Consumer Scoring: Machine Learning Models for Predictive Analytics

Artificial intelligence (AI), machine learning, and algorithmic decision-making have been present to some degree in the modern business environment prior to the COVID-19 pandemic.

When digitalization became a make-or-break sprint for businesses throughout the global economy upon worldwide shutdowns and stay-home orders, the demand for tools to manage the deployment and continuation of AI was strained further than previously thought possible, let alone socially acceptable.

As some of the limitations of the pandemic cease, businesses and organizations internationally have been forced to address the concerns of consumers and regulatory agencies alike. While the technological adoption has brought forth a necessary amount of backlash over privacy, fairness, and accountability issues, AI, machine learning models, and predictive analytics have enabled a digital revolution across a multitude of industries.

With regulatory agencies racing to keep pace with continuously-evolving big data analytics playing a critical role in this transformation, the net impact of this rapid digitalization on consumers and the overall customer experience is yet to be adequately determined.

The Internet of Things (IoT) has risen concurrently with devices that detail demographic, geographic, transactional, and behavioral data conducive to business use cases. With this rise in consumer data, businesses and organizations have been collectively challenged to find methods of processing data simultaneously with real-time analytical models to better understand and manage consumers.

This ecommerce ecosystem, one that is an accepted reality to our everyday lives, has cultivated increasingly sophisticated automation tools for predictive analytics. A primary usage of these predictive analytics, AI, and machine learning tools has been for algorithmic decision-making that “scores” consumers.

This practice is applied for the purposes of determining customer lifetime value, segmentation and modeling, risk assessment, and fraud detection.

Given the ever-growing volume of unstructured data constantly in our everyday lives, computational models leveraging machine learning technologies with automated decision-making has become a core feature of business predictions on consumer behavior. The business cases for these models is based on metrics within the perceived amount of risk, trust, and fraud associated with a consumer. These scores utilize complex data analytics with AI and machine learning to evaluate then apply these consumer measurements, and the decision is automated for business workflow. While historically, automation, machine learning models, and AI has coincided with data science and statistics, modern software applications allow business users across the enterprise to use these tools. Many of these platforms feature intuitive graphical interfaces that allow users to “drag and drop” automation use cases and workflow definitions.

While statistical modeling techniques, such as regression analysis and decision trees, have been a longstanding component of consumer behavior predictions in economics and business, the rapid evolution of AI and machine learning has permitted modern enterprises to process and analyze larger and increasingly complex data sets, often with data sources from social media content, web traffic, social networks, sensors data, and sometimes user gestures.

Models are developed using AI and machine learning techniques, beginning by training the model on historic data, then allowing the model to predict future behaviors based on the fit to the behaviors observed in the training data. Machine learning is typically used to examine these high volumes of unstructured data sources without being explicitly programmed to do so, detecting patterns to culminate to predictive analytics.

Many times, this machine learning model development is supervised and lead by a human, who builds out the training data set by labeling attributes and terms for the AI system to learn. However, the value in machine learning models does not transpose into this human-derived learning process; when an AI system is granted unsupervised learning, it is able to extract hidden patterns in data sets that is not legible to the human eye.

Businesses place a high value on this insight to help them recognize correlations and fits otherwise not possible through testing human hypotheses.

The business value in AI systems that employ machine learning to garner predictive analytics is clearly evident for modern enterprises and organizations. Adoption of AI and machine learning technologies is occurring in all industries, with financial services being one of the most mature industrial AI applications, with 32% of organizations already deploying AI tools in 2020.

Some of the most pertinent industry applications of consumer scoring are segmentation and modeling, customer value scores, risk and alternative credit, fraud, and facial recognition.

