Skip to main content

Using AutoMerge

How to find, validate, and merge duplicate data inside your Microsoft Dynamics 365 CRM.

This guide covers the AutoMerge experience inside your CRM: the duplicate views, the ribbon buttons, and what happens to your data when a set is merged. To configure the matching, ranking, and precision rules that drive the analysis, see Administering AutoMerge.

AutoMerge works on Leads, Accounts, and Contacts. Examples throughout use Contacts, but the views, buttons, and behavior are the same for all three.

Where to start

You areStart here
Setting up AutoMerge for the first timeInstall the Solution
An end user cleaning up your own recordsPotential Duplicates subgrid
An analyst merging duplicates in bulkDupes list views
Troubleshooting a failed mergeCommon errors
Just evaluating AutoMergeFAQs or a free Data Quality Report

How AutoMerge works, in short

  1. The service analyzes your CRM using your matching rules and tags records that belong to a duplicate set.
  2. Each set's members are sorted by your ranking rules. The record ranked #1 is the winner-designate.
  3. Each set is scored 0–100 by your precision rules, from loosely matched to closely matched.
  4. You review the sets and AutoMerge them. Winners stay active; losers are deactivated, and their field values are preserved onto the winner.

Nothing is merged until you say so.

Three ways a merge gets triggered

WhoWhereHow
Power userYour CRM, dupes list viewsSelect records from one or more validated sets, click AutoMerge
End userYour CRM, Potential Duplicates on the formCompare fields side by side, then merge that one set
AdministratorManagement AppSubmit an AutoMerge request across all tagged sets above a precision threshold

All three behave identically underneath: records ranked 2 and higher merge into the Rank 1 record, in ascending rank order.

Work high precision first

Precision scores let you merge confidently. Start with the high-precision sets, which need the least validation, working downward to the lower-precision sets need more human validation. Over time you will naturally zero in on a precision-threshold for your particular dataset that separates the high from the low.