Tax Answers You Can Verify: Why We Built SmartResearch

Today we released CereTax SmartResearch. It answers sales and use tax questions from two sources only: determinations our tax researchers wrote and the primary law behind them. Every answer cites that law, and SmartResearch tells you when the sources do not support one. Here is the short version. The rest of this post is why we built it the way we did.

A wrong tax answer shows up years later

Most tax questions sound simple. Is this service taxable in Texas? Under which tax? At what rate? As of which date?

The answer sits in a statute, a regulation, a letter ruling, and a state publication, and each of those changes on its own schedule. Someone on your team reads them, writes a memo, and moves on. A year later the memo is still in the shared drive, and nobody knows whether the law under it changed.

That answer then goes into your tax engine. If it is wrong, it charges the wrong tax on every transaction until someone notices. Often the someone is an auditor, three or four years on, with interest and penalties attached.

I have spent my career on this problem. I co-founded SureTax and Tax Partners before CereTax. The research has always been the slow, costly part, and the part a tax team has to defend.

An auditor will never accept "the AI said so"

AI should make tax research faster, and plenty of tools now promise it. Most of them fail a tax team in one of four ways.

• They read the open web. A general AI tool draws on blog posts, vendor pages, and forum threads, along with the law. It cannot tell you which one shaped the answer.
• They cite nothing you can check. An answer with no statute, ruling, or publication behind it gives you nothing to show an auditor or your CFO.
• They ignore the date. Tax law changes. A tool that answers with today's rule, when you asked about last March, gives you the wrong answer to your question.
• They always answer. A model built to be helpful will produce a confident answer when the sources run thin. In tax, a confident wrong answer costs more than no answer.

AI can do this work well when it has limits, and when the tax team can see them.

SmartResearch is not the first AI tax research tool on the market, and a quick search turns up plenty of others. Most start from a model and go looking for content. We started from the content our tax researchers built, added the primary sales and use tax law as sources, and instructed the models to work inside both. Our customers and partners rely on us to understand complex tax and calculate it correctly, and a research tool from CereTax has to get the answer right and show the law behind it. Building that took time, and SmartResearch is ready now.

Meet SmartResearch

SmartResearch is AI tax research with cited answers. Ask whether a product or service is taxable in a jurisdiction, under which tax, at what rate, and as of which date. You get a direct answer, a short summary, the controlling law, and a numbered citation for every claim.

Click a citation and you see the source itself: the CereTax determination first, then the statutes, rulings, and state guidance behind it.

Taxability answers are the center of SmartResearch, and six more modules sit around them:

• Matrix Builder: build a taxability matrix across products and jurisdictions, and export it.
• Nexus: see economic nexus rules by state, with the law cited.
• Tax Rates: look up rates from the engine that calculates the tax, down to the rooftop.
• Rate Files: download rate files in the coverage and format you pick, with sales and use rates in separate columns.
• Pulse: get daily briefings on legislative, court, and agency changes, set to what you follow.
• Library: read the forms and primary law text behind every citation.

How SmartResearch keeps the AI inside the law

We built SmartResearch around the four failures above. Each answer goes through the same steps.

1. It starts from a researcher's determination. CereTax tax researchers write the determinations. SmartResearch begins with the one that fits your question.
2. It reads only two sources. The models work from that determination and the primary law CereTax holds word for word. They never search the open web.
3. The models must agree. Models must agree on a cited answer, and a separate check tests that answer against the same sources.
4. It answers as of your date. We keep each version of the primary law we hold, so a team can see what the law said as of the date in question.
5. It tells you when it can't answer. When the sources do not support an answer, SmartResearch tells you, and you can take the question to our tax team.

The result is an answer you can verify. You can read the cited law yourself before you act on it, and so can your auditor.

All of this draws on hundreds of thousands of law documents, versions of primary law, and researcher-authored determinations. It is the same content that runs our calculation engine and SmartExempt, so all three products share one record behind every answer.

Who it's for

We built SmartResearch for the people who have to stand behind a tax answer.

• Corporate tax teams who decide how products and services are taxed and need to show the law behind each decision.
• Accounting firms and tax practitioners who answer the same questions for their clients and put their name on the advice.
• Finance leaders who own the audit risk and want to know an answer has a source before it goes into the tax engine.

If you already use CereTax, ask your customer success contact about access.

Where SmartResearch fits at CereTax

CereTax has been an AI-native company since our early years. SmartResearch is the most complete result of that work so far.

How we build. SmartExempt, our exemption certificate product, was our first AI product. It uses AI to read exemption certificates and pull out the data it needs. We have built every product since on the same base. AI also writes much of our code, and our engineers review and test it before it ships. It is the same rule we hold SmartResearch to.

How we sell. Our marketing and sales teams use AI every day to research a prospect's industry, draft, build pages, and answer the questions buyers ask before they talk to us. We get more done with fewer people than at any company I have run before.

How we deliver. Sales tax is a hard place to use AI, so we do not let a model answer on its own. Our researchers write the determinations. The AI works from those, cites the law, and tells you when the sources do not back an answer. The customer gets the researcher's work faster, with the law attached.

The calculation engine, SmartExempt, and SmartResearch now share one set of researcher-authored content and one record behind every answer. We plan to keep building on it, and we are not done.

See SmartResearch for yourself. Get a demo.