An AI recently claimed that 9.11 is bigger than 9.9. It sounds trivial, but it exposes a real business risk: models can sound confident while getting basic comparisons wrong.
For executives, the issue is not the math error itself. It is what happens when the same kind of mistake reaches pricing logic, inventory planning, credit decisions, or customer communications.
Why this mistake happens
Most language models do not “understand” numbers the way a calculator does. They predict likely text based on patterns. That works well for summaries and drafts. It works poorly when exact values matter.
Common failure points
- Token behavior: The model may treat 9.11 as a longer or more complex string, not a smaller decimal.
- Pattern over precision: It often favors what looks familiar over what is mathematically correct.
- No built-in verification: Unless you connect it to a rules engine or calculator, it can skip validation.
What this means for business automation
AI creates value when it reduces repetitive work and speeds up decisions. But leaders need to separate drafting tasks from decision tasks.
Good uses include:
- summarizing calls and emails
- classifying documents
- drafting responses
- extracting data from unstructured text
Riskier uses include:
- financial calculations without validation
- contract clause interpretation without review
- inventory or demand decisions with no system checks
- customer-facing answers that depend on exact figures
How to reduce the risk
You do not need to avoid AI. You need guardrails.
- Route numbers through deterministic systems. Use calculators, ERP logic, or rules-based workflows for exact comparisons.
- Validate outputs before action. Add thresholds, exception checks, and human review for high-impact cases.
- Test edge cases. Include decimal comparisons, dates, and outliers in your QA process.
- Limit the model’s job. Let AI interpret, summarize, or recommend. Let systems of record decide what is true.
The executive takeaway
The lesson is simple: AI is useful, but it is not reliable by default. If a model can be convinced that 9.11 is bigger than 9.9, it can also mishandle business logic in less obvious ways.
Use AI for speed. Use systems and controls for accuracy.
That is where automation creates real value: not by replacing judgment, but by making better decisions faster and with fewer manual steps.