What is spec matching — and why your tender team needs it
Spec matching is the core operation in any tender response: take a requirement from the procurement document, find the corresponding specification in your product data, and determine if it's a match. Simple concept. Brutal execution.
Why spec matching is harder than it sounds
A tender requirement might say: "The ultrasound system shall provide a frequency range of at least 2–12 MHz across all transducers."
Your datasheet says: "Broadband frequency: 1.8–15 MHz (varies by probe model). Standard convex probe: 2–6 MHz. Linear probe: 5–15 MHz. Phased array: 1.8–5 MHz."
Does this match? The answer isn't a simple yes or no. The system's overall range (1.8–15 MHz) exceeds the requirement, but no single transducer covers the full 2–12 MHz range. Whether this constitutes compliance depends on how the tender evaluator interprets "across all transducers."
This is one row. A typical medical device tender has 100–300 of these.
The three types of spec matching
1. Keyword matching (the spreadsheet approach)
Control+F in a PDF. Search for "frequency," find the number, paste it into the tender response. This works for exact matches ("voltage: 220V" → "voltage: 220V") but fails for anything requiring interpretation.
Error rate: 5–10%. Speed: 3–5 minutes per row. For a 162-row tender, that's 8–13 hours just for matching.
2. Rule-based matching (the custom script approach)
Write rules that parse numeric ranges, unit conversions, and known equivalences. "If tender says 'MHz' and product says 'MHz', compare the numeric ranges." This handles structured data well but breaks on free-text requirements and non-standard formatting.
Error rate: 3–5%. Speed: faster than manual but requires significant setup time for each product category.
3. Semantic matching (the AI approach)
Use language models to understand the meaning of both the requirement and the specification, then score the match based on semantic similarity and domain knowledge. "Broadband frequency 1.8–15 MHz" and "frequency range ≥2 MHz" are understood as related concepts, not just string patterns.
Error rate: under 0.5%. Speed: 162 rows in 46 seconds.
How semantic spec matching works
Semantic matching in MedStrato operates in four steps:
- Requirement parsing: Extract the technical requirement, identify the parameter being specified (frequency, weight, voltage, etc.), and normalize the units.
- Product search: Search the indexed product catalog for relevant specifications. Not keyword search — semantic search across datasheets, regulatory filings, and technical manuals.
- Match scoring: Score each candidate match on a 0–100% confidence scale. A 95% match means the product spec almost certainly satisfies the requirement. A 72% match means it's close but has gaps that need human review.
- Evidence linking: Attach the source document, page number, and section reference to every match. This creates a verifiable evidence chain.
The numbers that matter
| Metric | Manual | Semantic matching |
|---|---|---|
| Time per 162-row tender | 32–48 hours | 46 seconds + 2–4 hours review |
| Error rate | 4–8% | under 0.5% |
| Evidence trail | None (copy-paste) | Full source attribution |
| Consistency | Varies by person | Deterministic |
When you need spec matching
If your team responds to more than 5 medical device tenders per month, manual spec matching is your bottleneck. Not might be — is. The math doesn't work: 5 tenders × 40 hours = 200 person-hours per month on matching alone. That's more than one full-time employee doing nothing but cross-referencing PDFs.
Semantic spec matching doesn't replace your clinical engineers. It gives them back the 80% of their time currently spent on work that a machine does better, faster, and more accurately. They can focus on what humans are actually good at: interpreting edge cases, making strategic go/no-go decisions, and building relationships with procurement contacts.