Home>Credit Scoring Use Case v.1.0


A credit score may be called a person’s financial health that includes the summary of a person’s apparent creditworthiness. This is used while making underwriting decisions and predict the chances of events such as a default on outstanding credit.

Automated credit scoring systems, such as those created by FICO, have become one of the prime determinants for the financial success of Americans over the past 30 years. The credit scoring market is usually segmented into CRAs (consumer rating agencies) and companies that develop and license automated scoring methodologies. Credit reports are created by these CRAs for individual consumers by obtaining data from banks, credit card companies, mortgage lenders, and other potential sources. CRAs, estimates the Consumer Financial Protection Bureau (CFPB), receive almost 1.3 billion updates for over 200 million consumer files each month. The credit reports are then used to score individual consumers using proprietary scoring models that cater to the needs and information held by CRAs and lenders. Per the CFPB, in 2010, more than 90% lenders relied on FICO scores for their underwriting decisions.

Automated underwriting is a modern innovation that utilizes big data and its various applications. Automated scoring tools, similar to primitive roll outs of the FICO score, didn’t find widespread adoption till the early 1990s. However, they were viewed as better alternatives that could step up efficiency and help overcome biases in underwriting, credit scoring, and other forms of discrimination (primarily stemming from a lack of sufficient credit history and .adequate models)


Before the 1980s, individual loan officers and specialists personally screened loan applicants implying a risk of personal bias dependence on manual labor. Newer innovations such as the FICO score, while working well for certain quarters, has also unfairly deprived certain borrowers. Around 64 million customers, says an Experian report, were tagged “unscorable,” which denied them traditional forms of credit. Since traditional credit-scoring models factor in data points without much variety and depth, they can often be myopic while predicting the creditworthiness of consumers without a credit history or minimal credit history. This means lost opportunities for businesses and individuals.

FICO score is primarily based on a consumer’s payment history, the amount owed, the length of credit history, new credit, and types of credit used, while omitting factors such as employment history, salary, and other items that could point towards creditworthiness. Traditional credit scoring models also tread the risk of inconsistency since they are requested through one CRA versus another. Consumers are likely to be confused with different scores and may exclude them from specific opportunities tailored for a specific score, thus adding to the bias. Another form of credit assessment, termed ‘behavioral analysis’ or ‘behavioral scoring’ (creditworthiness by association), judges consumers based on their friends, neighbors, and people with similar interests, income levels, and backgrounds. This data-centric approach impacts a lender’s predictions, quite opaque as it were, and ignores a consumer’s individual merit, actions, employment history and other more rational factors.

To add to these, the inaccuracy of data used during traditional credit-scoring can be a major deterrent. 26% of consumers studied by a 2013 Federal Trade Commission (FTC) study had errors in their credit reports, out of which 13% had errors in paperwork. This meant loans were turned down, higher rates of interest were charged, and other terms and conditions were less less favorable than those who didn’t have the errors. Such errors also affected borrowers and prospective consumers with low levels of education. Also, even if an error is identified in a traditional credit report, mitigating the problem at the outset (and sometimes after repeated requests from consumers) can become challenging due to the amount of time and processes involved. This limits the consumer’s ability to maintain good credit in the future.


xpresso.ai is a Auto-ML AI Ops platform that excels in data collection to inference generation. The Auto-ML framework made available by xpresso.ai leverages the latest machine learning and Deep Learning tools while preparing credit scoring models. Traditional model development methods are lengthy, tedious and often prone to human bias and lack data accuracy in the absence of proper data collection, data models and data cleansing.

xpresso.ai’s automated ML platform enables global customers such as CRAs and financial institutions to analyze, manage, and prevent credit risk and market risk, and enhance investment stability for customers. The implementation supports the automation of all steps required to build ML models. These steps include data collection and analysis, data preparation and modification, algorithm selection, model versioning and performance comparison, model deployment, model monitoring and management.

Data Collection and Analysis

xpresso.ai’s Auto-ML AI Ops framework allows automating variable selection, data partitioning, model performance testing, model tuning and so on instead of manual coding. A major part of the data transformation journey while creating models involved setting up the required infrastructure and collecting raw, continuous data (unformatted, unparsed) from a huge number of social media channels, viz. LinkedIn, WhatsApp, Facebook, and data made available by scoring agencies and financial institutions.

xpresso.ai’s Auto-ML AI Ops framework was leveraged to collect and analyse this extensive data repository and analyse details. Details such as name, current employer details, designation, years of experience, employment history, as job rotation, years of experience, frequency of job change, key skill endorsements, group memberships, friends, online behaviour, buying patterns, from media such as Facebook and LinkedIn were collected, analysed and added as exploratory variables.

Data Cleansing

xpresso.ai extends data libraries that aid connecting to these data sources, collecting data off those as well as credit ranking and banks, and also helped in data correction such as cleansing, standardization, removal of stale and extraneous data.

Data Preparation

xpresso.ai can read factors from a varied recommendation text connection and generated an output. This was supplanted with additional data collected (with the aid of xpresso.ai libraries) while preparing a predictive model.

Model Processing / Evaluation

From all these variables obtained, models were created and versioned — enabled by xpresso.ai Auto ML-AI Ops framework.These models were also trained and evaluated with both standard and custom metrics, finally rendering a Credit Scoring model.

By adopting a clear, consistent scoring policy, credit managers were enabled to compare customer performance and historical trend analysis without exceptions, usually occurring due to human/personal bias, behavioral analysis, and lack of trained models and accurate data . The individual credit scores were used to create segments, place each segment in the right workflow and create the best actions for each customer.

Speak to AI expert