Some of the significant use cases of machine learning, AI systems, and predictive analytics for modern businesses include:

Segmentation and Modeling
- Increasingly rich data and metadata allows marketers to develop segments to better organize their audience and target personas
- Data sources include metrics on customer identity, behavior, purchase history, and psychographics
- Business usage of predictive analytics offers the oppurtunity to realize hidden correlations and vital relationships between consumer trends and their behavior
Customer Value Scoring
- These metrics provide a methodology for modeling various transactions of established customers and extrapolating new oppurtunities and leads
- Conducive for tailoring customer experience
- Example: A higher-value customer may be routed directly to customer service whereas a lower-value customer would be placed on hold
Risk and Alternative Credit
- According to estimates from the Consumer Financial Protection Bureau, approximated 45 million American consumers have thin files (limited credit history) or are invisible (no credit history)
- To build models to understand these consumers, alternative data is used to develop an alternative credit score
- Alternative data sources include:
  - Rental payments
  - Mobile phone payments
  - Bank account transactions
Fraud
- Risk scoring tools allow businesses to identify legitimate consumers and transactions, detect bots, and automate security decisions for transactions through utilizing scores
- Economist Intelligence Unit survey: 15% of banking executives internationally believe that customer fraud detection is the most valuable AI asset for banking
- Predictive analytics are used to identify patterns of abusive behavior in consumer returns to retailers
  - Automation of decisions on problematic returns for a flagged user
Facial Recognition
- Machine learning and AI techniques in facial recognition has created applications aimed at scoring the affective and emotional states of consumers
- Affectiva is a firm utilizing deep learning networks, computer vision, and speech analysis to categorize facial expressions and tone of voice
- Outputs include engagement and experience scores
- HireVue video interviewing software examines the facial movements, word choice, and tone of candidates to output an employability score
  - Machine learning model for high performers

As the data ecosystem surrounding automation of the customer experience and consumer behavior advances to include more detailed and proprietary data sets and ever-growing model granularity, it is discernible that consumer scoring by means of machine will continue rising in business use, as well as the role it plays in representing consumers in transactions. Understanding how this data is extracted, transformed, and leveraged for business use is vital to optimizing consumer experiences and outcomes on both sides of a transaction.

The Influence of Quantitative Easing on the Debt Ceiling

Recent news headlines have been focusing on the debt ceiling, but what exactly does this spending cap entail?

Congress sets limits on how much the government can spend and once that spending maximum is reached, the two actions can be taken on the ceiling. At this point, the debt ceiling must be raised or suspended before the Treasury is allowed to take on more debt.

Because of the negative economic consequences caused and exasperated by the coronavirus pandemic, a subsequent financial crisis ensued. To decisively address this crisis, the Federal Reserve had to intervene through the purchase of millions of dollars worth of government securities. This quantitative easing, which subsequently increases the demand for government securities, the price of these securities increased which, in turn, caused their yields to decrease. Given that investors constantly desire strategies that maximize their returns at the lowest initial investment possible, these raised security prices and consequentially lower yields has driven many away from the safety associated with government securities, oriented towards other alternatives, including corporate securities and the stock market. Thus, because of the high levels of quantitative easing conducting by the Federal Reserve on the market for securities, the American government finds itself nearing the debt ceiling.

High government spending has been kicked into high gear since the summer of 2020, however. So why is it suddenly a big crisis at this point in time?

The news has been emphasizing coverage on the debt ceiling recently due to the decision of the Federal Reserve to spend at least another $80 billion in Treasury securities, a target set during the FOMC meeting on September 21, 2021.

Unfortunately, this is not the only event that can potentially be detrimental to the markets.

After the recent FOMC meeting at the end of September, Federal Reserve chair Jerome Powell announced the plans of beginning the tapering process sometime in the near future, meaning the Federal Reserve will slowly begin to apply the brakes on the pandemic-induced quantitative easing strategies they used to stimulate the economy during the 2020 recession. This announcement is beneficial from the aspect of reaching the debt ceiling: decreasing spending on government securities means remaining below the debt ceiling, or at least delaying how soon it is hit.

But how soon is the tapering process going to kick in? In an FOMC statement regarding these actions, Chair Powell states that tapering “may soon be warranted.” Some interpret this to mean tapering will begin once this debt ceiling quarrel is resolved. The Federal Reserve is hesitant to spring into action immediately due to the unintended effects it might have on the markets. Instead, they must ease into new plans to ensure markets remain relatively stable. Once tapering begins, it is a good sign to investors and all Americans that the economy is strong enough to stay afloat without the help of federal quantitative easing.

