Scoring Methodology
Every deal on DealZap carries a 0-100 score and a quality verdict. This page explains exactly how those numbers are computed and what they mean.
The AI Score (0-100)
A composite of six sub-scores produced by our AI model. Each sub-score is bounded; the weighted sum is the headline score you see on every deal card.
- Discount score: how aggressive the markdown is vs the product's typical price
- Category score: how in-demand this category is right now (festival-aware)
- Commission score: the affiliate payout this deal would earn
- Timing score: day of week, hour, current festival window
- Price score: how the current price compares to historical observations
- Total score: weighted blend, capped at 100
Score Thresholds
| Score | What happens |
|---|---|
| ≥ 80 | Auto-approved. Eligible for DealZap Choice. |
| 70-79 | Approved. Visible on the site, eligible for publishing. |
| < 70 | Rejected. Never published, never appears in any SEO surface. |
Thresholds are tuned over time as we learn what scores correspond to user trust. The current floor (70) was set after we measured complaint rates against AI score buckets.
The Quality Verdict
A separate AI judgement that runs alongside scoring. The Quality Agent reads the deal's title, brand, image, and category and returns one of three verdicts:
- buy — clearly a real product at a real discount; safe to recommend.
- caution — uncertain; usually a non-mainstream brand or a deal that looks too good to be true.
- pass — clearly suspicious; looks like a fake brand, deceptive listing, or low-quality clone.
Only deals with verdict buy earn the green "Verified Good Buy" badge.
DealZap Choice (Top 5)
A composite ranking recomputed hourly. Selects the top 5 deals across the entire approved catalog by:
- 30% — AI score
- 25% — review strength (rating × log10(rating count))
- 25% — clicks in the last 7 days
- 15% — conversions in the last 7 days
- +15 boost — when the price matches our 90-day low (price history backed)
Who Reviews These
The pipeline is fully automated: AI scoring, AI quality judgement, and rule-based gates. No human reviews each deal. We do, however, audit the pipeline output weekly — sampling random approved deals and checking that the score correlates with real shopper experience. When we find a systematic miss, we update the scoring rubric.