Build smarter search without AI: Fuzzy matching for applications | Data Exposed
Microsoft Developer’s Data Exposed session examines how SQL search techniques handle imperfect business data, with practical comparisons to help developers choose an approach without adding AI, embeddings, or vector databases.
Where fuzzy matching fits
Business application searches often need to handle misspelled customer names, inconsistent product descriptions, duplicate records, incomplete entries, and user input that does not exactly match stored values.
The session explores the newest fuzzy string-matching capabilities in Azure SQL and SQL Server, placing them alongside established SQL search tools:
LIKEpredicatesCHARINDEXPATINDEX- Full-Text Search
- Regular expressions
These existing techniques solve many search scenarios, but they can struggle when the underlying data is imperfect. Practical demonstrations compare fuzzy matching with LIKE, Full-Text Search, and regular expressions, examining the strengths, limitations, and suitable uses of each approach.
Application scenarios
The comparisons focus on common business requirements:
- Customer lookup: Handling misspelled names and input that differs from stored customer information.
- Product search: Working with inconsistent product descriptions.
- Duplicate detection: Finding records that may represent the same entity despite differences in their text.
- Master data management and data cleansing: Addressing inconsistencies across business records.
- Application search: Providing typo-tolerant, user-friendly search experiences.
The central consideration is choosing a search technique that fits the problem. Fuzzy matching adds an option for imperfect text; it does not remove the need to assess existing SQL search approaches or their limitations. For applications whose main challenge is inconsistent strings, the session focuses on addressing that need without introducing unnecessary AI complexity.