AI's abilities are uneven. It can write a clean first draft of a board memo in seconds and then get a simple reconciliation wrong. Researchers call this the jagged technological frontier: tasks that look equally hard to people can be easy or impossible for AI.
What a large experiment found
In a pre-registered study with 758 Boston Consulting Group consultants, people using GPT-4 on tasks inside AI's frontier completed 12.2% more tasks, worked 25.1% faster and produced work rated more than 40% higher in quality. On a task deliberately chosen to be outside the frontier, consultants using AI were 19 percentage points less likely to get the right answer than those without it. AI made them confidently wrong.
What sits inside and outside the frontier in finance
Usually inside (with review):
- drafting variance commentary from numbers you supply
- summarizing contracts, policies and accounting guidance
- classifying and extracting data from invoices and receipts
- writing formulas, queries and first-draft analyses
Often outside, or risky without strong checks:
- arithmetic over many rows done "in the model's head" rather than in a tool
- judgements that depend on facts the model was not given (your contracts, your policies)
- anything that needs current information the model does not have
- novel accounting questions where a plausible answer is not the same as a correct one
Working with the frontier
- Test on your own tasks, not vendor demos. The frontier moves with every model release.
- Let tools do the math. Ask AI to write the formula or query; let the spreadsheet or system compute.
- Design review around the risk. Low-risk drafts need a quick read; numbers that reach the ledger need a control.