Your First ASC 740 Disclosure Under the New Rules Lands This Year. Can AI Do the Grunt Work?

If you're at a private company with a calendar year end, your first income tax disclosure under the new ASC 740 rules is the one you'll put out for 2026, which means the close you're walking into in about three months. FASB's ASU 2023-09 hit public business entities for annual periods beginning after December 15, 2024, and everybody else for annual periods beginning after December 15, 2025, so the grace period private companies have been enjoying while they watched the public filers go first has run out.

What the new requirement asks for is more DATA, sliced by jurisdiction, and that changes who should be doing the work. It's precisely the kind of assembly that has always fallen to a senior tax person hand-copying numbers between spreadsheets at eleven at night during provision week, which is a hell of a way to spend your life. That's also the exact category of work AI genuinely handles now, and it's why I think this particular standard is the best excuse you're going to get to finally build something.

What actually changed in ASC 740?

Two things, and the split between public and private matters. FASB's Improvements to Income Tax Disclosures project requires public business entities to present an annual rate reconciliation in a table, in both reporting-currency amounts and percentages, broken into eight specified categories. Private companies get a lighter version: qualitative disclosure about the nature and effect of those same categories of reconciling items, plus the individual jurisdictions driving a significant gap between your statutory rate and your effective rate.

The second requirement lands on everyone, and it's the one that will eat your time. Every entity now has to disclose income taxes paid, net of refunds received, disaggregated between federal, state and local, and foreign, and then break out any individual jurisdiction where that amount hits 5% or more of the total. The threshold gets tested on absolute values, so a big net refund in one state triggers the disclosure the same way a big net payment does. Income from continuing operations gets split domestic versus foreign, and income tax expense gets split federal, state, and foreign.

Read that again if you have entities in more than two states, because it means you now need jurisdiction-level cash tax data, net of refunds, accurate enough to test against a 5% threshold. Plenty of private companies have never tracked that at the precision a disclosure demands. The numbers exist, scattered across payment confirmations, estimated payment schedules, ERP subledgers, and somebody's damn email folder. Assembling them is the work.

Why is the provision the worst place to still be doing manual data collection?

Because it's the highest-stakes, most time-compressed, least-forgiving workstream in the close, and it's still largely held together by hand. Thomson Reuters' 2025 State of the Corporate Tax Department research found 58% of tax departments globally report being under-resourced, and 59% doubt they can upgrade their tax technology in the next two years. So the requirements are expanding, the teams aren't, and the tooling budget mostly isn't coming either.

That's the honest constraint, and the answer here is smaller than a platform purchase. Most of the pain in a provision comes from the assembly: chasing trial balances in four formats, reconciling a payment log against a GL account, rebuilding the same state apportionment workpaper you rebuilt last quarter. Same shape as the month-end close, and the same shit hurts for the same reason.

What is AI actually good at in a tax provision?

The extraction and the reconciliation. Point a model at a folder of payment confirmations and estimated tax vouchers and it will pull payer entity, jurisdiction, period, and amount into a clean table faster and far more consistently than a human doing it at midnight. Give it a GL detail export alongside last year's disclosure and it will flag which jurisdictions moved, which new ones appeared, and where cash taxes paid diverges from tax expense. Hand it a draft footnote and last year's version and it will tell you every place the language needs to change, which is useful in a year where the language has to change.

The judgment stays with you, and it's a longer list than the vendors admit: whether an item belongs in the rate reconciliation, what's material, how to conclude on an uncertain position, and who signs. The 5% jurisdiction test looks mechanical, and it turns on how you've defined the denominator and how refunds get netted, and those calls are yours. Vendors pitching a model that runs the whole provision are selling you bullshit, and the skeptics who won't let it near the data assembly are leaving real hours on the table. The useful ground sits in the middle, where it has sat for about two years now.

The build: a jurisdiction-level cash tax tracker

The concrete thing worth building before December is small enough to finish in a week of evenings.

Start with one file: every income tax payment and refund your entities made this year, one row each, with entity, jurisdiction, type (federal, state and local, foreign), period, and amount. Build it by pointing an AI tool at your payment confirmations and voucher PDFs and having it extract to a table, then tie the total back to your income taxes payable account activity. That tie-out is the whole control, and if it reconciles to the GL, the underlying extraction is trustworthy enough to disclose from.

Then add a column that computes each jurisdiction's absolute net payment as a percentage of the absolute total, so the 5% test answers itself and you can see now, in September, which jurisdictions you'll be naming in the footnote. If a state is sitting at 4.6% with a quarter still to go, you want to know that before year end rather than during the audit.

One table, one tie-out, one percentage column, and your first ASU 2023-09 disclosure turns from a scramble into a lookup. If you want to see which tools are actually built to be driven this way, our software directory scores them on features, value, support, and security with nothing pay-to-play.

What I'd do

Do the cash tax table this month, while the year is still open and fixable. Then treat the footnote itself as a second, separate build in January, because the disclosure language comes far easier once the data sits in a clean table.

And do it yourself, on your own time, before any project gets funded. The teams pulling ahead in this profession are the ones who opened the tool on a Tuesday and tried, which is the same pattern we found looking at rolling reconciliations and the month-end close - the best practice was always obvious and the execution was always the problem, and the execution is what got cheap. Build it small and build it yourself, which is also the lesson buried in what Klarna's Salesforce reversal actually taught everyone. Two-thirds of tax departments still aren't using generative AI at all, per that same Thomson Reuters research, which leaves a lot of room to be the person in your group who already figured it out. If you want a seat where that work gets valued, the remote tax roles on our board skew heavily toward teams that are actually building.

Frequently asked questions

When does ASU 2023-09 take effect for private companies?

For annual periods beginning after December 15, 2025. A private company with a calendar year end applies it for the first time in its fiscal 2026 financial statements. Public business entities were already required to apply it for annual periods beginning after December 15, 2024.

What does ASU 2023-09 require for income taxes paid?

All entities must disclose income taxes paid, net of refunds received, disaggregated between federal, state and local, and foreign. Any individual jurisdiction representing 5% or more of the total, tested on absolute values, must also be broken out separately.

Do private companies have to give the eight-category rate reconciliation?

No. The tabular rate reconciliation in both amounts and percentages across eight specified categories applies to public business entities. Private companies instead disclose qualitatively the nature and effect of those categories of reconciling items and the individual jurisdictions causing a significant difference between the statutory and effective tax rates.

Can AI prepare an income tax provision?

AI handles the data assembly well, including extracting payment and voucher details, reconciling cash taxes to the general ledger, and comparing this year's footnote against last year's. The technical positions, materiality judgments, and conclusions on uncertain tax positions still require a tax professional, and a person signs the disclosure.

What is the 5% jurisdiction threshold in ASC 740?

It's the disaggregation trigger for income taxes paid. Where the absolute value of taxes paid to an individual jurisdiction, net of refunds received, equals or exceeds 5% of the absolute value of total income taxes paid net of refunds, that jurisdiction must be disclosed separately.

Effective dates and disclosure requirements are drawn from FASB Accounting Standards Update 2023-09, Income Taxes (Topic 740): Improvements to Income Tax Disclosures. This is general information and not tax or accounting advice, so check your own facts with your own advisers. Compare accounting and tax tools in our software directory.