Matching records is the core procedure of data preparation. Usually it is the second step following data extraction from different sources. Data matching is the process of discovering records that correspond to the same unit, i.e. match records. It is the method that allows find, match and merge units that relate to the same record within one or many databases. Simply speaking it is a way of finding duplicate records and integrate multiple similar items into one. It is also a chance to reveal non duplicate records that look alike but indicate different values. Both ways are essential in the sense of clean, accurate and complete data.

The duplicated records can be in different databases, spread through multiple tables across database. For example, customer A, populated in five rows, may contain the same name and family name but address street or zip code may be different in two records and email address is different in three lines. Another example of data that requires matching, could be having inconsistent six product descriptions that refer to the same product id.

To match data, the first step is to identify the similarities for the identical items and link, connect them. Now when the duplicate records are found you need to be sure that they are really the same, one item. The question arises how to be sure that these five records are the same person or six product descriptions in reality represent one same product described differently. To answer it you need to evaluate all links between records, create all sets of matching data and merge the outcome. So the task is to find ‘identifiers’, attributes that characterize a certain record. However, many attributes are prone to change and it creates a challenge.

Today, to match records there are two common approaches. ‘Probabilistic Linkage’ of records is the approach that determines the odds, probability that evaluated attributes belong, link to the same customer (group, class, type), i.e. match. The identifiers, for example, can include address, date of birth, email, gender, initials, country of origin, etc.

Another approach is called ‘Deterministic Linkage’. It includes comparative analysis of attributes that reveals similarities. Comparing records across database allow to find matching sets. The first approach based on probability is more widely used.

Matching data usually starts with data standardization: sorting data based on the same attributes that are stable, unique and unexpected to be changed. It could be some sort of name, country of origin, ethnicity, volume, measurements, etc. Next step is to match attributes by similar features, marks . Follow up with evaluating, estimating the grade of importance, weight to each attribute and assume the probability of the match. Summarizing the grades will allow to define the best match by record that gains the heaviest score.

Another way of matching is recognizing one item from a group of identical items as a ‘master record’. It might be characterized by recency, frequency, highest value or other indicator. As a result this record becomes the true standard for other similar data that will be matched and merged. In some cases the ‘master record’ is randomly chosen among similar items that are being merged and matched accordingly.

Data matching is one of the techniques that refines and effectively cleans data. It allows to remove duplicates and consolidate data. It is a mandatory step in data preparation that helps maintain clean data and, therefore, improve data quality. It enhances the chances of purified metrics and have a huge impact on data driven decision making. It is vital for customer service departments, sales teams, business intelligence units, CRM and marketing groups and upper management affecting the efficiency of their work and organizations performance overall.

Looking for data matching service provider? Look no further, StrategicDB is a data cleansing company specializing in data matching between different sources and formats. Contact Us today at hello@strategicdb.com or 877-332-4923

Menu