Processing Questions
Once you have created a tender, the next step is to upload your question file. Tendrex parses the file, auto-classifies columns and metadata, and queues questions for AI answer generation.
Supported question formats
Section titled “Supported question formats”- Excel (.xlsx) - each row is treated as a separate question. Tendrex auto-detects the question column based on header names and content. Multi-sheet workbooks are supported — you choose which sheet to process.
- Word (.docx) - questions are parsed from the document structure (headings, numbered lists, or tables).
Uploading questions
Section titled “Uploading questions”- Open your tender.
- Click Upload Questions.
- Select your question file.
- Multi-sheet Excel files: If the workbook contains multiple sheets, Tendrex shows a sheet picker. Select the sheet containing your questions. You can switch sheets later to process a different sheet from the same file.
- Tendrex displays a preview of the parsed questions with auto-detected column mappings. Review and adjust if needed.
- Click Process to start AI answer generation.
Column classification
Section titled “Column classification”For Excel files, Tendrex automatically identifies and classifies columns:
| Column Type | Description |
|---|---|
| Question | The main question text |
| Section | The section or category (e.g. Quality, Social Value, Technical) |
| Reference | The question reference number (e.g. “3.1.2”) |
| Max Score | The maximum marks available |
| Weighting | The percentage weighting for scoring (e.g. “10%”, “15”) |
| Word Limit | Any word count constraints (e.g. “500 words”, “1,000”) |
| Context | Additional context or guidance |
| Metadata | Other columns (scoring type, evaluation criteria, etc.) |
You can manually override the auto-detection before processing.
AI metadata extraction
Section titled “AI metadata extraction”Beyond column classification, Tendrex uses AI to extract and enrich question metadata:
- Section identification - categorises questions by topic (Quality, Social Value, Technical, Commercial, etc.)
- Question number parsing - extracts structured reference numbers
- Word limit detection - identifies word count constraints from the question text
- Scoring type classification - determines if a question is scored, pass/fail, or information only
- Guidance notes extraction - pulls out buyer guidance on what they expect in the response
Monitoring progress
Section titled “Monitoring progress”After clicking Process, you can monitor progress on the tender page:
- A progress bar shows how many questions have been processed.
- Each question shows its status: Queued, Processing, or Complete.
- Processing typically takes 5-15 seconds per question, depending on complexity and the size of your knowledge base.
The AI priority chain
Section titled “The AI priority chain”When generating each answer, the AI follows a priority chain:
- Canonical store - checks your library of approved, canonical answers first
- Historical responses - looks at answers from previous tenders
- RAG chunks - retrieves relevant passages from your uploaded source documents
- AI fallback - generates from scratch when no strong source material exists
Rejected answers are also stored and fed back into prompts so the AI avoids repeating the same mistakes.
Export columns
Section titled “Export columns”When you export a tender’s answers to Excel, the spreadsheet includes the following columns:
| Column | Description |
|---|---|
| # | Question number |
| Section | The question section or category |
| Question | The question text |
| Answer | The approved answer text |
| Weighting % | The scoring weighting percentage |
| Word Limit | Any word count constraints |
| Confidence | The AI confidence level (High/Medium/Low) |
| Sources | Source document citations |
Column values from the original spreadsheet take priority. AI extraction fills any gaps for section, question number, scoring type, and guidance notes.
Re-processing
Section titled “Re-processing”If you upload additional source documents after processing, you can re-process individual questions or the entire tender to generate updated answers using the new material.