AI transparency

How AI is used

BidServe uses a set of narrow agents rather than one general assistant. Each has a single responsibility — discovery, extraction, classification, deduplication, eligibility, matching, bid/no-bid, commercial, compliance, risk, competition, social value, deadlines, evidence matching, partner discovery, notification and learning.

What we record for every material result

• Agent name, model and model version
• Prompt version and input record references
• Output, confidence level and timestamp
• Human reviewer, any correction, and the final approved result

How uncertainty is shown

Where a value or conclusion is uncertain, we label it as not confirmed, evidence required, low confidence or human review recommended. Low-confidence extracted fields are flagged rather than presented as confirmed.

What the AI never does

• Invent contract experience, customers, statistics, accreditations or staff qualifications
• Create policies you do not hold, or conceal a compliance gap
• Overwrite original tender documents
• Submit a tender without human authorisation

Grounding and evidence

Bid drafting is grounded only in the tender documents, your approved organisation profile, your verified case studies and policies, and evidence you select. Evidence references are shown next to generated claims, and unsupported claims are flagged.

Reporting inaccurate output

Every AI result can be reported as inaccurate. Reports feed our evaluation datasets for extraction accuracy, requirement detection, eligibility classification, matching, risk detection, hallucination prevention, evidence grounding and duplicate detection.