Vendor Matrix
AI-powered supplier evaluation — score vendors against a weighted rubric and eliminate bias from procurement decisions.
Overview
Vendor Matrix is TotalApp's AI-powered supplier evaluation screen inside the Purchase module. It reads raw text about a vendor — notes, audit reports, email summaries, capability documents — and scores them against a configurable weighted rubric, eliminating human bias from vendor selection decisions.
The screen is built on the MatrixEngine framework: a two-phase AI pipeline that first enforces hard non-negotiable gates, then runs a nuanced weighted scoring analysis. Results are stored per vendor and displayed in the main table with score badges and status labels.
Two-Phase AI Pipeline
Phase 1 is a strict Linter — if a vendor fails a Must-Have gate, the score is immediately zeroed and analysis stops. Phase 2 is a nuanced Critic that evaluates evidence quality across weighted categories. This design prevents a vendor with excellent pricing but missing quality certifications from appearing Approved.
Quick Start
- Switch to Purchase mode in the header.
- Open Vendor Matrix from the left sidebar (BarChart2 icon).
- Click Add Vendor → fill in name, category, and any initial notes → Save.
- Click the vendor row to open the Report Panel on the right.
- Paste or edit the vendor's raw content in the sandbox textarea (capability statement, audit notes, email thread, etc.).
- Click Analyze — the AI runs the two-phase evaluation and the result appears immediately.
- The vendor row updates: score badge, status (Approved / Rejected), and analyzed timestamp.
Evaluation Pipeline
Phase 1 — Linter (Gatekeeper)
The AI strictly checks every Must-Have item against the raw content. If any single Must-Have is absent or unverifiable, the result is immediate: passed_gate = false, total_score = 0. Phase 2 does not run.
Phase 2 — Critic (Scoring)
Only runs if Phase 1 passes. The AI scores the content against each weighted category from 0–100, providing reasoning text for each score.
Phase 3 — Math
The final score is calculated as: total_score = Σ (raw_score × weight / 100), capped at 100 and rounded to an integer.
Why Gating Before Scoring?
A Must-Have gate represents a non-negotiable requirement (e.g. "ISO 9001 certification"). A vendor scoring 95 on pricing but missing this certification should never appear Approved — the gate enforces that invariant regardless of scoring weights.
Evaluation Rules Panel
The accordion at the top of the screen controls all rules. Changes apply to the next analysis run only — existing results are not retroactively updated.
| Control | What it does |
|---|---|
| Threshold slider | Minimum total_score to count as Approved. Default: 70. Raise for stricter procurement standards. |
| Weight inputs | Per-category percentage. Must sum to 100 (a warning badge appears otherwise). |
| Must-Have tags | Type a gate + press Enter to add. Each tag is strictly checked in Phase 1. Examples: "ISO 9001 certified", "On-time delivery > 95%". |
| Nice-to-Have tags | Soft bonus hints passed to the AI. Not gating — a missing Nice-to-Have lowers the score but does not block Approved status. |
Default Scoring Categories
| Category | Default weight | What the AI looks for |
|---|---|---|
| Fiyat Avantajı | 40% | Competitive pricing signals, discount tiers, total cost of ownership |
| Teslimat Hızı | 30% | Lead times, on-time delivery history, logistics capability |
| Kalite Belgeleri | 30% | ISO certifications, quality audit scores, defect rate history |
Score Colours & Status Labels
Emerald — Approved
total_score ≥ threshold. Gate passed and weighted score met the minimum. Vendor is procurement-ready.
Amber — Borderline
total_score ≥ threshold × 0.6 but below threshold. Vendor shows potential but needs improvement in one or more categories.
Rose — Rejected
total_score < threshold × 0.6 or gate failed. Not procurement-ready. Review the Missing Gates list in the Report Panel.
| Status | Meaning |
|---|---|
| pending | Vendor added, not yet analyzed. |
| approved | Phase 1 passed + score ≥ threshold. |
| rejected | Phase 1 failed or score < threshold. |
Report Panel
Click any vendor row to open the right-side panel. It contains:
Gate Banner
Green (Passed) or red (Failed) with the total score displayed prominently.
Missing Gates
Shown only on failure. Lists exactly which Must-Have items were absent or unverifiable from the content.
Executive Summary
One-paragraph AI verdict summarising the vendor's overall procurement fitness.
Breakdown Bars
Per-category raw score, weighted contribution, and reasoning text explaining what evidence was found (or missing).
Content Sandbox
Editable textarea pre-filled with the vendor's saved notes. Edit and re-run Analyze as many times as needed. Each run overwrites the previous result.
AI Engine
The analysis uses the Writer Engine setting from Settings → Doc View → Writer Engine:
| Setting | Model used |
|---|---|
| API mode (default) | Anthropic claude-sonnet-4-6 |
| API mode + Cohere | command-r7b-12-2024 |
| Local CLI | claude --print fallback (no API key needed) |
Tips
- Use specific Must-Have gates. "ISO 9001 certified" works; "good quality" does not — the AI is a strict linter and will not infer intent.
- Paste the most authoritative evidence. A pasted audit report or capability statement beats a brief internal note every time.
- Re-analyze after vendor updates. When a vendor provides new certifications or updated pricing, paste the new data and click Analyze. The previous result is overwritten.
- Use the Missing Gates list as a remediation checklist. Send it to the vendor as a clear list of what they need to provide.
Content Limit
Raw content is truncated to 4,000 characters before being sent to the AI. For long vendor documents, paste only the most relevant section (certifications page, pricing sheet, delivery SLA, etc.).
Frequently Asked Questions
defaultWeights prop. Changing them requires a code change and redeployment. For custom categories, contact your TotalApp administrator.