Enhancing Business Central with Address Validation AI
Address validation using Azure OpenAI compares user-entered addresses against existing Business Central address records to identify matches and preserve associated metadata. The solution uses a system prompt to guide AI in matching addresses semantically rather than character-by-character, returning match confidence scores.
post Aardvark Labs
Key points
- Free-text ship-to addresses lose extra data tied to stored Business Central addresses, such as salesperson or regional details, so matching to an existing address code helps keep it
- The user address and the customer’s known addresses are sent as JSON to Azure OpenAI, which returns a match score between 0 and 1 for each address
- The system prompt carries the logic: ignore abbreviation and line-order differences, and treat a state mismatch as strongly non-matching
- In a test, a near-identical address scored 1, while the same street in Atlanta GA versus Atlanta TX scored 0.6, low enough to reject but high enough to flag for review
- The author’s earlier rule-based approach using USPS CASS, Levenshtein distance and hard-coded rules only reached about 70% accuracy
Context
- Features: Address matching with AI, Confidence scoring, System prompt engineering, JSON data formatting, Ship-to address validation, Sales quote integration
Source: Aardvark Labs, community blog. Summary, key points and quotes are derived (CONTENT-NOTICE.md); read the original for the full text.