All founders love “free” money, but with the global pandemic going on, the necessity of free money has taken on a whole new meaning this year. First, there was the scramble to secure PPP loans a few weeks back for U.S.-based startups, and then the second wave of PPP loans when Congress offered a second tranche of funding. Two weeks ago, I covered a company called MainStreet, which is helping startups apply for local economic development credits which cities offer to businesses relocating to their regions.
In the same vein, Neo.tax wants to help startups secure R&D research credits from the federal government — which tend to be fairly easy to acquire for most software-based startups given the current IRS rules for what qualifies as “research.”
The free money is good, but what sets this startup apart is its ambitious vision to bring machine learning to company accounting — making it easier to track expenses and ultimately save on costs.
It’s a vision that has attracted top seed investors to the startup. Neo.tax announced today that it raised $3 million in seed funding from Andy McLoughlin at Uncork Capital and Mike Maples at Floodgate, with Michael Ma at Liquid2 and Deena Shakir at Lux Capital participating. The round closed last week.
Neo.tax was founded by Firas Abuzaid, who spent the past few years focused on a PhD in computer science from Stanford, where he conducted research in machine learning. He’s joined by Ahmad Ibrahim, who most recently was at Intuit launching small business accounting products; Stephen Yarbrough, who was head of tax at Kruse Consulting, a popular consultancy for startups on accounting and financial issues; and Leonardo De La Rocha, who was creative director of Facebook Ads for nearly five years.
Neo.tax’s Stephen Yarbrough, Firas Abuzaid, and Ahmad Ibrahim. Photos via Neo.tax
Or in short, a perfect quad of folks to tackle small business accounting issues.
Neo.tax wants to automate everything about accounting, and that requires careful application of ML techniques to an absolutely byzantine problem. Abuzaid explained that AI is in some ways a perfect fit for these challenges. “There’s a very clearly defined data model, there’s a large set of constraints that are also clearly defined. There’s an obvious objective function, and there’s a finite search space,” he said. “But if you wanted to develop a machine-learning-based solution to automate this, you have to make sure you collect the right data, and you have to make sure that you can handle all of the numerous edge cases that are going to pop up in the 80,000 page U.S. tax code.“
That’s where Neo.tax’s approach comes in. The software product is designed to ingest data about accounting, payroll, and other financial functions within an organization and starts to categorize and pattern match transactions in a bid to take out much of the drudgery of modern-day accounting.
One insight is that rather than creating a single model for all small businesses, Neo.tax tries to match similar businesses with each other, specializing its AI system to the particular client using it. “For example, let’s train a model that can target early-stage startups and then another model that can target Shopify businesses, another one that can target restaurants using Clover, or pizzerias or nail salons, or ice cream parlors,” Abuzaid said. “The idea here is that you can specialize to a particular domain and train a cascade of models that handle these different, individual subdomains that makes it a much more scalable solution.”
While Neo.tax has a big vision long-term to make accounting effortless, it wanted to find a beachhead that would allow it to work with small businesses and start to solve their problems for them. The team eventually settled on the R&D tax credit.
“That data from the R&D credit basically gives us the beginnings of the training data for building tax automation,” Ibrahim explained. “Automating tax vertical-by-vertical basically allows us to be this data layer for small businesses, and you can build lots of really great products and services on top of that data layer.“
So it’s a big long-term vision, with a focused upfront product to get there that launched about two months ago.
For startups that make less than $5 million in revenue (i.e. all early-stage startups), the R&D tax credit offers up to a quarter million dollars per year in refunds from the government for startups who either apply by July 15th (the new tax date this year due to the novel coronavirus) or who apply for an extension.
Neo.tax will take a 5% cut of the tax value generated from its product, which it will only take when the refund is actually received from the government. In this way, the team believes that it is better incentive-aligned with founders and business owners than traditional accounting firms, which charge professional services fees up front and often take a higher percentage of the rebate.
Ibrahim said that the company made about $100,000 in revenue in its first month after launch.
The startup is entering what has become a quickly crowded field led by the likes of Pilot, which has raised tens of millions of dollars from prominent investors to use a human and AI hybrid approach to bookkeeping. Pilot was last valued at $355 million when it announced its round in April 2019, although it has almost certainly raised more funding in the interim.
Ultimately, Neo.tax is betting that a deeper technical infrastructure and a hyper-focus on artificial intelligence will allow it to catchup and compete with both Pilot and incumbent accounting firms, given the speed and ease of accounting and tax preparation when everything is automated.