Trestle provides a phone activity score through its Phone Validation API and Real Contact API, which indicates whether the phone is connected or disconnected and what activity has been seen. A score of 100 means consistent activity in the last 12 months, and a score of 0 means the number is likely disconnected or no activity has been seen in the past year. We’ve covered in a separate article how to read the Activity Score, so this article provides more details on how the score is calculated.
Trestle calculates this score using data points from publicly available records, carriers, and metadata attributes. Collectively, these data points are fed into a model to produce a score that dynamically updates over time as new data points are observed.
External Data Points #
External data points are collected from third-party providers to garner an understanding of each phone number available via the North American Numbering Plan (NANP). Anytime a phone changes carriers, the information for that number is updated within 15 minutes, and that data feeds into Trestle’s algorithm in near real-time. Trestle uses all of the following external data points to contribute to the phone’s score:
- Call connectivity data from certain carriers and cross-carrier network enablers
- Metadata:
- Line type
- Carrier Assignment
- Name associated and duration of the linkage
- Carrier Signals:
- Connection Status
- Porting Status
Most of these signals have apparent benefits for calculating an activity score, but some are less obvious. For example, line type can be used to understand the behavior patterns of certain types of phones. If a phone is a non-fixed VoIP, the model recognizes that without an identified owner, it will put significantly more weight on a lack of activity than a mobile phone with an identified owner that has not been used in a while.
Another example is phones without a carrier associated with them. The model will place a heavier weight on a lack of activity to bring that phone’s score down more quickly.
Our model has to consider many variables and weights, and all of the data we feed into it is carefully curated to produce the most accurate results possible.
Internal Data Points #
Trestle operates at a scale that enables us to identify data points within our network that contribute to a phone’s score. Certain use cases provide valuable insights into the type of activity associated with a phone. For example, here are a few internal data points that Trestle can feed into the activity score:
- The last time the phone was seen in a verified phone call on Trestle’s network
- The number of times a phone was seen in a verified phone call on Trestle’s network within a certain time frame
In addition to this data, we regularly work with our existing customers to get feedback on our model’s performance. This feedback loop enables us to improve the model, identify generally unnoticed patterns, and ultimately increase confidence in its capability.
What period of time does the Activity Score take into consideration? #
Trestle evaluates phones based on the last 12 months of data points. This means that phones do not suddenly get an activity score of 100 or drop immediately to 0. We look for consistent activity over that entire period and weigh the model for differing levels of consistency. There are some exceptions, of course, such as when the carrier provides a disconnected signal.
But what does consistent activity mean?
It means we see enough monthly signals to signal that this phone is active and in use. After one month or so of no activity signals, the score will begin to drop for that phone number. It doesn’t mean that one activity signal will suddenly bring the score from 30 to 100, but regular and consistent activity will eventually cause a score to rise and hit 100. This also means that very intermittent activity will have a low score, and at times, the model will dip the activity score to 30 (the recommended threshold for indicating likely disconnected numbers) or below.
Side note: Customers often ask us about a certain number where we provided a score of 30, but they dial the number, it rings, and maybe even someone picks up the phone on the other side. While this should be rare, it happens because the phone has had minimal activity across various networks we see and the limited data points we get from the carrier. This is also why our customers, even if not outright rejecting these leads, de-prioritize them, as the chances of them converting with such limited phone activity are pretty slim. They have seen this repeatedly after doing their call-down and other conversion tests.
What factors might impact the model’s ability to produce an accurate score? #
Some types of phone user activity could contribute to a lower-than-expected score. For example, FaceTime, WhatsApp, or Messenger calls are not logged as a data point for activity like a traditional phone call. Someone who uses these apps for all or most of their phone calls is less likely to produce a high activity score as less data is available.
Some regions, like Canada, have less data for calculating activity scores. The network data from Canada is limited, so it’s less likely that a Canadian phone number would have a high activity score.
Additionally, some people, particularly non-mobile phone users, behave inconsistently with their phones. If someone is inconsistent in their phone usage (e.g., using a Google phone number as a secondary phone used only for online sign-up 2FAs), it’s unlikely that the phone will receive a high activity score.
All these factors make it an interesting technical challenge. We are constantly experimenting with new data signals daily and using the latest machine learning models to iterate and improve the coverage and accuracy we provide to our customers for lead filtering, fraud detection, and call prioritization, among other use cases. If you are interested in learning more about or trying out our products, or if you have additional questions, please get in touch with us at support@trestleiq.com.