How to set scoring thresholds?
The scores your users receive will be affected by how much information we can analyse for that user, this in turn is dependent on the following:
- How many online profiles the user has consented us to analyse
- How rich the profiles & access granted by the user are
Having said this we broadly speaking advise our customers to set a general purpose overall threshold like this:
- Single-Source Check: 55% or higher where the user will only login with a single source
- Multi-Source Check: 65% or higher where the user will be challenged to provide multiple online profiles
In addition to the Overall score we also provide scores for the following attributes Name, Gender, Location, Date of Birth, Email and Phone number. We will also Flag certain observations from the data, such as New Profile to provide an indication of a profiles age. You can set thresholds for these parameters and flags in line with your own preferences.
How do we calculate our scores?
We often get asked this question and although we don’t want to give any potential fraudsters an advantage by revealing too much information we can give a high level overview. All profiles we process go through the following stages:
- Collection: collection of the raw data from the social networks (Facebook, LinkedIn, etc.)
- Data digestion: using a combination of processing techniques we turn unstructured data into structured data for analysis. Examples of digested data is ‘number of close friends’, ‘average number of posts’, etc.
- Scoring: applying a combination of scoring techniques to the data available
- Rules scoring: applying our proprietary rules to score the ‘facts’ built up in the fact building stage
- AI scoring: we use machine learning models to processes and scores all data signals
- Profile: After blending the Rules and AI based scoring results we compile the user's profile
In scoring the profiles we look for a number of different indications of realness, these include (but aren’t limited to):
- Correlation: data available correlates both within a single profile as well as across the user provided digital footprint
- Longevity: information available in the profile(s) is consistent with the expected age of a profile
- Activity: the user is showing activity within the profile
These factors are considered and rolled up into a score indicating the strength of the user profile. All scores returned will range from 50% to 100%. As we value consistency to our customers our start point at 50% is for historical reasons and may change in the future.
|Score (%)||What it means|
|0%||The user hasn’t provided access to any information and could not be scored|
|50% - 55%||Highly likely to be either a fake or unused profile|
|55% - 65%||Good user profile but with limited data to support it. This may be due to only a single source having been used or the user not being active on the accounts provided|
|65% - 80%||Good user profile with a significant amount of data to support it|
|80% - 100%||Great user profile with a significant amount of data across strong sources to support it|
For more information about how Veridu identity scores compare to other methods please see this following article; Three Approaches to Identity.
What do you score per user?
Data for the user is available in the Profile Resource endpoint, below is a description of attributes scored
|Overall||Score for the overall strength of the user profile which reflects both individual attributes as well scoring of information which isn’t specific to any of the below attributes|
|Gender||Value and score of how strongly we calculate the user profile indicates either male or female gender|
|Location||Value and score of how strongly we calculate the user to live in a certain location|
|Date of Birth||Value and score of how strongly we calculate the user to be of a certain age|
|Value and score of how strongly we calculate the user e-mail to be theirs|
|Phone number||Value and score of how strongly we calculate the user phone number to be theirs|
In addition to the above we also provide a number of indicator flags which highlight certain characteristics of the profile. Please refer to the Flag Resource for a full description of available flags.
Which providers should I use?
There’s no single answer to this, but we can provide some guidelines. Choosing which providers to use will depend on mainly the following:
- Your target audience: is a certain provider more prevalent within your user community?
- Attributes required: for your purposes are you most interested in the scoring of specific attributes?
The below gives you an indication of which providers provide the best scores for which attributes.
|Provider||Gender||Location||Date of Birth||Phone number|
Table: Attribute Scoring Strength
It is worth checking back regularly as we constantly update this information based on improvements in our scoring system.
- To minimise disruption to customers any significant changes to our scoring will always be implement as a separate version of the API allowing you to opt-in within a timeframe suitable to yourself
- Assumes Veridu Tokens are used or that customer tokens have equivalent permissions