards codification (rules such as ASC 606, which covers revenue recognition), SEC filings (summaries of significant accounting policies), prior working papers, and even thousand-page guidance documents published by the Big Four accounting firms across various topics (e.g., technical accounting).
Before genAI, practitioners needed to run basic keyword searches through all these databases, at best, or more likely check a pdf manual or email a colleague, to answer their own questions and make determinations. But in a post-LLM world, purpose-built copilots — trained on the aforementioned data sets and all the precedents set by them — should be able to answer these queries deterministically. Over time, they can also be trained on judgment calls that fit the firm or professional. There are numerous startups offering solutions across this spectrum. SPRX and Neo.tax, for example, specialize in R&D tax credits, whereas Materia helps provide better guidance and accelerated research.
Report generation and filing
Once practitioners have categorized their clients’ data, they next need to analyze the data and produce internal and external reports. These can range from journal entries for the enterprise resource planning system (ERP) and disclosure reports, to audit checklists and technical accounting memos for tax-filing purposes. Much of this work can now be automated. In the tax example, while genAI may not be necessary for populating boxes on a templated form to file taxes, it can be immensely helpful when it comes to writing summaries. For example, all professional services firms typically try to do things in a distinctive style, whether it’s the format of their documents, the syntax of their correspondence, or the tone of their delivery. LLMs can easily be trained to summarize information in the style of a particular firm (or even an individual partner), and generate audit-ready reports and checklists for clients that look exactly like those previously prepared by hand.
Client service and advice
Providing analysis, support, and advice is arguably where genAI can potentially add the most net new value in accounting. Client service practitioners have fairly self-explanatory jobs: they summarize key performance data for their clients (be they internal clients, in the case of in-house finance teams, or external clients, in the case of accounting firms), answer ad hoc questions, and offer finance and tax optimization advice. Introducing GenAI into this equation is particularly interesting in two ways:
- In our conversations with accounting firms, one theme stood out: accountants are constantly looking for ways to convert their client relationships from annual, transactional touch points that center around tax season to recurring, ongoing engagements focused on optimizing the business throughout the year. If accountants are able to produce high-quality insights every month for their clients at scale using GenAI tooling, clients may find the additional advice good enough to pay more for. This seems particularly relevant for accounting firms that serve startups and SMBs, who likely don’t have sophisticated in-house finance functions and could use more regular financial analysis. Our portfolio company Black Ore helps accounting firms free up time by automating redundant tax preparation processes so that practitioners can double down on higher value client advisory work.
- To accomplish the above, accountants will need access to fine-tuned models that can combine industry-specific detail (e.g., tax laws and policies) with quantitative capabilities (e.g., calculating key business metrics, ratios, and the like). While general purpose LLMs excel at many of the written tasks needed for an accountant’s workflow, and can even do some of the math required via tools like Code Interpreter, they’re still a long way away from putting it all together (and ensuring that recommendations comply with applicable law!). This presents a clear opportunity for startups looking to build fine-tuned models trained on industry-specific data sets. The question will be how they can get the requisite data in the first place when starting from scratch.
Beyond the jobs to be done
Importantly, as we plan to explore further in a future post, genAI is not yet ready to replace two absolutely crucial capabilities in professional services: judgment and sales (although both can be AI-assisted). At the end of the day, it will still be up to the seasoned professional to win new business, sign their name next to a recommendation (even if said recommendation is 90% automated via AI), and stand by it. This has real implications for a company’s willingness to adopt these solutions, as judgment and sales are typically currently responsibilities reserved for senior staff, whereas day-to-day workflows are managed by individual contributors (ICs). Founders must be wary of this as they navigate two elements of their GTM strategy:
Incentive alignment with buyers
Similarly to Legal AI, Accounting AI threatens to cannibalize billable hours with greater efficiency. Therefore, what buyer personas and what business units AI companies pitch matter immensely. Tax departments, for example, typically structure their engagements by hourly rates. Auditors, on the other hand, often have fixed-fee engagements (i.e., no hours to worry about there!) Given these differences, the same end work product can command the same fee from the client whether you put in 1 hour or 100 (that is, until the AI-enabled price wars begin!).
Top-down vs. bottom-up adoption
Senior IC practitioners are likely not the right path in the door for sellers of GenAI accounting and workflow products. Although they stand to benefit from GenAI saving them or their junior staff hours of manual work, they also are likely fearful that an AI-native product could diminish the scope of their utility and/or eventually replace them.
Finally, there’s one more crucial intangible element at play here that may impact buying decisions at CPA firms: talent development. While the idea of automating away a lot of relatively expensive grunt work with software is attractive from a margin perspective, all senior accountants were once junior accountants, and the grunt work is how they learned the ins and outs of the job. Much as software companies train their sales reps through the Sales Development Representative (SDR) program, accounting firms rely on the analyst program to expose junior CPAs to the entire lifecycle of an engagement. Companies take great pride in their training programs, so we expect that in the short term, AI-native tooling will mostly seek to augment junior staff rather than entirely replace them.
Where are we now? A Generative AI in Accounting Market Map
If we haven’t yet convinced you that there’s a big opportunity here, look no further than the number of early stage startups building in this space. It’s nearly impossible to be comprehensive here given the sheer number of companies sprouting up every day, but below you’ll find dozens we’ve come across, segmented by core product focus and GTM approach.
Needless to say, if you’re building in this space (or think we got something wrong!), we’d love to chat. You can reach us at mandrusko@a16z.com and samble@a16z.com.
