2026-08-26 · 4 min read

AI for consultants: all the leverage, zero client exposure

What can AI do for a business or financial consultant?

The analysis and writing layer: organizing findings into a deck, drafting an executive summary, summarizing meetings, scaffolding a model, and explaining concepts to a client. None of these capabilities depends on the client's name — they work just as well on COMPANY_001.

  • Executive summaries — turning a long analysis into one sharp page of findings.
  • Deck structure — the right narrative order for findings and recommendations.
  • Table work — spotting trends and phrasing conclusions from tokenized data.
  • Meeting summaries — raw bullets into a clean recap with action items.

Which materials at a consultancy are the riskiest to paste?

The materials the client would show no one: internal financial statements, models with real assumptions, strategic plans and payroll data. For financial advisors serving individuals — also portfolios and personal data of clients like John Miller.

  • Internal financials — true margins, cost structure, debt.
  • Financial models — projections and assumptions that are the heart of the trade secret.
  • Strategic plans — market entry, a merger, a restructuring — before they're public.
  • Employee and payroll data — which is also personal data under privacy law.

Why is 'just a small summary' still a problem?

Because a confidentiality clause doesn't distinguish a big paste from a small one — any transfer to a third party counts. And even a short excerpt gives things away: the client's name in a header, a competitor in a paragraph, a margin figure that appears only in the internal report.

Vendor privacy policies don't resolve this either: terms vary between tiers, and control of the server side is never yours. What is yours is the content at the moment of pasting — which is why the rule is to clean it first.

What does a safe workflow look like for a consultant?

  1. Replace the client's name with COMPANY_001, competitors with COMPANY_002 onward, and people with PERSON_001 — consistent tokens keep the document coherent.
  2. Replace telling numbers — margins, salaries, prices — with tokens or consistently shifted values.
  3. Clean metadata: the Author field, comments and tracked changes expose names even in a 'clean' file.
  4. Do the AI work — analysis, wording, structure — on the clean copy.
  5. Restore real names and numbers only in the final file, on your own machine.

The principle serves any organization adopting AI — see the safe AI adoption guide — but for a consultant it's doubly critical: confidentiality is the product. The same approach carries over to startups handling NDAs, and the general never-paste list makes a good appendix to any engagement policy.

Frequently asked questions

May a consultant use ChatGPT on client material?

Only after cleaning it: confidentiality clauses forbid disclosure to third parties, and an external AI tool is a third party. On anonymized material — identity and telling numbers replaced — the work is safe.

Are numbers without the company name safe enough?

Not always. Industry, scale and distinctive figures combined can give the client away to anyone who knows the market. It's safer to tokenize or shift telling numbers as well.

The client would never know — what's the real risk?

A contractual breach doesn't depend on immediate discovery: it can surface in an audit, a dispute, or a future leak. And beyond the legal side — a consultant's reputation is built on trust, which is very hard to win back.

Can AI analyze a client's spreadsheet?

Yes, after names and identifying numbers are replaced in the analysis copy. You apply the conclusions to the real file on your own machine — the tool doesn't need the real data to spot structure and trends.