In a perfect world for investors, the debt ceiling is increased, and subsequent tapering measures are implemented soon afterwards. Instead, the reality is that we must wait for Congress to determine how it seeks to settle this monetary and fiscal discrepancy.

Rising Inflationary Concerns in American Markets

During the month of September 2021, only 194,000 jobs were added to the U.S. economy.

This job report proved a disappointing result due to the 366,000 jobs that were added over the month of August, as well as the over 1,000,000 jobs added in the month of July.

As the pandemic continues to evolve, the United States continues to recognize an increase in economic growth as desirable to economic recovery and expansion. However, the recent stagnancy in the recent job report appear to suggest the existence of limiting factors preventing a full economic bounce-back in American markets.

As people slowly return to work and the unemployment rate begins to decline is when more economic growth is expected. With labor shortages across the country, there are still people remaining unemployed and refusing to work. As a result of labor shortages, businesses are affected and are forced to make decisions like raising prices or shut down. Inflation is a growing concern to economists because of cost-push inflation, demand pull inflation, and the labor market.

Cost-push inflation, often referred to as wage-push inflation, can be described as a rise in consumer prices due to increases in labor wages that impact businesses. The Federal Reserve is concerned of wage-push inflation in the present environment because it may force businesses to increase the wages offered to new hires, reducing their ability to leverage these resources in other significant areas.

As a result of pandemic-instilled policies lenient with the distribution of unemployment benefits, millions of people remain on unemployment because of health concerns over COVID-19 variants, aided in part by increased benefits. Workers in industries such as leisure, food, and entertainment were receiving more weekly income working zero hours a week on unemployment than if they were employed full-time on their weekly wage for their respective workplaces. Unemployed workers decided to stay at home and earn more money than return to work, a financial decision logical given the labor-payout variances. As a result, businesses have been severely affected.

Desperate to attract workers, businesses have been increasingly forced to raise wages to a level proportionate with their competitors, frequently just to catch the attention potential human resources. When the wages provided at a firm increase, the firm is then often forced to raise the prices it charges on goods or services to maintain profits. To address this, several states decided to prematurely end the additional federal unemployment benefits. By doing so, state government hoped workers would be encouraged to return to work, resulting in a decline of unemployment rates.

While more people have returned to jobs and the current unemployment rate is at 4.8%, the Fed remains concerned. The issues brought forth by cost-push inflation issues remain key concern until the unemployment rate continues to steadily decline. Additionally, an unemployment rate near 4.0% and 500,000 jobs per month appears to be the monthly job target for the Fed based on their job reports. These unemployment numbers can assist businesses with hiring workers while maintaining stable prices beneficial for consumers.

Demand-pull inflation is very similar to cost-push inflation. Demand-pull inflation refers to inflation occurring as a result of increasing prices because of limited supplies. As seen in the graph, supply shortages affect demand which in turn raises prices. For firms and businesses across the country, shortages in supplies are a result of the pandemic. Different market items across different industries are experiencing shortages. For example, raw materials like plastic and lumber are currently in shortage. These shortages in lumber occurred because as workers in the lumber industry returned to work, the demand for housing was so high lumber industries were unable to keep up. As a result of the lumber shortage, house prices have increased significantly. The median price of a house rose 15%, by the end of 2020. The current housing market is experiencing high rates of inflation because of the lumber shortage, working from home, and the low mortgage rates at the beginning of the pandemic.

Additional industries like the restaurant and bar industry are dealing with supply shortages of their own. Chicken shortages are affecting the restaurant industry the hardest. Chicken wings across the country are severely low in supply because of winter storms in Texas. Unable to produce as much chicken wings for businesses, firms are forced to raise their prices in order to obtain chicken. This in turn forces a restaurant to raise their price on the consumer to compensate for the increased price of the supplier.

Although demand-pull inflation and cost-push inflation is concerning for the Fed, many economists believe these inflation numbers will only be transitory. The Federal Reserve aims for a 2.0% flat rate for inflation each year. Before the pandemic began, inflation numbers were substantially below 2%. Some economists are accepting the fact of a transitory inflation period of around 3-3.5% for 2021 and some of 2022. These higher numbers may be a concern for most economists, but others believe a year of high inflation will average out to 2.0% over a several year outlook because of the low rates the previous years.