What AI can get wrong
AI can produce answers that look confident but are wrong. This is a known limitation of current generative AI, and every AI feature in Revometrics is subject to it. Treat outputs as a starting point: verify facts, numbers, and citations before acting on them, and involve a qualified person for anything financial, medical, legal, or business-critical.
- Fabricated facts: invented numbers, names, dates, or events that sound plausible.
- Made-up quotes and citations: quotations, papers, and URLs that may not exist.
- Stale knowledge: the model may not know recent events or the latest versions of things.
- Confidence without evidence: the tone can be authoritative even when the content is wrong.
- Misread intent: the AI may answer a slightly different question than the one you asked.
- Reasoning errors: arithmetic, unit conversions, logic, and code can be silently wrong.
- Context drift: in long conversations, earlier constraints and preferences can be forgotten.
When AI queries your data
AI can read your uploaded files, query your connected databases, and call your APIs. Because the numbers come from your real data, outright fabrication is rare, but interpretation errors are common and easy to miss.
- Wrong column or table: a similarly named field (for example
net_revenueinstead ofgross_revenue) changes the answer entirely. - Wrong aggregation: summing instead of averaging, or counting rows instead of distinct values.
- Missing business logic: forgetting to exclude test accounts, deleted rows, or internal users.
- Wrong time range or filter: the wrong date bucket, time zone, or segment.
- Stale scheduled results: when your schema or business logic changes, scheduled reports and dashboards can silently go wrong.
Review generated queries before relying on their results. Column descriptions and clear column names noticeably improve accuracy.
When AI generates content
Text, analysis, slides, dashboards, web pages, code, and diagrams generated by AI are drafts, not finished deliverables. Before using them:
- Fact-check statistics, dates, quotations, citations, and attributions.
- Check that numbers and charts reflect your real data, not placeholders added to fill a layout.
- Treat explanations such as "revenue grew because…" as AI interpretation, not evidence.
- Review summaries of your documents; summarizing can drop nuance, reverse meaning, or skip caveats.
- For diagrams, check that the connections shown exist in your real system or data.
- Test and review generated code before shipping it. It has not been security-audited.
Generated images
Images and logos are generated from text prompts by a third-party AI provider. They can look realistic but are synthetic.
- Do not use generated images as evidence of real events or people.
- Do not generate images of real or identifiable people without their consent.
- Generated logos may resemble existing trademarks; check before commercial use.
Using AI responsibly
- Verify, don't trust: cross-check important facts, numbers, and citations against primary sources.
- Ask for sources: ask the AI to show its reasoning or data, then check it. If it cannot, treat the answer with extra skepticism.
- Use AI as a draft: refine outputs rather than shipping them unedited.
- Keep people in the loop: have a qualified person review financial, legal, medical, hiring, or safety-critical decisions.
- Share only what the task needs: never paste passwords, API keys, or government IDs into a conversation. Connect data sources through their connection settings instead.
- Re-ask if unsure: rephrase the question or ask the AI to check its own reasoning.
Report a problem
If a response is wrong, harmful, biased, or otherwise problematic, use the feedback button on that message, or email [email protected] with a screenshot or copy of the response and what went wrong.