Automatching is an intelligent process that works behind the scenes in the following matching processes: 

• User-led with one-sided approval
• Admin-led
• Roulette

You can read about each match process in Matching Process Overview. 

How automatching is used

• Admin-led: Admins can click "Run automatching" to optimize all remaining unmatched users, taking into account how much capacity they have left. The results require admin review and can be approved in bulk or one-by-one. 
• User-led with one-sided approval: The same automatching algorithm is used, but the results are sent to the users as a recommended 'top match' instead of being sent to the admins.
• Roulette: The same automatching algorithm is used, and the results are announced to users on a regular cadence as their new match. There is one twist: past matches that were assigned before are not allowed until you've matched with everyone else available. 

Note: The automatching algorithm is not used in User-led with two-sided approval. Users create a shortlist of their favorite matches, using percentage match scores as a guide. 

How automatching works

Automatching analyzes all possible combinations of matches across all users and then assigns the optimal set of final matches among them. 

To illustrate with a small example, assume Mentees 1 and 2 have match scores between 1-10 with Mentors A and B. How would you best match them?

• Mentee 1 with Mentor A: 8
• Mentee 1 with Mentor B: 5
• Mentee 2 with Mentor A: 9
• Mentee 2 with Mentor B: 4

In this case, starting with Mentee 1 and assigning Mentor A would be incorrect, because Mentee 2 would be stuck with the lower fit with Mentor B of 4. 

Now imagine doing this for 1000+ users! This problem is very difficult to solve. 

Automatching also takes into account "capacity": How many mentees can each mentor take on? Note that if someone is already in a match or a matched staged by an admin, this will subtract from their capacity when automatching runs.