AI tax software for in-house tax teams: research, indirect tax and transfer pricing (September 2026)

Three jobs AI tax software does for a tax department

Most tools in the category do one of 3 jobs. Research assistants take a question in plain language and return an answer with citations to statutes, case law or commentary. Indirect tax engines sit between the ERP or billing system and the tax authorities: they calculate VAT or sales tax per transaction, prepare filings and now use AI to classify products and explain results. A third group covers specialist work such as transfer pricing research and return preparation in accounting firms.

Our directory sorts its 62 tax tools along the same lines. A few are bookkeeping tools for small businesses that we also file under tax, such as sevdesk and FreeAgent; they serve founders and freelancers more than a corporate tax department, so this guide mentions them only in the data.

The explorer below shows the category as of September 26, 2026. Filter by main job or by headquarters region to see how many tools publish a price, offer a free trial or keep data in the EU. The research tools from German tax publishers pull the share of published prices up, the enterprise engines pull it down.

Sources: [1][2][3]

Why tax departments are automating now: nexus, e-invoicing and the minimum tax

Three rule changes created much of the volume these tools handle. In South Dakota v. Wayfair, decided on June 21, 2018, the US Supreme Court allowed states to require sales tax collection from sellers without a physical presence. South Dakota's law covered sellers that deliver more than $100,000 of goods or services, or 200 or more separate transactions, into the state each year. Since then sellers track thresholds state by state, and Vertex says its platform covers more than 20,000 jurisdictions.

In Europe, the VAT in the Digital Age package (ViDA) was adopted on March 11, 2025, and entered into force on April 14, 2025, from which point member states can introduce mandatory e-invoicing under specific conditions. Digital reporting requirements follow from July 1, 2030, with the main impact on businesses making cross-border B2B supplies, and national real-time reporting systems must be brought into line with the EU system by January 1, 2035. The Commission expects the package to cut VAT fraud by up to 11 billion euros a year.

For large groups, Council Directive (EU) 2022/2523 sets a 15 percent minimum effective tax rate for groups with combined revenues above 750 million euros a year, applicable to fiscal years starting in 2024. Each of these rules means more data per transaction or per entity, and that is where automation pays.

Tax leaders expect change to come fast. A survey of 170 corporate tax leaders by the Thomson Reuters Institute and Tax Executives Institute, published on March 16, 2026, found that 67 percent said technology investment had let their department shift toward more proactive tax work. Respondents expect AI to become central to their workflows within 1 to 2 years, although current use was still described as relatively low and fewer than half had received a budget increase.

Sources: [4][5][6][7][8]

Tax research assistants: cited answers, few published prices

Research assistants are the most visible use of generative AI in tax. They answer a question in plain language, cite the sources behind the answer and draft memos or client letters from the result. The value depends on the content underneath, which is why tax publishers and younger AI companies compete on sources as much as on models.

Blue J covers US federal and state and local tax, builds on primary law and Tax Notes material and updates its database daily. It is one of few that publish a price: $1,498 per user per year on the Individual plan, with a 7-day free trial without a credit card, while the Team plan is quoted. TaxGPT sells 3 plans (Pro, Pro Plus and Enterprise) priced by team size, number of clients and capabilities, with no amounts on its pricing page.

Thomson Reuters builds Checkpoint Edge on 12 million expert-authored content pieces maintained by more than 680 editors, according to its product page, and quotes prices through sales. German tax publishers add AI to their own databases: Haufe includes CoPilot Tax at no extra cost in all versions of Haufe Steuer Office, and Deubner TAX KI draws on more than 55,000 tax expert opinions for a flat 99 euros a month per firm.

Treat every answer as a draft, check that the cited source says what the answer claims and keep that check in the file.

Sources: [1][9][10][11][12][3]

Indirect tax engines: where the AI sits in VAT and sales tax

Indirect tax engines were rules engines long before generative AI. The AI layer now takes the parts that rules handle badly: sorting products into tax codes, spotting unusual transactions and explaining why a rate applied. The large vendors all describe AI features, but they place them in different parts of the workflow.

Avalara lists AI item classification for taxability, AI tariff code classification, a rules builder that works from prompts and Aviator, which it calls its agentic command center. Its Core Compliance plan costs $79 per state per month or $799 per state per year and is aimed at businesses under $50 million in annual revenue; larger companies get a custom plan.

Vertex names 4 agents: a Product Mapping Agent, a Configuration Agent that flags how tax content changes affect a customer's setup, ERP and workflow agents that detect transaction patterns, and a Text-to-Tax Agent that creates rules from natural language. Ask Sovi from Sovos diagnoses issues, explains what went wrong and applies fixes with the user's approval. Fonoa Knowledge sends cited alerts on indirect tax changes and names coverage of more than 95 countries.

For smaller sellers, Kintsugi offers a free plan with exposure monitoring, AI product categorization and its Kin assistant, and charges from $75 per filing or registration on its Starter plan. The difference to the enterprise engines is scope: the free plan watches thresholds, the paid plans register and file.

Sources: [13][2][8][14][15][16]

Transfer pricing and returns: narrower tools, closed data

Transfer pricing and the global minimum tax put the heaviest research load on corporate tax teams. Groups above the 750 million euro revenue threshold now compare their effective tax rate with the 15 percent floor, and questions on OECD rules and their national implementations are where a closed, curated source set helps most.

Transfer pricing also shows how far a vendor trusts its model. Aibidia runs its TP Aurora assistant on a closed, validated dataset of IBFD content and the OECD guidelines, and the assistant does not access the internet. It is sold per seat at a price Aibidia does not publish, with a 30-day free trial that includes 10 complete answers with sources.

Return preparation is mainly a category for accounting firms rather than corporate departments. Filed charges a flat monthly fee scoped to a firm's client book, with no charge per seat, per return or per credit, and offers a 60-day money-back guarantee during the pilot. TaxGPT's Pro Plus plan adds an agent called Cowork and tax preparation to its research tools.

Sources: [3][6][17][9]

Pricing: what AI tax software vendors publish

Price transparency differs sharply between tax and audit tools. In our tax category, 52 percent of tools list their price as on request, and 26 offer a free trial. Published prices cluster at the small end of the market: research tools for tax firms and sales tax tools for online sellers.

Among the AI tools for tax automation in our directory, the enterprise engines from Vertex, Sovos and Fonoa quote after a demo, and Avalara publishes its compliance plans but moves companies with $50 million or more in revenue to a custom plan. In our audit category, by comparison, 84 percent of tools quote on request.

Published prices are also hard to compare because the units differ: per user per year at Blue J, per state per month at Avalara, per filing at Kintsugi, per firm at Deubner. Convert every offer into an annual cost for your own team size and filing footprint before you build a shortlist.

Sources: [2][1][16][12]

What to check before you buy AI tax software

Start with the sources. Ask which content the model may cite, how often it is updated and whether every answer links to a document your team can open. Blue J updates its database daily, Fonoa cites every response and TP Aurora stays off the internet; each answer to that question tells you how much checking your team still has to do.

Then test with your own cases. Most vendors offer a trial or pilot: 7 days at Blue J, 30 days at Aibidia, 4 weeks at Haufe, 14 days at Deubner. Put questions to the tool that your team has already answered and compare the results, citations included.

Settle data handling before any client or company data goes in: where it is stored, whether prompts are used for training and who at the vendor can see them. Deubner, for example, states that customer questions are not used for AI training. The same questions come up with AI tools for auditors, where evidence and sign-off rules are stricter.

Sources: [1][15][3][11][12]