Deloitte admits AI wrote parts of Wellington council report — but stands by findings

Wellington mayor Andrew Little has publicly stated that “large chunks” of a $435,000 Deloitte report on council staffing were written by artificial intelligence. The admission adds to mounting criticism of the Future Fit Pōneke report, which recommended cutting 330 staff — a figure now revealed to be based on flawed data.

Key errors found in the report

The council's own review uncovered multiple data-handling mistakes:

• Deloitte double-counted current employees alongside vacant roles.
• Employment data was three years old, sourced from the Taxpayers' Union.
• All employees were counted at their full-time salary, regardless of actual hours worked.
• Staffing expenses were overstated by $21.5 million.
• Roles with “project” in the title were automatically assumed to be duplicates.

Because of these inaccuracies, the council concluded that the recommendation to cut 330 FTE roles was “inherently flawed.” A more accurate figure might be 284, though even that does not account for other potential errors.

AI disclosure remains vague

Mayor Little told Newstalk ZB that the AI-generated sections were “just a waste of time” compared to the chapter on contract management systems, which he described as a “disaster” but useful. Deloitte has not specified which sections were AI-produced or how the AI was used.

In a prepared statement, Deloitte said: “Deloitte uses a range of tools and technologies across its work, including AI-enabled tools where appropriate,” while requiring that “a human needed to sign off any analysis, findings and conclusions.” They maintain the report's conclusions are sound.

Implications for AI use in consulting

This case highlights the risk of using AI without rigorous verification — especially in high-stakes public-sector reports. For developers and teams building AI-assisted workflows, the lesson is clear: AI can draft, but humans must validate every data point and assumption. The Wellington council's experience is a cautionary tale about the reputational cost when AI-generated content is accepted without thorough audit.