Customer Identity Unification - Resolving Scattered Customer Data with AWS Entity Resolution
An explanation of matching customer data records (entity resolution) with AWS Entity Resolution. Covers the three matching workflows (rule-based, ML-based, and provider service-based), ID mapping, pricing, privacy protection, and integration with Clean Rooms.
The Challenge of Matching Customer Data
A company's customer data is scattered across multiple systems such as CRM, e-commerce sites, call centers, and marketing tools. The same customer is registered with different notations in different systems ("Taro Tanaka" and "TANAKA Taro," "Minato-ku, Tokyo" and "Minato-ku"), and linking these to the same person, known as record matching (entity resolution), has been a long-standing challenge. Previously, you had to implement exact-match or fuzzy-matching logic yourself, but because the patterns of notation variation are so diverse, achieving both accuracy and coverage was difficult. AWS Entity Resolution is a managed service that matches customer data records with three types of matching workflows: rule-based, machine learning (ML)-based, and data service provider-based.
Matching Methods and Configuration
Entity Resolution provides three types of matching workflows (as stated in the official documentation / as of August 2026). Rule-based matching applies the matching rules you define in order from the top in a hierarchical waterfall manner. There are two rule types, Simple and Advanced, and the rule type cannot be changed after the workflow is created. Advanced assumes that one column is mapped to one input type, can handle both exact matching and fuzzy matching, and lets you build conditions with AND / OR and parentheses, but it cannot be used for real-time processing or ID mapping. Simple lets you combine multiple columns into the same input type, and while its conditions are limited to exact matching and AND only, it supports real-time processing and ID mapping in addition to batch processing and incremental processing. In ML-based matching, a machine learning model prepared in advance by AWS comprehensively evaluates attributes such as name, email address, phone number, address, and date of birth, and outputs match groups. Because the model handles notation variations, abbreviations, and differences in format, no effort to define rules is needed. Data service provider-based matching is a method that matches records using identifiers held by a provider such as LiveRamp, and it requires a separate subscription with the provider. Separately from these matching workflows, there is an ID mapping workflow that maps records in one identifier system to another system, and it can be run in two ways: rule-based, and provider service-based using LiveRamp transcoding. Input data refers to data sources on S3 through the AWS Glue Data Catalog, and you can specify up to 20 data inputs in one workflow. Schema mapping maps the columns of the input data to Entity Resolution input types (name, address, phone number, email address, etc.).
Use Cases and Clean Rooms Integration
The main use cases are customer unification for marketing (unifying customer data from multiple channels to build a 360-degree customer view), fraud detection (detecting that different accounts belong to the same person), and data cleansing (detecting and merging duplicate records). Through the integration with Clean Rooms, you can perform record matching without sharing data between organizations. For example, you can use it for a use case in which an advertiser and a publisher identify their common customers and measure advertising effectiveness without disclosing each other's customer data. In this setup, each member brings data that is under its own control and only the matching is performed within the collaboration, so no raw data itself is handed over. If you choose provider service-based matching, you can use an identifier such as LiveRamp's RampID as a token that represents a customer, so you do not have to pass customer data directly to an advertising platform. Note that Entity Resolution used on Clean Rooms has a separate billing structure from when it is used on its own.
Entity Resolution Pricing
Entity Resolution is billed by counting the number of records processed in units of 1,000 (as of August 2026 / measured in N. Virginia and Tokyo). One record refers to one row regardless of the number of columns, and records that did not match are also billed as processed records. The unit price is $0.25 per 1,000 records for both rule-based matching and ML-based matching, and $0.10 per 1,000 records for data service provider-based matching (the provider's subscription fee is separate). ID mapping workflows are also billed based on the number of records processed. Because the unit price is not stated on the official pricing page, checking it with the AWS Price List (Pricing API) is the reliable way. The official pricing page states that prices do not vary by AWS Region, and we have confirmed that the rule-based and ML-based unit prices are the same in the N. Virginia and Tokyo price lists. Note that data service provider-based workflows (both matching / ID mapping) are offered only in three US Regions, Ohio / N. Virginia / Oregon, as of August 2026, and they cannot be used in the Tokyo Region. Also note that they are not covered by the Free Tier. From these unit prices, processing 300,000 records works out to $75, and processing 1 million records works out to $250. The first matching run covers all records, so the cost is large, but with rule-based matching, if you enable incremental processing, only new / updated / deleted records are billed. If you add 100,000 new records to a table in which 900,000 records have already been processed, you are billed only $25, which corresponds to the incremental 100,000 records. This incremental processing cannot be used with ML-based matching. When you use Entity Resolution on Clean Rooms, the billing basis changes: data preparation is $0.10 per 1,000 processed records, and data matching with the rule-based method is $0.50 per 1,000 matched records. A data matching base fee of $100 per collaboration is added to this (the base fee applies only to the rule-based method). Data matching with the provider-based method is $0.10 per 1,000 processed records. In either case, standard charges for S3, where the input data is stored, and for the AWS Glue Data Catalog are incurred separately.
