# AI for consultants: checking the research before it reaches the client

Most lists of AI tools for consultants cover drafting help: proposals, meeting notes, slides. This page on AI for consultants is about the other half of the job, the research an independent strategy or M&A advisor hands a client under their own name. It goes through the three deliverables a solo or small-firm advisor produces most, a market brief, a competitor scan and a commercial diligence memo, and for each one says what a client will push back on, where the public evidence lives, and what to check before a number goes in the deck. A checklist for any AI-drafted research follows.

## The risk sits with the person who signs

The [Bureau of Labor Statistics](https://www.bls.gov/ooh/business-and-financial/management-analysts.htm) counts 1,077,100 management analyst jobs in 2025, projects 10% growth to 2035, and finds 14% of the work is done by the self-employed. A solo advisor has no research desk and no second reader.

AI drafting does not remove that gap. In a [BCG experiment with 758 of its consultants](https://www.bcg.com/publications/2023/how-people-create-and-destroy-value-with-gen-ai), those given GPT-4 did 23% worse than the control group on a business problem-solving task built so the model's first answer would be wrong, and BCG reports that many accepted the erroneous output at face value. The cost of that shows up in fees. [Fortune reported](https://fortune.com/2025/10/07/deloitte-ai-australia-government-report-hallucinations-technology-290000-refund/) in October 2025 that Deloitte's Australian member firm would refund the final instalment on a $290,000 government report after a university researcher found references to papers that do not exist and a fabricated quote from a court judgment; the revised report disclosed that a generative AI system was used to create it.

## The market brief

**What the client challenges.** The definition of the market, the year behind each figure, and whether the headline size is a government count or a vendor's model. Vendors rarely agree, for the reasons in [why market size reports disagree](/guides/why-market-size-reports-disagree).

**Where the evidence lives.** Most paid market figures rest on a few free Census programs:

- The [Economic Census](https://www.census.gov/programs-surveys/economic-census/about.html) runs every five years, for years ending in 2 or 7, and is the widest count of establishments, employees, payroll and revenue. The 2022 census is the latest.
- [County Business Patterns](https://www.census.gov/programs-surveys/cbp/about.html) is annual and covers establishments with paid employees by industry code, down to county and ZIP code. It excludes crop and animal production and most government employers.
- [Nonemployer Statistics](https://www.census.gov/programs-surveys/nonemployer-statistics/about.html) counts businesses with no payroll and receipts of $1,000 or more, which is where owner-operators in fragmented service markets show up.

**Before a number goes in the deck.** Check that the industry definition matches the client's market, that every figure carries its year, and that an establishment count from County Business Patterns is not presented as the total number of firms when most of the market is nonemployers. The [TAM SAM SOM guide](/guides/how-to-calculate-market-size) shows a bottom-up build from these files, and [how to check a market research report](/guides/how-to-check-a-market-research-report) traces one vendor's headline figure back to Census and SEC data. For the tool side of this job, see [AI market research](/ai-market-research).

## The competitor scan

**What the client challenges.** Who is missing, whether the revenue figures are comparable, and how you know anything about the private rivals.

**Where the evidence lives.** For public competitors, the 10-K business section is a required description. SEC rules ([Regulation S-K, Item 101(c)(1)](https://www.law.cornell.edu/cfr/text/17/229.101)) list "trends in market demand and competitive conditions", dependence on key customers or products, and seasonality among the topics it may cover. [EDGAR full-text search](https://www.sec.gov/edgar/search/efts-faq.html) reaches every electronic filing since 2001, exhibits included, so you can find every filer that names a rival or a product. The SEC's [XBRL APIs](https://www.sec.gov/search-filings/edgar-application-programming-interfaces) return reported financial facts without an API key, and [account research from a 10-K](/guides/account-research-from-a-10-k) works one filing through.

Private rivals leave a thinner trail. A company raising money under an exemption files a [Form D notice](https://www.sec.gov/resources-small-businesses/exempt-offerings/filing-form-d-notice) within 15 days of the first sale. An employer with [100 or more employees](https://www.law.cornell.edu/uscode/text/29/2101) must give [60 days' written notice](https://www.law.cornell.edu/uscode/text/29/2102) of a plant closing or mass layoff to the state and the local elected official, and those notices are a public signal of a competitor shrinking.

**Before a number goes in the deck.** Line up fiscal years, separate segment revenue from company revenue, and label any private-company figure as an estimate with its source. The [restaurant POS competitor analysis](/samples/restaurant-pos-competitor-analysis) is an unedited example of a cited scan, and [AI competitive analysis](/ai-competitive-analysis) covers this deliverable on its own.

