Trullion tackles applying AI to accounting workflows

Most of us don’t think of accounting as a separate line of business. But accounting is big business, in and of itself.

The global accounting services market was expected to grow from $544 billion in 2020 to $573 billion in 2021 at a compound annual growth rate (CAGR) of 5.4%, according to the Accounting Services market global report by Research and Markets. The market is expected to reach $736 billion in 2025 at a CAGR of 6%.

This is a sizable market and is a target for digital transformation. The size of the accounting software market is projected to reach $22.9 billion by 2027, from $14.2 billion in 2020, at a (CAGR) of 7.1% during the forecast period 2021 to 2027, according to Valuates Reports.

The account software market includes companies like Microsoft, Oracle and SAP to Intuit and Xero, plus many others. Yet, there’s something none of these names are applying, says Isaac Heller: AI in accounting. That’s why Isaac Heller and Amir Bolto co-founded Trullion in 2019.

Trullion is attempting to unify “unstructured and structured worlds of accounting by reading PDF or Excel-based contracts and translating them into financial workflows, connected to the data source.”

Trullion has more than 100 clients, from pre-IPO to multinational Fortune 500s and audit firms that procure directly for their clients, and has crossed the seven-figure mark in recurring revenue.

In February, Trullion announced that it closed $15 million in Series A funding, co-led by Aleph and Third Point Ventures, with participation from existing investors Greycroft, and seasoned financial executives including Bob Mylod, Artie Minson, Jody Padar and Guzel Lumpkin.

Heller met with VentureBeat to discuss Trullion’s use of AI to innovate in the accounting software market.

Accounting is disconnected

Heller lives in the world of finance and accounting. After earning an MBA and working as a teaching assistant and guest lecturer at the University of Texas at Dallas, he learned about Generally Accepted Accounting Principles (GAAP) and how those big standards are used. He then made his way into the corporate world.

This is where he ran into more challenging standards and business models — things like revenue recognition, revenue accounting and lease accounting. After a stint at Visual Lease, a lease-optimization software provider, Heller was inspired to do the same for accounting. Heller describes accounting as three parts that are disconnected:

The first part is the corporate accounting world. They might be doing some of their processes in spreadsheets or offline. The second part is ERP — SAP, Oracle and WorkDay. And then the third part is the audit leaders and the audit team.

There are many reasons why the ecosystem is getting more manual and riskier. “The question is, how could you make it more accurate and transparent? And the way to do that, in our view, would be to find ways to connect those three parts, meaning everyone in all parties could connect on one source of truth,” Heller said.

“But in order to do that, if you tried to come to market with a heavy tool that did it all, it wouldn’t be successful. Or you’d have to spend $100 million just to start. So what we do is focus on a pain point. In this case, we focus on things like lease accounting and revenue recognition,” Heller said.

Accounting and finance teams are still bound by legacy processes, including manual data entry, spreadsheets and PDFs, which mean that they’re at high risk for critical and costly errors. Meanwhile, continually evolving standards – including new lease accounting standards – are estimated to have brought $2 trillion of lease liability to S&P 500 balance sheets, adding further complexity and potential data inaccuracy.

Trullion has been emphasizing evolving standards as a way to raise awareness, drive innovation and create a sense of urgency. Heller likened esoteric accounting standards such as ASC 842 and IFRS 16 to GDPR in that respect. The end goal, he said, is more transparency on the balance sheet.

Diving deeper on ASC 842 highlights the reasons why Trullion thinks applying AI to accounting, and lease accounting specifically, is a good idea. ASC 842 defines leases as contracts, or portions of contracts, granting “control” of an identifiable asset for a specific period of time in exchange for payment.

Trullion notes that this means that contracts have to be carefully scanned in order to identify leases as per the above definition. That also includes contracts which may not be about leasing per se. That poses an interesting problem, for which AI may lend a helping hand.

Applying AI for accounting

Before getting to AI, there are a lot of data pipelines and data stores that have to be in place – in other words, creating the “source of truth” that Heller referred to.

Oftentimes, when you rely on an ERP or an old database as a source of truth, it doesn’t have any of the actual source within it, Heller said. With source-based accounting, that means more than PDF documents, he said. It could also mean Excel files, or systems like Salesforce or QuickBooks. Trullion ingests data from all those sources, and normalizes it into one source-based accounting system.

Of course, it’s not as easy as it sounds. Relevant data from all those sources has to be parsed. In many cases, that includes scanned hand-written documents, which is why Trullion has worked on honing its OCR. But to do things such as identify lease definitions in a contract, it takes some state-of-the-art technology.

Trullion’s whitepaper cites named entity recognition (NER), which is a technique used to identify specific instances of interest in data sources. To build this, and even to normalize ingested data into one source of truth, a domain entity model is needed. Trullion’s whitepaper refers to combining deep learning with domain-specific ontologies.

In fact, as Trullion cofounder and CTO Amir Bolto confirmed, it takes much more than NER to be able to identify leases in contracts. Bolto referred to pre-processing, which helps identification by running the data through multiple steps, such as data cleaning, data integration, data transformation and more before it goes through NER.

Bolto said that this helps improve the accuracy and reduce errors in the final result, even though there aren’t currently any publicly available performance metrics or ontologies to share. Trullion’s solution for identifying leases in contracts has been in development for about a year.

“There’s a lot of startups out there that do sales software,  and marketing software and, customer support software, and those are use cases that we all have experience in early-stage companies. But accounting — that’s big business. That’s the mature companies. It’s not as well represented with startups,” Heller said.

There are some use cases where AI is applied, like accounts payable and invoice automation that are more common for us. “I don’t really know why, but we just haven’t seen a lot of AI. We saw a lot of people talking about it. People wrote some really nice papers about it, but we didn’t see it practically applied. It’s definitely harder than everyone thinks,” Heller said.

Trullion doesn’t claim to wave a magic wand and get 100% of a document extracted. But if you can connect to the actual financial workflow and get something like 70% improvement, that’s still a big deal, Heller said.

Heller said that what Trullion does is complementary to solutions like QuickBooks, Xero and ERP software, in the sense that lease accounting is a gap in their ecosystem. The differentiation with more specialized vendors is the use of AI. Trullion, which has a presence in New York and Tel Aviv, Israel, will use its Series A funds to hire across a variety of engineering, sales and marketing roles, as well as accelerate product development.