Every finance leader I talk to is trying to move faster with AI. For a lot of them, that starts with a hard look at their tax stack, and that’s often how they end up talking to us. Once they start comparing vendors, they find a lot of different versions of “AI in tax,” and they’re hard to tell apart.
The legacy players have bolted AI onto products that were built long before anyone was thinking about it. The newer “AI-native” companies have built AI, but only for a slice of the compliance chain. The rest gets handed off somewhere else, and that handoff is where accuracy slips, things slow down, and the numbers stop reconciling. Both can look great in a demo.
What’s hard to see in a 30-minute sales call is where the AI actually lives and what it can see. AI is complicated, and it gets even harder to evaluate when it runs through every layer of a platform, which is how we built Anrok. I think that’s a big part of why half of the Forbes AI 50 run their tax on us. Our customers include Anthropic, Cursor, Notion, Harvey, Perplexity, Webflow, and many more. They understand better than anyone what AI can and can't do.
Where sales tax got harder
The fundamentals of sales tax and VAT haven't changed much. Every company still has to figure out where it owes, get registered, charge the right amount, file and pay, and then prove it was right. I think of that as a chain, because each step depends on the one before it. Miss a registration, and everything after it in that state is wrong too.
Most companies still run that chain across a few tax tools, a filing firm and a spreadsheet, and tax dollars fall through at every handoff. That's how tax that should have been a pass-through ends up coming out of your margin, usually with penalties and interest attached.
That's the whole reason we built Anrok: one platform for the entire chain, with AI built into every step.
As fast as companies move today, I don’t think that’s optional anymore.
Three questions that separate AI in tax
You can sort tax vendors a lot of ways: old and new, US and global, enterprise and startup. I don't think those labels help much for understanding AI. When I look at how any tax platform uses AI, I ask three things.
Does it run the entire chain? Most platforms are strong at one link, usually calculating the rate, and hand the rest to separate modules, partners or a filing firm. AI can only work with what it can see. If monitoring, filing and reconciliation live in different systems, the AI is answering questions about a slice, and nobody checks what you charged against what you filed and paid.
Does it show its work? A tax answer is only as good as the reasoning behind it. When AI tells you a product is taxable in Texas, you should be able to see why, including the rule it applied, the source it came from, and the transactions it looked at. The same goes for every number on a return. If you can trace it back to the invoice and the rule behind it, you can defend it. If the AI just hands you an answer, someone on your team ends up redoing the research by hand when an auditor asks.
Who stands behind the answer? AI is good at research and at spotting things that look off. It shouldn't be guessing a tax rate. I'd want a rules engine on the calculation and real tax experts reviewing every rule change, so someone is accountable when an auditor asks.
Most of the market can say yes to one of these. Leading AI in tax has to say yes to all three.
AI in every layer
We built Anrok to say yes to all three, and that meant putting AI into every layer of the platform. If the compliance chain is the work, the layers are where the AI sits inside it. In Anrok, there are five.
| Layer | What our software does | Where our AI works |
|---|---|---|
| Data | Brings in every transaction from your billing, ERP, and HR systems. | Reads exemption certificates from PDFs and photos, and helps catch bad addresses, invalid VAT IDs, and unmapped products on the way in instead of at quarter end. |
| Engine | Holds the rules for 11,000+ US jurisdictions and 150+ countries, and puts the right tax on every transaction in real time. | Monitors tax authorities for new rates and rules, and drafts the updates, including which products they affect. Our tax experts review every change before it ships. |
| Operations | Registers, files, remits, and reconciles what you charged against what you filed and paid. | Runs and maintains filing automations across jurisdictions and flags exceptions, with our tax experts on the hard ones. |
| Your team | Gives finance and tax a clear view of exposure, filings, and discrepancies. | Anrok Atlas and MCP answer tax questions grounded in your own data, show its reasoning and sources, and can investigate a discrepancy or rank your exposure in one query. |
| Your agents | Lets the tools you already use act on your tax data. | The Anrok MCP server and our APIs put Anrok inside Claude, ChatGPT, Cursor, or the agents you build. |
One place you won't find an AI model is the moment of calculation. We don't let AI guess a tax rate. It keeps the rules current and flags things that look off, with a human in the loop, and then a rules engine applies them the same way every time.
All five layers write to the same record. We're the only tax platform with reconciliation built in, so what you charged, what you filed, and what you paid sit in one place and have to agree. That's a big part of why I trust our AI with real work. It's looking at everything, not a slice.
The MCP server we shipped this month is a good example. Connect Anrok to Claude, ChatGPT, or Cursor, ask it to reconcile your billing system against Anrok within your AI assistant. Just a few minutes later, you've got the answer and the transactions behind it. Webflow's tax team took their monthly Stripe-to-Anrok reconciliation from four hours down to 30 minutes. It only works because the data was already complete and matched. Point an agent at a messy record, and it just gives you the wrong answer faster.
Anrok Atlas, which is in open beta, is for the people on your team who'd rather just ask within Anrok. It answers from your own transactions, shows its work, and cites where the answer came from, using a knowledge base our in-house tax team built and Big Four CPAs reviewed. Ask why tax in one state jumped last quarter and you get the reason, plus the transactions that caused it. You can hand that to a financial statement auditor.
We run the company the same way we built the platform. Our engineers and our tax ops team use AI every day, and it's a big reason a rule change in one state turns into tested, working software quickly. The law doesn't slow down for us, so we don't slow down either.
Trusted by the people building AI
More than 1,400 companies now run on Anrok across 150+ countries, with more than $100B in processed volume. Their achievements prove the difference between a bolted-on AI solution and one that's built into the platform.
For example, Webflow's old tax engine led their team to pay tax out of the business' own pocket. Christie Perez, who's been in indirect tax for 18 years, switched Webflow to Anrok and stopped leakage while cutting down their time on tax. "Anrok shaved off a good two to three days a month," she told us. Nine months in, her monthly analysis hasn't found one discrepancy between what Webflow charged and what it filed. "The analysis we've done each month since putting Anrok in place shows very little coming out of pocket," she said. That's what a pass-through is supposed to look like.
Synthesia, the AI video company, was running two tax systems side by side. Today one accountant handles their global sales tax and VAT in about four hours a month. "Having everything in one platform gave us a great warning about geos where we're getting close and should think about registering," said Jeff O'Connor, Head of Accounting.
People assume switching is painful, but Notion was on Anrok in three weeks. The benefits from the time you get back compound. Statsig's founder and CEO, Vijaye Raji, told us, "Anrok has cut the time we spend on sales tax by 90%."
Where this goes next
The promise of AI gets tested in tax because the answers have to be right and the money is real. If the data is right, books close with tax reconciled, you get an alert the day you launch in a new market and a board question can get answered in seconds, with the work shown.
If you're evaluating tax platforms right now, take those three questions into every demo. You'll learn a lot from how each vendor answers them.
