CallTime.AI will display, for research purposes, the public political giving records for contacts in your database, to help you understand one of the key drivers of the Donor Score and Ask Amounts generated by the software. At a minimum, however, CallTime.AI requires a first and last name, and some location information (city is best) in order to accurately display data. Occasionally, the system is able to infer location for you, by linking an email address to social media data.
If you're curious to learn more about this data- its source, accuracy, completeness, etc.- then this article is for you!
We display data that we pull directly from the public record. For federal contributions, we pull directly from the Federal Elections Commission, and for state and local data, we pull from the state and local equivalents of the FEC.
Over-matching vs. Under-matching
Our goal is that when you import a contact record, we display contribution histories that are truly associated with that individual. That is, we don't want to display contributions on a contact record that were actually donated by somebody else. This can happen, however, if multiple donors have the same name. So we compare other identifying information to be sure it's a true match.
Unfortunately, the source of this data has very little other identifying information about an individual donor. But we also don't want to only match data back to your contacts when all identifying information matches perfectly-- because this data is very frequently reported inconsistently (i.e., Donald is an Attorney for one donation, and a Lawyer for another donation), or inaccurately (i.e, the same donor is reported as both Brian and Bryan), or the identifying information has changed (i.e., Alicia used to work at the University of Tennessee and lived in Memphis, and now she works at Stanford and lives in Palo Alto).
This is the same problem you would have if you were manually searching each of these dozens of databases by hand, looking up a donor's contribution history. You would execute a search, and have to try and make some determination about which results were true matches, and you would often be missing some matches in your search, depending on your precise query.
How do we try and solve this problem?
In addition to dramatically speeding up the time it would take to run all those searches yourself, CallTime.AI has built complex algorithms designed to instantaneously determine true matches. For example, we have built programs that:
- Calculate how common or rare a name is, and determine similar versions of the same name (read more about how you can add contact aliases to those individuals who use various names, to capture additional data);
- Generalize an occupation into an "occupation category" to determine when seemingly different occupations (i.e., Teacher and Educator) might actually be the same, and belong to the same person;
- Evaluate the proximity between seemingly different locations, to assess the likelihood they might actually be the same (i.e., Los Angeles and Brentwood).
All of the above- and much more- are then considered in combination with one another, calculating the overall likeliness that two donations belong to the same person, or two different people.
While we strive for only "true matches," our system is designed to err on the side of under-matching, rather than over-matching. That is, we would rather inadvertently omit a donation belonging to a contact, rather than inaccurately tell you that one of your contacts has made a donation that he/she has not really made.
While we strive to be far quicker and more accurate in finding giving histories than you would be doing it by hand, we occasionally fall short of the latter. Here's how we work to continuously improve, and how you can help:
- Every week our data science team runs simulated imports and searches to test the quality of the data for consistency, accuracy, completeness, and timeliness.
- We have a whole team dedicated to investigating any instances where you encounter data that doesn't seem right.
These examples help us better train the models to distinguish between true and false matches, so please email screenshots (including the URL) with a note explaining your concern to firstname.lastname@example.org.
Please sign in to leave a comment.