KPMG retracted an agentic AI study in early 2025 after independent researchers flagged fabricated citations and hallucinated claims in the published report. The firm issued a public apology and pulled the paper from circulation. For consulting firms racing to adopt AI tools, this incident isn’t a cautionary tale about avoiding AI. It’s a blueprint for what happens when you skip the human review layer.

The study claimed to analyze the capabilities of agentic AI systems. Researchers outside KPMG noticed citations that didn’t exist, statistics with no source, and conclusions that contradicted the underlying data. The firm’s response was swift, but the damage to credibility was real. When your business model depends on trust and expertise, publishing work that can’t withstand scrutiny is an existential risk.

Most consulting firms aren’t publishing academic papers, but they’re producing AI-assisted client deliverables every week. Proposals, market research briefs, industry reports, pitch decks. The same risk applies. An AI tool generates a paragraph with a plausible-sounding statistic. A senior consultant skims it, assumes the source is valid, and the deck goes to the client. Three weeks later, the client’s team fact-checks it and finds nothing. You’re now explaining why your firm didn’t catch a basic error.

The solution isn’t to stop using AI. It’s to build a review protocol that treats AI output the same way you’d treat work from a junior analyst. You check the sources, verify the claims, and rewrite anything that doesn’t meet your standard. The firms that figure this out will compress their cost-of-sale by 30 to 50 percent while maintaining quality. The firms that don’t will either avoid AI entirely or publish work that erodes trust one deliverable at a time.

The Real Cost of Publishing Without Review

A partner at a mid-sized strategy firm told me his team spent 40 hours on a major proposal last quarter. They used an AI tool to draft sections of the methodology and competitive analysis. The tool pulled in case studies, industry trends, and a few statistics that looked credible. No one on the team verified the sources because the deadline was tight and the output read well.

The client didn’t award the work. During the finalist presentation, their CFO asked about one of the statistics in the competitive section. The partner couldn’t provide a source. The CFO pushed. The statistic turned out to be a hallucination, a plausible number the AI generated to fill a gap in its training data. The firm didn’t lose the deal solely because of that moment, but it shifted the tone of the conversation. Trust evaporated.

That’s a $200,000 engagement that didn’t close, and the reputational cost compounds. The client talks to other buyers. The story spreads. Your firm becomes known as the one that doesn’t check its work. The cost isn’t just the lost deal. It’s the three deals you never hear about because someone heard the story secondhand.

Consulting firms typically lose $80,000 to $300,000 per year to inefficiencies in proposal development, research duplication, and knowledge management. When you add the risk of publishing AI-generated errors, the number climbs. A single retracted deliverable or public correction can cost more than a year of operational waste. The KPMG incident is a reminder that the downside of getting this wrong is asymmetric.

What a Human Review Protocol Actually Looks Like

The firms that use AI successfully don’t treat it as a replacement for expertise. They treat it as a research assistant that needs supervision. The protocol looks like this: the AI generates a draft, a human reviews every claim, verifies every source, and rewrites anything that doesn’t meet the firm’s standard. The output is faster than writing from scratch, but it’s never published without review.

A Research Agent can pull industry reports, company filings, and competitor analysis in 20 minutes. It produces a structured brief with sources, summaries, and key findings. The senior consultant reads the brief, checks the citations, and decides which insights are relevant to the client’s context. The AI compresses the research phase from two weeks to two days, but the human still makes the judgment calls.

A Proposal Generation Agent can draft a tailored proposal by pulling past case studies, pricing models, and methodology sections from your firm’s archive. It generates a 15-page deck in an hour. The partner reviews every section, rewrites the executive summary, adjusts the pricing based on the client’s budget, and verifies that the case studies are accurate. The AI saves 30 hours of drafting time, but the partner still owns the final product.

The protocol isn’t complicated. It’s the same process you’d use with a junior analyst. You don’t publish their work without review. You don’t assume their sources are valid. You check, you edit, you approve. The difference is that an AI agent can produce a draft in minutes instead of days, so the time saved is real. But the review step is non-negotiable.

If you’re looking for a structured way to deploy your first agent with a built-in review protocol, we’ve put together a practical worksheet that walks through the setup. You can grab it here: Deploy Your First Business Agent. It includes a checklist for verifying AI output before it reaches a client.

The Three Places Consulting Firms Leak Revenue Without Realizing It

Most consulting firms don’t have a revenue problem. They have a cost-of-sale problem. The work is profitable once it’s won, but the process of winning it is expensive. Three areas drive the majority of waste: proposal development, research duplication, and knowledge management debt.

Proposal development consumes 20 to 40 hours of senior time per major opportunity. A partner or director writes the deck from scratch, pulls case studies from memory, and formats slides manually. The win rate might be 30 to 40 percent, which means 60 percent of that time produces no revenue. A Proposal Generation Agent can cut that time in half by drafting the structure, pulling relevant case studies, and formatting the deck. The partner still reviews and tailors the content, but the manual work is compressed.

Research duplication happens at the start of every engagement. A consultant spends two weeks reading industry reports, analyzing competitors, and synthesizing trends. Three months later, another consultant does the same research for a different client in the same sector. The firm pays for the same insight twice. A Research Agent can run that analysis in 20 minutes, with sources and summaries, so the second engagement starts with the work already done. The consultant reviews it, adds client-specific context, and moves to the strategic work.

