The CFO of a PE-backed company should put finance's first AI on the 13-week cash forecast: score the current forecast against actual cash every week, run an AI forecast of customer receipts and supplier payments beside it, keep the AI only if its error is lower on the same weeks, and send the sponsor the error table every Monday.
Protiviti, a consulting firm whose services include finance transformation and AI, published its 2026 Global Finance Trends Survey on August 25, 2026, from 902 finance executives across North America, Europe and Asia-Pacific: 77 percent of finance organizations are using AI, but only 14 percent of this group have a defined AI strategy. The firm reports AI adoption still outpacing measurement, with finance leaders reporting greater effectiveness at measuring the ROI of broader transformation initiatives than of AI-specific investments.
The Association for Financial Professionals, a professional society, ran its 2026 Treasury Benchmarking Survey, sponsored by PNC Bank, in May 2026 and released it on September 15, 2026, from 425 responses by treasury practitioners at organizations of varying sizes, industries and ownership structures. Cash and liquidity forecasting was the most challenging treasury activity, cited by 49 percent, and the top priority overall, at 63 percent.
Nine-67 starts finance AI on the 13-week cash forecast because it is checked against the bank every week, so the company has a baseline before any tool arrives and a verdict by week 31, and because it feeds liquidity decisions.
Before week 1, the CFO writes down the test and approves the tool's license or build as a line in the finance budget. In weeks 1 to 13, the treasury manager, or the controller where the company has no treasury staff, saves each Monday's forecast and records actual cash the following Monday against the forecast from four weeks earlier for five lines: customer receipts, payroll, supplier payments, debt service and taxes. From week 6, each line's four-week error is scored in dollars and, where actual is nonzero, as a share of actual. Over the same 13 weeks, a finance systems analyst reporting to the controller, or the controller and the tool's vendor where there is none, connects the tool to the receivables ledger, the payables ledger and the bank feed, and the controller approves which accounts it reads.
From week 14, the analyst, or the vendor where there is none, runs the AI each Monday to forecast customer receipts from open invoices and each customer's payment history, and supplier payments from the payables aging and the payment run calendar, sending its figures to the CFO alone, unseen by the manual forecaster. Both are scored on the 13 weeks ending in week 30, each against what it said four weeks earlier. In week 31, the CFO keeps the AI only if its average weekly sum of the two lines' dollar errors, each made unsigned first, is below the manual forecast's on the same basis and weeks. The CFO, or the analyst where there is one, scores both weekly.
From week 1, the CFO sends the operating partner the forecast each Monday with the error table for the weeks scored so far, so the week 31 decision rests on numbers the sponsor has seen weekly. An operating partner can ask each portfolio company for the same table and compare each company only with its own manual baseline, since some companies' receipts are more predictable than others'.
In May 2026 the same 425 practitioners ranked AI and automation among the top five treasury priorities, at 30 percent, and managing AI opportunities and risks among treasury's most significant challenges, at 38 percent.