## The commercial diligence memo

**What the client challenges.** How much of the target's or comparable's growth is organic, how concentrated its customers are, and whether the market is growing as fast as the company claims.

**Where the evidence lives.** Public comparables disclose growth in their filings, but "organic growth" is usually a company-defined, non-GAAP measure. [Regulation G](https://www.law.cornell.edu/cfr/text/17/244.100) requires a company that publishes one to present the most directly comparable GAAP measure and a quantitative reconciliation, so the reconciliation table is the place to read the split. For a small private target, lien filings ([what a UCC lien is](/guides/what-is-a-ucc-lien)) and the scoping in [commercial due diligence for a small acquisition](/guides/commercial-due-diligence-small-acquisition) cover what public records can and cannot show.

Public records do not cover private deal data. PitchBook publishes no list price; a 2026 [Department of Energy order](https://api.usaspending.gov/api/v2/awards/CONT_AWD_89703026FAR000017_8900_89703025AAR000001_8900/) paid $96,000 for 12 users, $8,000 each, for one year ([what PitchBook costs](/alternatives/pitchbook-pricing)). If the memo turns on private transaction comps, that database is the right tool. [AI due diligence](/ai-due-diligence) and [AI for private equity](/ai-for-private-equity) go further on deal work.

### A worked example: organic and acquired growth from a filing

This is one question an M&A advisor looking at pest-control roll-ups might ask, sent to [Answers](/answers), the Hyperresearch tool for a single cited fact. The answer and quotes below are as returned on 19 November 2026, trimmed but not edited.

**Question:** What revenue and organic growth did Rollins (the pest control company) report for its most recent quarter, and how much of its growth came from acquisitions?

**Answer:** Rollins reported Q3 2026 revenue of $1.1 billion ($1,089.7 million), up 6.2%, with organic growth of 4.4% and acquisitions contributing 1.8 percentage points of growth.

**Sources, each quote marked verified against the page:**

- [Rollins Form 10-Q for the quarter ended September 30, 2026](https://www.sec.gov/Archives/edgar/data/84839/000008483926000045/rol-20260930.htm): "Third quarter revenues were $1.1 billion, an increase of 6.2% over the third quarter of 2025 with organic revenues* increasing 4.4% compared to the prior year."
- Same filing, the revenue reconciliation table: "Revenues from acquisitions | (18,348) | … | 1.8 | …" (the acquisitions line, in thousands of dollars, and its 1.8-point share of growth)

The asterisk on "organic revenues" is the filing marking it as a non-GAAP measure and pointing to its reconciliation, which is the check described above. The [HVAC acquisition report](/samples/buying-an-hvac-business) shows the same kind of work at the length of a full memo.

## A pre-delivery checklist for AI-drafted research

Run this on any draft, whoever or whatever wrote it:

1. Open every cited link and find the sentence that supports the claim. A real page that does not support the claim is harder to spot than a dead link.
2. Check every quotation word for word against the source.
3. Confirm the year and the geography of each figure, and label mixed years.
4. Check units: thousands against millions, fiscal against calendar years, segment against company.
5. Mark company-defined measures (organic growth, adjusted EBITDA) and read their reconciliation.
6. Recompute every derived number, such as a share, a growth rate or a multiple.
7. Make sure two citations are not the same press release reprinted.
8. Label estimates, interviews and the client's own data as such, apart from published facts.

[Verifying citations in AI-generated text](/guides/verify-citations-in-ai-output) covers steps 1 and 2 in detail. To price the hours this takes, see [consulting rates](/guides/consulting-rates) and [GSA consulting rates](/guides/gsa-consulting-rates).

## Where a research tool fits

Hyperresearch does the gathering and the first pass of that checklist. You ask one question; a Light run keeps about 20 to 50 sources and a Deep run targets 150 to 200 over about 3 to 5 hours (targets, not guarantees). Both tiers read each cited claim against its source and correct what does not hold before the report ships; a Deep report adds a [verification receipt](/verification-receipt) with a verdict for each citation. On [DRACO](/benchmarks/draco), a benchmark where factual accuracy is about half the score, Hyperresearch Deep scored 77.7 across 10 tasks.

It reads public pages only, logged out. It does not read PitchBook, Crunchbase or a client's data room, interview anyone, build your model or write the deck, and a checked citation means the source says what the sentence claims, not that the source is right. Steps 3 to 8 of the checklist stay with you. [How it works](/how-it-works) has every stage.

## Starting

Runs are priced per run with no seat licence; see [pricing](/pricing). Your first Light report is free when you sign up at [console.hyperresearch.ai](https://console.hyperresearch.ai/login), with no card. A useful first test is a question you have already answered for a client, so you can compare its sources with yours.
