Case 11

Master Data Cleansing and Entity Resolution

One company, three spellings, two IDs. An exact match will not find it.

Today

Customers, suppliers, products and items live in several systems, with different spellings and IDs.

Who does the work

Today7 of 7 with a human

With workflow1 of 7 with a human

01Entity extraction
02Candidate matching
03Semantic similarity
04Deterministic constraints
05Conflict detection
06Approval of uncertain matches
07Master data update

One mark: one work step, in workflow order. Today a person handles each of them.

What it is measured by

Duplicate rate
Manual review time
Errors in downstream processes

The workflow

Comes in
Records from several systems
AI understands
Possible matches
- Extract entities
- Semantic similarity
- Context for the assessment
Software checks
Permitted matches and critical data
- Detect conflicts
- Write only after approval
Lands with a human
Uncertain matches
- Decision before the leading system changes

AI finds candidates. Rules limit the matching. Humans decide when in doubt. Only then does the system write.

Case 01
Data capture from approved documents
Models propose, code decides. Code reads what it can read reliably. AI fills in the rest. The human reviews instead of typing.
Case 02
Articles from Researched Sources
Not every draft has to succeed. Only what passes the review gets published. AI writes. Rules decide what moves on. Humans set the rules.
Case 03
Social Media Posts for Every Article
Nobody writes posts anymore. A human decides what a post may do. AI picks and writes. Code checks and posts. A human sets the rules.
Case 04
Inquiries Captured in Conversation
A form asks what it knows. A workflow rarely fits into the first sentence. The agent asks. Software checks the brief. Only what a human approves is sent.
Case 05
Invoice and Purchase Order Verification
The standard case is rarely the problem. The effort comes from the deviations. Automate the standard. Give humans the exceptions.
Case 06
Quote Preparation from Customer Inquiries
The information for a quote already exists. It is just scattered. AI prepares the quote. The expert makes the decision.
Case 07
Technical Document and Compliance Review
The review takes expertise. Finding the evidence does not. AI prepares the compliance review. The decision stays with the responsible human.
Case 08
Customer Support → Engineering
The bottleneck is not the technical analysis, but the missing context. Engineering gets the relevant context before the actual analysis begins.
Case 09
Supplier Comparison in Purchasing
Every offer arrives in a different format. Comparison starts once they all look alike. AI structures the comparison. The purchasing decision stays with the human.
Case 10
Contract and Order Review
The goal is not to approve contracts automatically, but to find deviations reliably. AI finds, structures, compares and escalates. Humans decide and take responsibility.