AI features

Limitations of AI evaluation

Tender Navigator's AI outputs, the evaluation verdicts, reasoning, summaries and assistant replies, are informational and advisory only. They can be less accurate for non-English notices and highly technical requirements, may not reflect very recent regulation, and can occasionally state something plausible but wrong. Verify against the original tender documents before acting.

Last updated

Advisory only

Tender Navigator's AI Transparency Notice, published under Article 50 of the EU AI Act, states that AI outputs in the product are informational and advisory only. They are not legal advice, procurement decisions or binding determinations. In practice:

  • The AI does not make procurement decisions on your behalf.
  • The AI does not submit bids or applications.
  • The AI does not access external databases or live sources beyond what Tender Navigator has ingested.
  • The AI does not profile or evaluate you as an individual; it processes the business content you provide.

Tender Navigator is classified as a limited-risk AI system under the EU AI Act for its standard use by private companies discovering and responding to tenders.

Known limitations

The notice discloses these limitations and their mitigations:

LimitationWhat it means for you
Language and jurisdiction accuracyAccuracy may be lower for non-English tender documents and jurisdiction-specific rules. Always verify outputs against the source documents.
Regulatory recencyModels have training cut-off dates and may not reflect very recent regulatory or legal changes. Do not rely on AI outputs for compliance verification.
Highly technical specificationsOutputs involving specialised technical, engineering or scientific requirements may be less accurate. Expert review is recommended.
Training data biasOutputs may reflect patterns in training data that do not apply to your context, market or organisation. Human review is essential.
Hallucination riskLike all LLM-based systems, outputs can occasionally contain plausible-sounding but incorrect information. Verify critical facts against the source tender documents.

Limitations specific to evaluation

  • The model reads the notice text Tender Navigator holds. It does not open attached tender documents, so requirements that only appear in attachments are invisible to it.
  • The verdict can only apply the criteria in your prompt. Missing criteria are not inferred.
  • The same tender can receive different verdicts from filters with different prompts or models, and a verdict is not re-computed when you edit the prompt.
  • Notices with little text produce weak reasoning; see Why some tender fields are missing.

What to do with this

Treat Matched as "worth a person's look", not "bid on this". Read the reasoning, open the original notice, check eligibility and deadlines at the source, and keep sharpening the prompt when the reasoning shows a criterion being applied the wrong way. See Why a tender was matched or unmatched.