Process taxes: what to automate, what stays human, and what the data needs

Software can already take on much of tax return preparation, deferred tax arithmetic, tax journal entries and the compliance calendar. Planning, uncertain positions and conversations with authorities still need a person. None of it works until ledger accounts, legal entities and asset records carry clean, consistent tax attributes.

Where software earns its place

Return preparation is the obvious candidate. Once trial balances map reliably to return lines, tax engines can populate forms, roll forward prior year schedules and run validation checks before anyone signs. The preparer's job shifts from typing figures to reviewing exceptions.

Deferred tax calculation suits automation for a similar reason. Temporary differences between book and tax basis follow rules. A provision tool can compute them, apply enacted rates by jurisdiction and produce the rate reconciliation. It does this faster and more consistently than a spreadsheet passed between three people at quarter end.

Accounting for taxes follows naturally. Current and deferred entries can post from the provision tool to the ledger without rekeying. Reconciling tax payable accounts against payments and returns is also mechanical work that matching software handles well.

Compliance monitoring is mostly a calendar problem. Filing obligations by entity and jurisdiction, payment deadlines and extension status can live in a system that sends alerts and tracks evidence of filing. This replaces the inherited tracker that only one person truly understands.

Where AI helps but should not decide

Foreign taxes sit in the middle. Software handles withholding calculations and foreign tax credit arithmetic. Transfer pricing documentation, permanent establishment exposure and treaty positions depend on facts about how the business operates, and those facts change with every new contract or hire abroad.

Tax inquiries are similar. AI can read a notice, classify it, pull the relevant return and payment history, and draft a response. A person must check that draft before it goes anywhere. Letters to an authority become part of the record, and a confident but wrong explanation is hard to retract.

Research is another fair use. Summarising a new rule or comparing treatment across jurisdictions saves real time. The output is a starting point for a qualified reviewer, never a position in itself.

What stays with people

Tax planning and strategy remain human work. Choosing an entity structure, timing a restructuring or deciding how aggressive a position to take involves judgement about risk appetite, reputation and business plans that no model holds.

Within the provision, valuation allowances and uncertain tax positions need someone accountable. Whether future profits will absorb a deferred tax asset is a forecast and a judgement call, and auditors will ask who made it and why.

Sign off on returns stays with a named individual too. Automation can prepare almost everything; responsibility cannot be delegated to it.

What has to be true about the data first

Every ledger account needs a tax mapping that someone owns. When accounts get added without one, figures land in suspense or in the wrong return line, and the tool quietly produces errors.

Legal entity data must agree everywhere it appears. The ledger, the tax engine and the compliance calendar should hold the same entities, jurisdictions and registration numbers.

Fixed asset registers must track tax basis alongside book basis, with depreciation methods recorded per jurisdiction. Deferred tax figures are only as good as these records.

Transaction level tax codes on sales and purchases need to be correct at the point of entry. Fixing them later, in the tax team, is where most indirect tax effort disappears.

Intercompany balances should reconcile before tax work starts. Unmatched intercompany charges undermine transfer pricing support and foreign tax calculations alike.

Questions to ask the people who run it

What documentation describes and what actually happens usually drift apart in tax. These questions surface the gap:

• Which numbers get adjusted by hand after the system produces the provision, and why?
• Where do figures for foreign subsidiaries really come from, and who checks them?
• Which spreadsheets would cause trouble if their owner left tomorrow?
• How does a new filing obligation get onto the calendar when the business enters a new country?
• When a notice arrives, who sees it first, and how does it reach the right person?
• Which ledger accounts get remapped every period because the standard mapping is wrong?
• What do the external advisers prepare that the internal team never reviews in detail?
• Which judgements in the provision were last revisited long ago?

Answers to the first and last questions often reveal more than the rest combined. Manual overrides point to broken data upstream. Stale judgements point to risk that automation would simply carry forward faster.

Sequencing the change

Fix data before buying tools. A provision engine fed with unmapped accounts produces wrong answers with great efficiency. Start with the account mapping and the entity master, then automate the calculations that depend on them. Leave planning, uncertain positions and correspondence with authorities in human hands, supported by better information drawn from the cleaner records.

APQC's Process Classification Framework® (PCF) is an open standard developed by APQC, a nonprofit that promotes benchmarking and best practices worldwide. To download the full PCF or to view definitions and measures, please visit www.apqc.org/pcf.