Company Profile / Fintech x AI
Neo.Tax
The startup teaching machines to read the tax code - and turning your engineers' commits into audit-ready tax credits.
The startup teaching machines to read the tax code - and turning your engineers' commits into audit-ready tax credits.
It is the week before a corporate tax deadline. Somewhere, a controller is not staring at a spreadsheet of engineering hours, not chasing twelve developers for survey answers they half-remember, and not praying the numbers survive an audit. That controller is using Neo.Tax. Software is doing the staring instead.
Neo.Tax sells one deceptively simple promise: connect the systems your company already uses - GitHub, Jira, payroll, the general ledger - and let a set of AI agents figure out which work qualifies for the R&D tax credit, calculate it, and write the documentation an auditor will accept. No interviews. No FIN48 theater. The work already happened; the software just reads the receipts.
We want to apply machine learning to taxes, upgrading them from an ancient pain into a modern advantage.- Neo.Tax, on its founding idea
The U.S. R&D tax credit is real money. For tax year 2024 it is projected to top $17 billion nationwide. A software startup can claim up to $250,000 against payroll taxes. The catch is that claiming it has historically required surveys, interviews, consultants, and a stack of documentation thick enough to discourage most people from bothering.
Then Congress made it worse. Changes to Section 174 forced companies to capitalize and amortize R&D expenses instead of deducting them, turning a once-friendly line item into a compliance maze. The credit was sitting on the table. Most companies, understandably, decided the reach wasn't worth the paper cut.
The credit is fairly accessible to software startups. The process of claiming it is the part that's daunting.- The gap Neo.Tax was built to close
The team that came together in 2020 is almost suspiciously well cast. Ahmad Ibrahim, the CEO, spent his career in tax tech and a stint as a product manager at Intuit - the company that taught America to fear April a little less. Firas Abuzaid brought a Stanford machine-learning PhD, the part of the equation responsible for making software actually understand engineering work.
And then there is Stephen Yarbrough, who used to run R&D tax credit audits for the IRS. The man who once decided whether your claim survived now helps build the claims. It is the corporate equivalent of hiring the goalkeeper to design your penalty kicks. Their bet: the credit isn't hard because the rules are unknowable - it's hard because nobody had wired the rules to the data.
Neo.Tax is like having a former IRS Agent walking our customers through the process.- Steve Yarbrough, Co-Founder & Head of Tax
Pictured in spirit: three people who agreed that the tax code is a database problem wearing a paperwork costume.
Neo.Tax's platform runs a set of AI agents, each trained on IRS code, regulations, and case law, continuously across whatever engineering and financial signals a company can offer. They qualify projects under Section 41, handle the Section 174 capitalization math, and classify software work for ASC 350-40 financial reporting. The output is a single source of truth for every R&D and software-cap decision.
Audit-ready documentation assembled automatically from engineering and payroll data - the survey replaced by the source.
Automated calculation and documentation for the capitalize-and-amortize rules that turned R&D accounting into a headache.
Classifies engineering effort for software capitalization so the financial close stops being a dreaded annual ritual.
Six agents reading 20+ passive signals from GitHub, Jira, payroll, and the ledger - continuously, not once a year.
The unglamorous miracle: software that finds your tax credit by reading commit messages nobody else wanted to read.
The company is founded with a thesis: the tax code is a machine-learning problem hiding behind a wall of paperwork.
Uncork Capital, Floodgate, Liquid 2 Ventures, and Lux Capital back the first product: automating the startup R&D credit.
The team grows from 9 to 20 people as startups discover claiming the credit no longer requires a consultant on retainer.
Infinity Ventures leads, with GV, Acrew, Fin Venture Capital, and earlier backers joining. Total raised reaches $13M.
AI agents expand across Section 41, 174, and ASC 350-40; a Thomson Reuters partnership opens the enterprise door. SOC 2 Type II in hand.
Skepticism is the correct response to any sentence containing both "AI" and "taxes." Neo.Tax seems to know it. Rather than promise magic, the company publishes accuracy figures for its workflow and leans on the least exciting credential in software: it has actually survived enterprise procurement, complete with SOC 2 Type II certification, single-tenant deployments, and closed-off LLMs with no data co-mingling.
Note the 74%. A tax-AI company that admits one number is harder than the others is, oddly, the more believable one.
The only AI in tax that has actually been through enterprise procurement.- Neo.Tax, on why the boring credentials matter
The customer list does the rest of the talking. Early on it was the startup crowd - Pipe, Stedi, Taika, Casa, Hatch. The company now points to enterprise tax teams at names like Adobe, Block, Capital One, SoFi, and partner Thomson Reuters. The through-line: the bigger the engineering org, the more credit gets left on the table, and the more a machine that reads everything is worth.
Ask the CEO what's hard about tax AI and he doesn't say accuracy or scale. He says trust. It's one thing to let software draft a marketing email; it's another to let it sign off on what you tell the IRS. Neo.Tax's whole design - the case-law training, the audit-ready paper trail, the closed LLMs, the former auditor on the founding team - is an argument that the machine can be trusted with the highest-stakes paperwork a company files.
That's the real mission hiding under the tagline. Not "do taxes faster." Make a category of work that has always required blind faith in a consultant instead require something a CFO can verify, line by line, and defend in an audit.
The R&D credit is a beachhead, not the destination. If software can be trusted to read a company's raw activity and produce a defensible tax position for the single most scrutinized incentive in the code, the same approach extends to nearly every determination a tax department makes. Continuous, automated, audit-ready compliance stops being a slogan and starts being the default.
So return to that controller, the week before the deadline. The spreadsheet of engineering hours is gone. The survey nobody filled out is gone. The consultant's invoice is gone. What's left is a filing the software built by reading work that already happened - and a person who got to spend the week doing something other than reconstructing the past. Neo.Tax didn't make the tax code simpler. It made the tax code readable by machines, which, for the people who file taxes, amounts to nearly the same thing.
The work already happened. Neo.Tax just reads the receipts - and hands you a credit the IRS will accept.- The closing argument