Intelligent loan qualification based on future servicing capability

A system comprising:

• a processor adapted to:
• send an access request to access a depository account of a user, the depository account hosted on a banking platform external to a source of the access request, the one or more anonymized historical transaction records originating separately from the depository account of the user;
• and
• upon receiving permission to access the depository account, receive a plurality of transaction records in the depository account of the user, the received transaction records including a chronological listing of debits and credits;
• and
• one or more machine learning model adapted to:
• receive one or more anonymized historical transaction records;
• train on at least on a portion of the anonymized historical transaction records, thereby generating training output;
• and
• generate a weight for one or more loan tier metrics for each of a plurality of loan tiers based on the training output, each loan tier associated with a respective loan tier qualifying score;
• determine a plurality of loan tier metrics based on the chronological listing of debits and credits, the plurality of loan tier metrics including at least one of income, income frequency, income stability, and overdraft history;
• generate a respective metric value for one or more of the loan tier metrics, each respective metric value based on the transaction records in the depository account of the user;
• determine that based on prior activity of the user recorded in the transaction records, a first loan tier metric of a first loan tier has greater importance in determining a loan servicing capability of the user than a second loan tier metric of the first loan tier;
• adjust a selected weight chosen from a first weight of the first loan tier metric and a second weight of the second loan tier metric, thereby generating an adjusted weight, so that the first weight is greater than the second weight;
• compute a user score based in part on applying the adjusted weight to the respective metric value associated with the transaction records that corresponds to the first loan tier metric of the first loan tier, wherein the user score represents a prediction of the user's future loan servicing capability;
• and
• determine that the user score meets or exceeds a loan qualifying score for the first loan tier.

A non-transitory computer readable medium having stored thereon software instructions that, when executed by a processor, cause the processor to perform a series of operations comprising:

• generating a weight for one or more loan tier metrics for each of a plurality of loan tiers, each loan tier associated with a respective loan tier qualifying score, wherein generating the weight for a respective loan tier metric comprises;
• (i) accessing one or more anonymized historical transaction records;
• (ii) training one or more machine learning models based at least on a portion of the anonymized historical transaction records;
• (iii) determining the weight of the respective loan tier metric based on output from
• training of the one or more machine learning models;
• sending an access request to access a depository account of a user, the depository account hosted on a banking platform external to a source of the access request, the one or more anonymized historical transaction records originating separately from the depository account of the user;
• upon receiving permission to access the depository account, receiving a plurality of transaction records in the depository account of the user, the received transaction records including a chronological listing of debits and credits;
• determining a plurality of loan tier metrics based on the chronological listing of debits and credits, the plurality of loan tier metrics including at least one of income, income frequency, income stability, and overdraft history;
• generating a respective metric value for one or more of the loan tier metrics, each respective metric value based on the transaction records in the depository account of the user;
• determining, by the one or more machine learning models, that based on prior activity of the user recorded in the transaction records, a first loan tier metric of a first loan tier has greater importance in determining a loan servicing capability of the user than a second loan tier metric of the first loan tier;;
• adjusting, by the one or more machine learning models, a selected weight chosen from a first weight of the first loan tier metric and a second weight of the second loan tier metric, thereby generating an adjusted weight, so that the first weight is greater than the second weight;
• computing a user score based in part on applying the adjusted weight to the respective metric value associated with the transaction records that corresponds to the first loan tier metric of the first loan tier, wherein the user score represents a prediction of the user's future loan servicing capability;
• and
• determining that the user score meets or exceeds a loan qualifying score for the first loan tier.