Data Preparation and Schema Mapping
To match customer data records, you first need to prepare the input data. Because customer information scattered across multiple systems often has inconsistent field names and formats, you perform schema mapping to map which fields correspond to name / email / address / phone number. Normalization, which evens out notation variations, is also important for improving matching accuracy. For example, you even out the ways addresses are written and the differences between full-width / half-width characters. The quality of the input data greatly affects the results of record matching. Carefully preparing the data before matching is the prerequisite for highly accurate entity resolution.
Tuning Matching Accuracy and Interpreting Results
You tune the accuracy of record matching using cues that differ by method (as stated in the official documentation / as of August 2026). In rule-based matching, the order in which the rules are arranged hierarchically itself becomes the indicator of accuracy. Each output match group is labeled with the number of the rule that produced that combination, so you can interpret records matched by rule 1 as having been matched more precisely than records matched by rule 2. How far records are regarded as the same is controlled by whether you make the conditions of each rule stricter or looser and by the order in which you arrange them. In ML-based matching, on the other hand, a confidence score from 0.0 to 1.0 is output for each match group. This score is an output value that represents how likely it is that the model judged the records to be the same entity, and you cannot customize the model itself. Therefore, accuracy is tuned not by retraining the model but by designing where to draw the line on the output scores before passing them to downstream processing. In addition, normalization of input data is enabled by default, and removal of special characters and extra spaces and conversion to lowercase are applied automatically. You can also disable it if needed, but because it is effective at absorbing notation variations, you normally keep it enabled. It is important to tune the method and the line you draw according to the use case while balancing the risk of mistakenly merging different people against the risk of missing the same person.
Privacy and Governance
Because matching customer data records involves sensitive personal information, consideration for privacy is essential. Entity Resolution has mechanisms that let you process data while keeping it under your own control, so you can match records without moving data outside. This lets you integrate data securely while keeping its location within your company. For the personal information you handle, you keep to the purpose of collection and the scope of consent for use, and you comply with applicable laws and regulations. You manage who accessed which data and prevent use for purposes other than those intended. While integrating data for convenience, you are required to also maintain controls that protect customer privacy.
Integration and Use
Customer profiles integrated through record matching can be used in a variety of situations. By bundling behavior histories that were scattered across multiple channels into one customer, you can grasp the whole picture of that person, and consistent handling becomes possible. In marketing, this leads to accurate targeting with duplicates removed and to initiatives that span channels. If you connect it with an analytics foundation, you can perform accurate aggregation per customer and evaluate lifetime value. There are also uses that combine it with mechanisms that let multiple companies match data securely. An integrated view of the customer raises the quality of data use by a level.
Summary - Guidelines for Using Entity Resolution
AWS Entity Resolution delivers record matching of scattered customer data as a managed service. Its main strengths are that you can choose from three types of matching workflows, namely precise control with rule-based matching, automatic handling of notation variations with ML-based matching, and matching that uses the identifiers of a data service provider. Its other main strengths are that it offers an ID mapping workflow that spans identifier systems and that the integration with Clean Rooms lets you match records without sharing data. It suits organizations in which customer data is scattered across multiple systems and building a unified customer view has become a challenge.
References (Official AWS Resources)
The primary sources for this page are the official AWS website and documentation. Check the official pages below for the latest specifications and pricing.
- AWS Entity Resolution official page
- AWS Glue official page
- AWS Clean Rooms official page
- AWS Documentation (official)
If this page and the official documentation disagree, treat the official documentation as authoritative.