Agent StoreUtilitiesData Preprocessing
Live

Data Deduplication and Cleansing Agent

UtilitiesData Preprocessing

Identifies duplicate, malformed, and inconsistent records across uploaded datasets and produces a cleansed dataset with a change log for review before reuse.

4
Process steps
5
Integrations
3
Data inputs

Teams frequently work with CRM exports, spreadsheets, and merged data files that contain duplicate customer records, inconsistent formatting of names and addresses, and missing or malformed fields, and cleaning that data manually before it can be used for analysis or outreach is tedious and error-prone at scale

Downstream errors from dirty data, such as duplicate outreach to the same customer or failed matches in a merge, damage both efficiency and customer experience

This agent scans uploaded datasets for exact and fuzzy duplicate records, standardizes formatting for common fields like names, addresses, and phone numbers, and flags records with missing required fields for review rather than silently dropping them

It outputs a cleansed dataset alongside a detailed change log so users can audit exactly what was modified or merged

The agent ingests the uploaded dataset and profiles each column to detect type, format patterns, and null rates. It applies fuzzy matching algorithms across likely duplicate-prone fields such as name, email, and phone to identify probable duplicate records, then standardizes formatting for structured fields using configurable normalization rules. Records with unresolvable issues are flagged for manual review rather than auto-modified, and the agent outputs both the cleansed dataset and a structured change log documenting every merge, standardization, and flag applied.

1

Profile Dataset

  • Ingest the uploaded file and detect schema and column types
  • Calculate null and inconsistency rates per field
  • Identify likely duplicate-prone key fields
  • Flag columns with high format variability
Outcome: A full data quality profile of the dataset is established before any changes are made.
2

Detect Duplicates

  • Run exact-match detection on unique identifier fields
  • Apply fuzzy matching across name, email, phone, and address fields
  • Score match confidence for each candidate duplicate pair
  • Group high-confidence duplicates for merge review
Outcome: Duplicate and near-duplicate records are identified and confidence-scored.
3

Cleanse and Standardize

  • Apply standard formatting rules to names, addresses, and phone numbers
  • Merge high-confidence duplicate records, retaining the most complete data
  • Flag records with missing required fields or low-confidence matches
  • Preserve original values alongside cleansed values for traceability
Outcome: A standardized, deduplicated dataset is produced with flagged exceptions set aside.
4

Output and Log

  • Generate the cleansed dataset file for download or system load
  • Compile a structured change log of every modification and merge
  • Summarize flagged records requiring manual review
  • Deliver both outputs to the requesting user
Outcome: The user receives a ready-to-use cleansed dataset and a full audit trail of changes.
Salesforce
Loads cleansed records back into the CRM and pulls source exports
Google Sheets
Accepts spreadsheet uploads and returns cleansed output
Snowflake
Writes cleansed datasets directly into the data warehouse
Amazon S3
Stores raw uploads and cleansed output files with the change log
Slack
Notifies the requesting user when cleansing is complete