Accrual, a tax prep and review platform for firms that launched today with $75 million in funding, supports every federal and state tax form, as well as the unstructured data to fill it, according to its co-founder and CEO Cosmin Nicolaescu.
"What kind of forms do we support? The TLDR is we support all the forms. Every federal form, state form. But we also support any unstructured data. So you can give us emails, spreadsheets, handwritten notes, checks — we have people that give us donations, notes, expenses, literally anything and everything — and the model is smart enough to interpret what it is, and know how to use that information throughout the tax process," he said in an interview.
He said this is because their AI-native platform does not rely on optical character recognition and template-based form mapping, which involves extracting data from fixed locations on tax forms. Nicolaescu found this approach inefficient, as you need to program a different procedure for every single form, which he said can introduce a higher rate of error than he is comfortable with. Instead it relies on a combination of AI models that ingest and analyze the data then applies it to the proper form on their own.
"OCR is, for lack of a better term, kind of a dumb extraction. You extract some text from these coordinates. What we try to do is say, 'OK, what is this actual document? OK, this is a W-2 document, and that means you have an employer. Who is your employer?' And we kind of build a graph around the client on all that knowledge. And by using that foundational block, it allows us to take any type of document, because models are very smart right now, to be able to understand the meaning of that document," he said.
He said the AI is 100% accurate, versus OCR-based solutions that he said top out at around 98% accuracy. The models make multiple passes over their outputs to ensure fidelity and reliability.
"We will be 100% accurate, because I think anything else is kind of unacceptable and requires you to validate every single piece of document," he said.
In general, the user will either import the previous year's return or onboard a new client with the relevant documentation. The model takes the information to first generate a pro-forma return. Then it examines the data and asks questions based on it before constructing a completed return and flagging what needs a follow-up; it also produces a comparison from last year so people can see where the biggest differences are. Noting that many accountants dislike the black box aspect of AI, Nicolaescu added that the software shows its work and even provides references that, when clicked, takes the user to the exact place the model acted.
"We will also give full references. So, not only explain [the reference], but in a single click you can navigate to exactly the box that our system took the form for. This allows users to audit things in place at the end of the run," he said
This does not just apply to simple forms that can be done assembly line style but even more complex ones like the K-1. In fact, according to Nicolaescu, this is where Accrual started. Rather than begin with a simple product and gradually scale up in complexity, its approach was to start with some of the most complex use cases, based on the idea that if you can master a complex thing early, mastering simpler things later will be much easier.
"It is very hard to go from simple to complex in this space… I think it is very hard to say, 'Oh, I'm going to build [a product for], let's say, W-2s and 1099s, and then a year from now K-1s. It's a different beast," he said.
Due to this philosophy, the company has started with selling to top 50 accounting firms that do large numbers of highly complex engagements.
"The reason why we started with the top firms is because solving for the complex use cases actually gets you the simpler ones by default," he said. Accrual is only starting there, and ultimately Nicolaescu would like all kinds of firms to be using his solution.
Because accounting firms are the primary market for the solution, the billing is based on the firm's own engagement fees. Rather than charging per return or per user, which Nicolaescu said can lead to unpredictability in costs, Accrual aligns charges with the firm's margin, which he felt was much simpler.
"We are taking an enterprise approach with each firm that we work with: we ask them for the
complexity matrix for their returns since they each have different matrices. From there, we generally do two or three tiers of complexity that we price according to their engagement fees," he said.
All of this, ultimately, is not to replace accountants but allow them to spend more time solving problems and less time filling in boxes. While Accrual automates a lot of things, Nicolaescu noted that it does not automate 100% of them, though it really doesn't need to, as he stressed that accounting remains a relationship-driven profession. The point, he said, is not for accountants to press a button and have an AI do all their work for them, but to deepen client relationships in a way that ultimately helps not just individuals but the profession as a whole.
"Professional services is a relationship business. The firms that we work with have worked with some of their customers for literally decades and there is a foundation of deep trust. Our view is how do we empower accountants to focus their time and attention on things that really move the needle for their customers?" he said
Accrual currently integrates with CCH Axcess and is exploring further integrations, with the intention of working with the tax engines their clients demand. The funding round was led by General Catalyst, with participation from Pruven Capital, Edward Jones, and a group of industry executives and founders. The funding will support continued investment in product and AI development, team expansion and onboarding accounting firms as the company scales. Accrual is actively onboarding accounting firms across the U.S. and plans to expand the platform beyond individual returns over time. The company is also hiring engineers, product leaders and accounting experts to help build out the infrastructure.
