What a 'Warranted Accountant' Actually Means — and Why Your Tax Layer Needs One
There's a word that does a lot of quiet work in tax: warranted. It means a credentialed professional has put their name — and their professional standing — behind a piece of work. If it's wrong, there's a real person, with a real licence, who is accountable.
This is the thing AI cannot manufacture, and it's exactly the thing tax requires.
"An app told me" is not a defence
You can build a tax engine that's right 95% of the time and presents every answer with total confidence. Your users will love it right up until the 5%. When a tax authority disputes a filing, "the software calculated it" carries no weight. The obligation sits with the taxpayer, and behind any reliable filing there is, by design, a credentialed human who reviewed it and is accountable for it.
So a tax layer that only outputs AI numbers isn't a tax layer — it's a very confident calculator with a liability problem. The moment a user files something or makes a decision based on it, the absence of a real accountant stops being a UX detail and becomes the entire risk.
Where the human actually sits
The mistake is thinking "add an accountant" means "put a human in front of every answer." That's slow, doesn't scale, and isn't what most questions need. The right model is layered:
- Most questions never need a human in real time. "What's the higher-rate threshold?" or "roughly how much should I set aside?" runs on verified rules and answers instantly. The accountant's work is already baked in — they verified the rule the answer is built from.
- Anything filed or relied upon gets a named accountant. When a user is about to submit a return, claim a relief, or act on a number that has consequences, a credentialed professional reviews it and stands behind it. Not a faceless "reviewed by our team" — a specific, named, licensed person.
In our model these are the same people, doing two kinds of work: verifying the rules once, so they're trustworthy at scale, and reviewing individual filings where a human signature is what makes the output usable.
What "verified" means here (and what it doesn't)
We're deliberate about language, because the words carry risk.
- Source-cited draft (Q2): a rule that's been researched and cited to its source, but not reviewed by a credentialed accountant. Useful, honest, and clearly labelled as unreviewed.
- Accountant-verified (Q1): a credentialed CPA or EA (or equivalent) has reviewed the rule against the source and stands behind it.
We never call a draft "verified," and we never call a verified rule "certified," "approved," or "guaranteed." A tax layer that blurs those lines is selling confidence it hasn't earned — and that's the failure mode that ends in a furious user and a regulator's letter.
Why this is the moat, not the friction
It's tempting to treat the human as the slow, expensive part to engineer around. It's the opposite: the credentialed human is the only part a competitor can't clone overnight, and the only thing that makes "we'll handle your tax" a promise rather than a liability. A network of real accountants standing behind the rules — and available to review what gets filed — is what separates a tax layer from a tax toy.
For a platform, the good news is you don't build that network. It already exists. OpenAccountants connects the verified rules and the credentialed accountants behind them into one integration, so the human accountability is part of the layer you embed — not a department you have to hire.
If you only take one thing from this: an AI can give your users a number. Only a warranted accountant can let them rely on it. Build the second thing, and the first thing becomes safe to ship.
Curious how the verified rules and the accountants behind them fit into a product? Read the three-layer model or explore the platform integration.