Before funding an AI layer over finance, measure what checking, re-keying and reconciling data between systems costs finance in hours per check: that measure is the first saving the AI project has to beat, and it names the data to fix first.
KPMG International, which sells audit, tax and advisory services, published its 2026 Global AI in Finance survey on May 11, 2026, from 1,013 C-suite and senior finance leaders, 58 percent in technology or financial services, at organizations with at least 250 million US dollars in annual revenue (500 million in the United States), surveyed online in March 2026: 36 percent cite data quality as both their top barrier and opportunity. The same release reports active AI use in finance rising from 30 to 75 percent since 2024, with 71 percent of leaders saying it meets or exceeds return on investment expectations.
In a separate survey, EY, an audit and consulting firm, published its 2026 DNA of the CFO survey on June 3, 2026, from 1,610 CFOs, finance directors and heads of finance from 28 countries, at organizations with a billion US dollars or more in annual revenue, total assets, assets under management or gross written premiums (a quarter of that AUM floor for private equity funds), surveyed between February 16 and March 30, 2026: asked about AI investment barriers, 61% of CFOs surveyed cited data quality and bias as their top barrier.
An AI layer, meaning software that reads the finance systems and drafts reports or entries on top of them, is sold on finance time saved; none of the three releases gives an hours figure, so the count is the controller's own. In the week after one month-end close, the controller samples a week of checks between two systems: item, customer and vendor records matched between the ERP and the contract or matrix price file; vendor rebates and chargebacks re-keyed from supplier portals into spreadsheets; inventory value tied between the warehouse system and the general ledger. The controller owns the log, the CFO ranks it by hours per check, and the ERP administrator owns fixing the top pair; the cost is finance's time, from the finance budget.
At a distributor with several branches, one line might read: check, vendor rebate accruals; systems, the supplier's portal and the general ledger; who, the assistant controller; hours per check, the sample's average; when skipped, the rebate is missed at quarter end and branch margin is wrong until it is found.
The controller samples one close in month one; in month two the ERP administrator starts the top pair's fix, making one system the master; in month three the controller resamples the next close and the CFO reads any AI proposal against the recount. An operating partner can ask every portfolio CFO for the same log; its columns, shown in the line above, do not depend on the systems a company runs, and no source sets a threshold for funding a fix, so each portfolio sets its own; none of the sources here says how a fixed pair shows up in diligence or a quality of earnings, so that stays outside this count.
AFP published a benchmarking report on January 20, 2026, from 332 corporate finance practitioners of varying sizes worldwide, surveyed in August and September 2025: despite widespread adoption of planning technologies, the average time to produce a budget across all organizations surveyed remains unchanged from three years ago at 8.7 weeks.
