Three layers of hotel mapping: SQL, scripts and AI explained
Why Syncro maps hotels in three layers, what each layer does, and how a human review step keeps automation trustworthy.
There is no single technique that maps every hotel correctly. Rules are fast but literal, fuzzy scoring is flexible but needs tuning, and AI can reason about messy records but is the most expensive step. Syncro uses all three, in order, so each technique does the work it is best at.
Layer 1: the SQL algorithm
The first layer runs directly on the database and clears the obvious matches in seconds. It uses rules such as:
- a normalised-name match after cleaning punctuation, accents and filler words
- geo-radius proximity between coordinates
- country and city gating, so hotels in different cities are never compared
- phone, email and website matches
- chain and brand codes
- address standardisation
Most straightforward matches are resolved here, quickly and transparently. Whatever this layer cannot decide with confidence moves on.
Layer 2: script batching
The second layer runs as scheduled batch scripts over the unresolved candidates. It scores each pair across many attributes at once:
- fuzzy matching for typos and small spelling differences
- token and n-gram similarity, so word order matters less
- phonetic matching for names that sound alike
- weighted multi-attribute scoring across name, geo, address, postal code and stars
- blocking and candidate pruning, so only plausible pairs are compared
- multi-language transliteration, including Arabic and English
Because it runs in batches, this layer can afford heavier comparisons without slowing anything down for your users.
Layer 3: AI-based mapping
The hardest cases reach the third layer: re-branded hotels, names translated rather than transliterated, vague addresses, and look-alike sister properties that sit next to each other. Here the AI layer reasons over the whole record, not one field at a time, and returns a confidence score with its reasoning.
The human in the loop
Automation is only useful if you can trust it. In Syncro:
- high-confidence matches go live
- uncertain ones wait for a person on your team, who sees the supplier record and the candidate side by side
- every correction is kept, so the next decision is better informed
Why layers instead of one model
Layering keeps mapping fast, explainable and affordable. Easy matches never wait for an expensive step, the hard ones get the attention they need, and your team only reviews the cases that genuinely need judgement.
Want to see it on real-looking data? Try the simulator or book a demo.