From sparse data to coherent recommendations
Most loyalty systems need months of data before they can personalise. This household visited just three times in 18 months — and Kornerway already knows what to send them.
Family 1
Real Partner DataSPARSE — 6 RECEIPTS
- The Gallop ×4
- Moon Koon ×1
- HVRC Topaz ×1
4 reservations (3 confirmed, 1 cancelled)
Hash: 862b0682c9961134
Member count: 2
Universal Taste Vector
We used two independent methods — one reads what the member ordered, the other reads how they behaved. Both arrived at the same taste profile.
| Axis | Family 1 | Pop Mean |
|---|---|---|
| LUX | 0.339 | 0.333 |
| NOV | 0.154 | 0.144 |
| AES | 0.325 | 0.328 |
| SOC | 0.353 | 0.302 |
| AUTH | 0.401 | 0.410 |
| SVC | 0.323 | 0.327 |
| SENS | 0.349 | 0.370 |
| PLAN | 0.306 | 0.331 |
| VAL | 0.300 | 0.288 |
| WELL | 0.247 | 0.222 |
Social, Wellness-Oriented Spontaneous Households
48 households
- The Gallop Summer Cantonese Menu dining 0.9906
- Moon Koon Restaurant Premium Homestyle Menu dining 0.9904
- Provincial Cuisine Dining - Happy Valley dining 0.9903
- Guangdong Indulgence Delicacies at Oi Suen dining 0.9901
- Pak Hop Chinese Cuisine - Sha Tin dining 0.9875
Cohort label reflects shared dimension signature; recommendations reflect Family 1's specific household centroid within the cohort.
Does the same engine work when there’s more data?
The engine works at both ends of the data spectrum. Family 1 has 6 receipts — sparse — and still receives specific, relevant routing. Family 2 has 24 receipts — dense — and receives more precise routing with higher confidence. They land in different cohorts (7 vs 5), as expected. The difference is confidence, not capability — and both beat sending everything to everyone.
Family 1
6 receipts · Agreement 0.95 · Cohort 7
Family 2
Synthetic Comparison24 receipts · Agreement 0.97 · Cohort 5
Representative dense household for comparison
Family 2 is a representative synthetic dense household for demo comparison purposes.