Danai Data Book a discovery call

Blog In the news

After the Big Four's AI citation scandals: how consulting firms can use AI agents for desk research without a fake source reaching the client deck

Herman Wa · · 3 min read

Illustration

The news. In October 2025 Deloitte agreed to refund part of an A$440,000 Australian government contract after a report it produced with generative AI was found to contain fake citations and a made-up quote from a court judgment. Through 2026 more reports from large firms have been flagged for suspected AI-generated references, with at least one withdrawn and others corrected.

The lesson is not that consultants should avoid AI. It is that a citation nobody checked should never reach a client. That is the failure our standing orders are written against: every finding carries a source, sources are checked live before the minutes are signed, and anything uncited is marked as opinion.

Management consulting is, at its core, structured research and argument under deadline. A team gathers evidence, frames options, challenges them from several angles and presents a recommendation a client can act on.

That is also a fair description of what a governed AI agent team does. The fit is close, and so are the risks: a confident but unsourced claim in a client deck can cost a firm its reputation.

Where agent teams help

The heaviest time sinks in consulting work tend to be:

  • Desk research: market scans, competitor reviews and literature surveys.
  • First-pass analysis: framing options and their trade-offs.
  • Internal challenge: the red-team review before anything reaches the client.
  • Write-up: turning the work into a document with every claim supported.

A governed agent team can take the first pass at all four, with each finding sourced and each objection recorded, so consultants spend their time on judgement and client relationships rather than assembling the base.

What a consulting agent team might look like

An illustrative set-up for a strategy engagement:

  • a market research agent working from the sources the firm licenses and approves,
  • a financial analysis agent building and checking the numbers,
  • an operations agent testing whether each option can be delivered,
  • a risk and regulation agent looking for legal, regulatory and reputational problems,
  • a critic assigned to argue against the leading option,
  • and a chair that structures the session and writes the minutes.

The assigned critic matters. On our own research programme, critics assigned to each position paper produced the challenges that most improved the final plan. Built-in challenge is something a single assistant cannot give you.

Sources a partner can sign

The rule that matters most for consulting is simple: every finding carries a source. Uncited claims are marked as opinion in the minutes. A partner reviewing the work can see where each point came from and how confident the team was, and can trace a disputed figure back to its document.

Client confidentiality

Consulting firms hold several clients' confidential material at once, sometimes from competitors in the same sector. Two controls matter most:

  • A separate deployment, memory and log for every client. Nothing learned on one engagement can surface in another.
  • Controlled web access. Search queries can reveal what a client is working on, so agents on confidential work get no web access, or generalised queries only, against an allowlist.

What it does not replace

An agent team does not replace the partner's judgement, the client relationship or the room where a recommendation is tested. It replaces the hours spent building the base before that conversation can happen, and it leaves a record of how the base was built.

Starting point

A good first use case is a type of analysis your firm repeats across engagements, such as market-entry scans or due diligence first passes. Our two-week Discovery Sprint designs the team around it and ends with a working session on a real question.

Sources

Read next