Aipedia: Management consultants

AI-built financial models and spreadsheet errors

Aipedia, by the Apex broking team · Last reviewed 2026-10-27

AI assistants can write spreadsheet formulas and build whole financial models, but they can also introduce errors that look correct and go unnoticed. A consultant remains responsible for every number in a model a client relies on.

Part of: AI in Management consultants

In short

Consultants use AI assistants inside spreadsheets to write formulas, build cash flow and business case models, and check existing models for errors. Mistakes such as wrong ranges, hard-coded values, sign errors and broken links can survive because the output looks tidy. The ICMCI AI Code requires meaningful human review (3.1), accountability and documentation (3.6) and no unreviewed output (5.2). Claims about calculation errors are often easier for a client to prove than claims about judgement, because the mistake is objective. PI usually responds to both, subject to the wording, but the defence looks very different.

How consultants use AI in models

Financial models underpin business cases, investment appraisals, pricing reviews and restructuring plans. AI assistants built into spreadsheet software can now write complex formulas from a plain-English request, generate a three-statement model from a description, create scenarios, explain what an inherited model does and flag apparent errors. Consultants also ask generative AI chatbots to produce formula snippets or macros that are pasted into client workbooks.

This lowers the skill threshold for building models, which is both the benefit and the risk. A junior consultant can now produce a model that looks sophisticated but that nobody fully understands, including its author.

Typical AI-introduced errors

Formulas that reference the wrong range or stop short of the last row.
Hard-coded numbers replacing links, so scenarios no longer flow through.
Sign and units errors, such as treating thousands as units or costs as income.
Circular references resolved silently, or incorrect handling of timing, such as annual rates applied monthly.
Error-checking tools that report a model as clean when they only checked syntax, not logic.

The duties that apply

ICMCI 3.1: meaningful human review and the ability to explain AI-assisted results. A model you cannot explain fails this.
ICMCI 3.6: accountability and documentation; in a model this means version control, an assumptions sheet and a record of checks.
ICMCI 5.2 and 5.3: no unreviewed AI output and no over-reliance on AI in place of competence.
Common law: reasonable skill and care. A reasonable modeller tests a model before a client uses it.

The ICMCI AI Code applies to individual CMCs, and the Institute of Consulting code requires competence from its members.

What can go wrong

Illustrative: a consultant uses an AI assistant to build a business case model for a client's new service line. The assistant writes a revenue formula that applies an annual growth rate in each month. The model shows a strong return, and the client commits to leases and hiring. When actual results come in well below plan, the client's finance team finds the error and claims the consultant's model was negligent.

Unlike a disputed strategy recommendation, there is little room to argue that this was a reasonable professional view. The argument shifts to causation and loss: would the client have gone ahead with the correct figures, and what did the error actually cost?

How PI treats calculation and judgement claims

PI for consultants usually covers both a calculation error and a judgement call, subject to the policy terms. The difference is in the defence. Judgement claims can often be defended on the basis that the advice fell within a range of reasonable opinion; calculation errors are harder to defend on liability, so the focus moves to causation, contributory negligence by the client and quantum. That is why model claims sometimes settle earlier.

Check the limit of indemnity against the scale of decisions made on your models, and the aggregation clause if one model is reused across several clients. If a client claims a refund of fees for the model, that may be treated separately from damages. See does PI cover AI mistakes.

Practical controls

Have someone other than the builder test every model before it is issued.
Rebuild key outputs with a simple independent calculation.
Keep inputs, calculations and outputs separate, with an assumptions sheet.
Record which parts were AI-generated and who checked them (ICMCI 3.6).
Lock and version-control issued models.
State the model's purpose and limits in the covering report.

How Apex helps

Apex arranges PI for management consultants and can check how your policy treats modelling work, aggregation and fee claims. See the AI in management consulting overview.

Sources

Frequently asked

Am I liable for an error in a formula an AI wrote?

Yes, in the same way as if you had written it. ICMCI 2.2 says AI does not reduce accountability, and the common law asks whether you exercised reasonable skill and care, which includes checking models before clients rely on them.

Are calculation errors treated differently from bad advice?

Often in practice. A calculation error is objective, so liability is harder to dispute and the argument usually turns to causation and loss. Judgement claims can be defended as reasonable professional opinion. PI usually covers both, subject to the wording.

What documentation helps defend a model claim?

An assumptions sheet, version history, a record of independent testing, notes of which parts were AI-generated and who checked them, and a covering report setting out purpose and limitations. These reflect ICMCI 3.6 on accountability and documentation.

Check your PI covers how you use AI

Tell us about your firm and the work you do. A broker will review how your current PI treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.

Start a PI quote Request a callback

Apex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.