Understanding match scores

ⓘ Reading Time: 5 mins

Overview

Spade’s unique approach to transaction enrichment involves matching each transaction to a real merchant entity (”counterparty”) in our database, and returning granular merchant, category, and location information.

Matching accuracy is critical, and we have developed a “match score” model that assesses how confident we are that each match is correct. This is a machine learning model that is constantly fine-tuned to create more and more accurate predictions.

In addition to using this model to filter out matches that don’t meet a quality bar, we surface results via our API to allow you to make decisions about when and how to use our data.

How do I interpret match scores?

Match scores range in value from 0.00 to 100.00 — the higher the match score, the higher the likelihood that a counterparty or location returned was the one involved in a transaction. We return two types of match scores: counterparty match score and location match score.

What is the counterparty match score?

• Counterparty match score is an assessment of how confident we are that a specific counterparty we return is the one involved in a transaction (e.g., how likely it is that WALMART002191BRYANOH is a transaction occurring at Walmart)
• Counterparty match scores appear in the counterparty portion of the response.

What is the location match scores?

Note: location match scores are currently in beta.

• Location match score is an assessment of how confident we are that a specific location we return is the one involved in a transaction (e.g., how likely it is that WALMART002191BRYANOH is a transaction occurring at the Walmart at 1215 S Main Street, Bryan, Ohio).
• Location match scores are only returned when we match on a location that is identified as physical in the spendingChannel field (it is null for digital transactions)
• Location match scores appear in the location portion of the enrichment

When is a match score not returned?

• In the rare event we are unable to match a transaction to a counterparty or location in our backing data we will return a null match score.
  • This indicates that the returned data is not based on a match and is instead derived from what was in the request.
  • It can also be identified by the presence of a counterparty or location ID - if no ID is present, no “match” has been made.
• At times, we may be able to find a counterparty match but not a location match. In these events, the match score in the location portion of the response will be null but the match score in the counterparty portion will be a numerical value.

How should match scores be used?

Match scores can help you make decisions about how to use enriched data in your systems. For example, for decision-making processes (e.g. card authorization flows or fraud assessments) we suggest only using counterparties with a match score of >=90, whereas for customer analysis, budgeting, or UX/UI improvements, counterparties with scores >=80 are sufficient.

*Note that as our model improves and becomes more accurate, these guidelines may be adjusted.