Loading page…
Most companies have the same customer recorded more than once, and records that look alike but are not the same organisation. The duplicate finder works through two lists of companies and says which are true duplicates, which are related, and which only look alike, with the evidence for each.
Step by step
For example, “Find the duplicates in our CRM,” or a question about one company in particular.
It looks for records with similar names, addresses, websites or phone numbers, across both lists and within each one.
For each pair it goes through the details one field at a time: what agrees, what differs, and what the difference means.
Before anything is called a duplicate, a second agent makes the case that the two are different. The finder has to answer it.
Merge, link as related (a parent and its subsidiary, say), or leave alone, each with the reasons laid out.
Judgment
These are the moments a simple search tool gets wrong. The demo data is built to include them.
A hospital and its charitable foundation can share a building and most of a name, and still need to stay separate.
Franchises often list the same head-office number. That makes them related, not the same customer.
A missing letter at the same address is usually a typo. The same name in a different city usually is not.
Who does the work
A large, capable model that shortlists pairs, compares them and gives the verdict.
A smaller, faster model whose only job is to find reasons two records are not the same.
Every agent here is built the same way: a leading AI model that plans the work and a smaller one that checks it, a set of tools each is allowed to use, and rules on what it must prove before it answers. They run on leading models from Anthropic or OpenAI, inside this website.
Safeguards
Two invented lists: 33 accounts from a CRM and 21 records from a purchased company file. Real duplicates, near-misses and look-alikes are planted in them on purpose.