Contact quality scores
What is Contact Quality?
Contact Quality is a representation of our software's confidence in the accuracy and completeness of a new contact. Machine learning is not a perfect science, so our scores help you segments data for different actions.
We have three possible quality scores:
- High - we found both a full name and an email address and could confidently associated them together.
- Medium - we found a name or an email and could confidently infer the other part of the contact record.
- Low - we just found an email address and couldn't confidently infer the name.
Phone Numbers and Job Titles:
Phone numbers and job titles do not impact Quality score, but will be mined and associated with contact records when available.
Example of High Quality
In my absence please contact Susan Smith.
She can be reached at email@example.com
We have a first name, last name, an email address that includes the last name, and a phone number that appears near the email address. We can confidently associate all of this into a contact record
High quality contacts are colored green in the app.
Example of Medium Quality
Susan Smith will be handling my
responsibilities while I'm out.
Here we have a first name / last name but no email address. The Siftrock inference engine will attempt to determine the contact's email address based on patterns it recognizes from the sender's address and create a medium quality contact. This can work the other direction as well.
Medium quality contacts are colored yellow in the app.
Example: Low quality
While I'm gone please get in touch with our
Steve in our HR department at firstname.lastname@example.org.
This is the opposite of the last example. We are given an email address but not a name. To make matters worse, the email alias doesn't really provide any insight as to what the person's first name / last name might be. If the email address had been email@example.com the parser would have picked up on the pattern.
Low quality contacts are colored red in the app.
Background on Entity Recognition Technology:
It's worth touching on a few general considerations with entity recognition and extraction of people's names.
- Some names look like regular words. Consider a name like "Rose Green". Both the first and last name are common English words and will often not be seen as a name. If an email address like firstname.lastname@example.org is provided we can easily make the connection. However, if just the name is provided we'll have a harder time interpreting it as a person's name.
- Name variations are continually evolving and changing. Obscure or strangely spelled names are a disaster for machine learning technology because they can be so unique they aren't recognized. The engine has no problem recognizing "Amy" but it's considerably more difficult to see "Aimee" as a person's name.
Ultimately it's the combination of names, proper nouns, and email addresses that can be used to determine an individual's contact details.
This combination and the assumptions that are made is what fundamentally determines the quality (or confidence) of an individual contact record.