Knowledge management debt is the silent killer. Every project produces IP. Decks, memos, frameworks, client insights. Almost none of it is reusable because it’s locked in email threads, Dropbox folders, and individual laptops. A Knowledge Agent can index every document the firm produces and answer questions across the corpus. A partner preparing for a pitch can ask, “What frameworks have we used for digital transformation in manufacturing?” and get a list of relevant decks with summaries. The firm stops paying for the same insight twice.

These three areas typically account for $80,000 to $300,000 in annual leakage for firms doing $1 million to $25 million in revenue. The number scales with firm size. A 20-person consultancy might lose $150,000. A 50-person firm might lose $400,000. The waste isn’t always visible because it’s distributed across dozens of proposals, engagements, and missed opportunities. But it compounds every quarter.

How Omni Builds Review Protocols Into Every Agent

We’ve built agents for consulting firms that compress proposal time by 50 percent, research time by 70 percent, and make the firm’s entire knowledge base searchable in natural language. The difference between those agents and the AI tools that produce hallucinations is the review layer. Every agent we build includes a human checkpoint before output reaches a client.

A Proposal Generation Agent doesn’t send a deck directly to a prospect. It generates a draft, flags sections that need partner review, and highlights any claims that lack a source. The partner reviews the draft, adjusts the pricing, rewrites the executive summary, and approves the final version. The agent saves 30 hours of manual work, but the partner still owns the quality.

A Research Agent doesn’t produce a final report. It produces a structured brief with sources, summaries, and key findings. The consultant reviews the brief, verifies the citations, and decides which insights are relevant to the client’s context. The agent compresses the research phase from two weeks to two days, but the consultant still makes the judgment calls.

A Knowledge Agent doesn’t make decisions. It surfaces relevant documents, summarizes past work, and answers questions across the firm’s archive. The partner reads the summaries, decides which frameworks apply to the current client, and tailors the approach. The agent eliminates the manual search, but the partner still applies the expertise.

The protocol is simple: AI generates, human reviews, human approves. The firms that adopt this model get the speed of AI without the risk of publishing hallucinations. The firms that skip the review step will eventually publish something they have to retract. The KPMG incident is a preview of what that looks like at scale.

If you want to see what this looks like for your firm, book a 60-min Omni Audit. We’ll map your current workflow, identify where AI can compress time, and show you the review protocol we’d build into each agent. No deck, no pitch. Three outputs: a workflow map, a priority list, and a build plan. You’ll leave with a clear picture of what’s possible and what it takes to implement it safely.

What the Omni Audit Looks Like for Consulting Firms

The audit is 60 minutes. We walk through your proposal process, research workflow, and knowledge management setup. We ask about the manual work that consumes senior time, the research that gets repeated across clients, and the IP that’s locked in individual files. We map the workflow, identify the highest-impact agents, and show you what the review protocol looks like for each one.

The first output is a workflow map. We document the current process for proposals, research, and knowledge management. We highlight the steps that consume the most time and the points where AI can compress the work. You’ll see exactly where the 20 to 40 hours per proposal are going and which steps an agent can handle.

The second output is a priority list. We rank the agents by impact and implementation complexity. A Proposal Generation Agent might save 30 hours per major opportunity and take four weeks to build. A Research Agent might save 10 hours per engagement and take two weeks. A Knowledge Agent might save 5 hours per week across the firm and take six weeks. You’ll know which agent to build first and what the return looks like.

The third output is a build plan. We outline the data sources, the review checkpoints, and the integration points for each agent. You’ll see the timeline, the cost, and the internal resources required. No surprises, no scope creep. You’ll leave the audit with a clear decision: build it, defer it, or pass.

Most consulting firms that go through the audit decide to build at least one agent within 30 days. The ones that don’t usually defer because of timing, not because the case isn’t clear. The cost-of-sale is brutal, the research is repetitive, and the knowledge management debt compounds every quarter. The firms that solve this first will compress their cost-of-sale by 30 to 50 percent while maintaining quality. The firms that wait will keep paying for the same work twice.

You can see the full Omni Audit for consulting firms here. It’s designed specifically for firms doing $1 million to $25 million in revenue, with a focus on proposal generation, research synthesis, and knowledge management. The audit is free, the outputs are yours to keep, and there’s no obligation to build anything.

The Firms That Win Will Build Review Into Every Workflow

The KPMG retraction isn’t a story about AI failure. It’s a story about process failure. The firm used AI to generate content, skipped the review step, and published work that couldn’t withstand scrutiny. The tool didn’t fail. The protocol did.

Consulting firms that adopt AI successfully will treat it the same way they treat junior analysts. The AI generates a draft, the human reviews every claim, verifies every source, and approves the final output. The speed is real, the cost savings are real, but the review step is non-negotiable. The firms that skip it will eventually publish something they have to retract. The firms that build it into every workflow will compress their cost-of-sale by 30 to 50 percent while maintaining the trust that keeps clients coming back.

If you’re ready to see what that looks like for your firm, book your Omni Audit here. Sixty minutes, three outputs, no deck. You’ll leave with a clear picture of what’s possible and what it takes to implement it safely. The firms that figure this out first will win the next five years. The firms that wait will keep paying for the same work twice.

For more on how AI agents are reshaping professional services, check out our insights on AI adoption and explore the Omni platform we’ve built specifically for knowledge-intensive businesses. The tools exist. The question is whether you’ll build the review protocol that makes them safe to use.