Credit Industry Ripe for Disruption: AI & Big Data
- There are industries where the prediction of an individual’s behaviour plays a key role (e.g. credit and insurance)
- There is now an abundance of digital data which can be analysed, uncovering new insights to predict behaviours
- Artificial Intelligence is able to analyse these sets of big data in an automated, cheap, and efficient way
- Due to these new behavioural insights, there is a lot of room for improvement in these industries
- Privacy concerns based on this data remain in a grey area and could pose problems
- Hackers and cyber-security firms race to create the most sophisticated AI algorithms
An Outdated Industry
According to the McKinsey Global Institute, 43% of all financial and insurance operations can be automated with existing technologies. We have seen an astonishing rise in the use of AI in many industries – of which financial services is one of them. The credit industry is ripe for AI intervention as traditional methods have clearly untouched values and inefficiencies (we’ll explain below). Furthermore, the playing field is relatively small when it comes to credit reference agencies, there is more than enough room for competition.
Artificial Intelligence Basics
There has been an influx of startups entering the credit industry, leveraging AI tech to offer more efficient and personalised offerings. But how exactly does AI work in order to do this? There are many aspects to AI. We’ll cover the 3 basic concepts – Machine Learning, Deep Learning, and Neural Networks – so that you’re up to speed. If you’re an AI expert feel free to skip ahead.
Machine Learning (ML)
In the simplest terms, Machine Learning is about teaching computers to learn from data and make intelligent decisions or act based upon what they’ve learnt. This is done by supplying them with a large amount of data, a set goal and indicators on how to achieve that goal. Lets take facial recognition as an example. You could teach a computer to learn what an image of a human face is, by providing large data sets of pictures with and without human faces, then by labelling each picture ‘face’ and ‘no face’, the algorithm will be able to learn what a face looks like. With each trial, it will adapt and improve its strategy for identifying a face. Once it has learnt this, you can give it a new set of pictures so that it can predict whether any contain faces or not. If a mistake is made, a human can intervene and correct it – whereas deep learning is when the computer will correct itself by analysing the predictions against a vast amount of other pictures with faces.
Deep Learning is applying the skills learnt from ML in order to make improvements on new data sets. Therefore, when the computer is predicting if the picture has a face, it can then reinforce its decision or reduce its uncertainty via Deep Learning. The Deep Learning algorithm performs a task repeatedly, each time tweaking it a little to improve the outcome. It does this by continuously assessing data via layers of artificial Neural Networks that mimic the decision-making processes in our human brains. In effect, the computer is correcting and improving its strategy. Deep Learning requires big data in order to work, a low supply of data will not be sufficient.
An artificial Neural Network tries to simulate the processes of densely interconnected brain cells. Instead of being built from biology, these neurons, or nodes, are built from code. They are designed to recognise patterns, this is done via three basic layers in the Neural Network; input, hidden and output (there can also be multiple layers for each). Each of these layers contains thousands or millions of nodes. The patterns that are recognised from the data are numerical. All bits of data – whether it is sound, image, text or time related – are converted into numerical vectors so that the algorithm can read it. Neural Networks cluster and classify the inputted data as they spot these numerical correlations. By grouping unlabelled data based upon similarities which human analytics may not have considered, new insights can be uncovered.
The typical way in which the credit industry issues loans is via credit scores. Credit reference agencies gather information such as your repayment history, bills, mortgage and address. Then based on this info, you’re given a credit score. This is the score which banks and credit providers base their decision making on. However, this doesn’t show the whole picture of an individual. Furthermore, human bias can play a role in this process.
AI Potential in Credit Industry
AI credit scoring is cheaper, faster and more accurate than traditional methods as it is automated, based on a larger pool of data and has rules that are more sophisticated and complex. Furthermore, it is not subject to human bias* as all analyses are carried out by a computer based on inputted data. As AI can interpret unstructured data (text, audio, image, video, location ect), this increases and diversifies the variety of data that can be used to determine how credit-worthy someone is. But how is this achieved?
*Machine learning bias is a hot-topic issue in many industries. Machines assess reality and adjust their behaviours based on predictions from past experiences. In this respect, there is a thin line between “bias” and “intelligence”.
As previously explained, AI technology is programmed to uncover patterns. For example, profiles with the same traditional credit score plus a similar behaviour in internet activity, social media or eCommerce transactions, could be clustered together. If a negative behaviour pattern is recognised with their repayments, then this precise digital footprint and traditional credit score combination could be considered as a negative influence for future applications. Whereas the traditional system would not be able to consider this risk factor as it only looks at the credit score. This is especially useful for people who don’t have much credit history, yet have a digital footprint. Their unstructured data could be correlated with patterns of positive repayment behaviours, meaning that their chances of receiving credit would be higher.
As AI technology is specialised in labelling data and finding patterns, anomalies are also easily detected. When it comes to the credit industry, anomalies mean fraudulent behaviour. Furthermore, due to AI’s advanced technology, predictive analytics can be leveraged to prevent fraud from happening in the first place. Such an abundance of data is created meaning that AI technology is able to extract these insights. Whereas a human taskforce wouldn’t have the ‘processing power’, or time for such large analyses.
One worrying factor to consider is the exponential growth in cyber-attacks. It has been reported that hackers are also using AI algorithms in their cyber-crimes… This is where we come to a standpoint, fighting fire with fire. If more attention isn’t payed into cyber-security, then the credit industry’s sensitive info could be left vulnerable.
AI Startups in the Credit Industry
Upstart has a founding team of ex-Googlers. They leverage AI and Machine learning to obtain Credit scores based on additional variables. For students they base their analysis upon their future potential, instead of just their past. They consider factors such as education, employment and income when approving loans. You don’t need to have any credit history, however, a minimum wage of $12k and a low credit score of at least 620 is required. It could be argued that an analysis team could sift through this data, meaning the AI tech wouldn’t be needed. On the other hand, it is more time and cost efficient to automate the process, plus more accurate.
ZestFinance also has a founding team of ex-Googlers, they have a similar product and have made an attempt to reduce bias in their offering. “The traditional way of reducing bias is a slow and manual process that often comes late in the modelling process. It requires making lots of hard choices between fairness and performance and then re-running the model. Lenders end up simply tossing out offending credit signals, which leaves a lot of performance on the table. With ZAML Fair, lenders are in control and able to pick a better model at a fraction of the time and effort required by legacy techniques.”
ZestFinance have stated that using AI predictive analyses in a regulated industry is difficult. This is because it is typically tricky to explain how the computer reached a certain outcome. However, ZestFinance believes they have overcome this ‘black box’ issue. They claim to have transparent decision-making processes, allowing you to validate, interpret and document the reasoning behind the model’s credit decisions. Therefore complying with regulatory guidelines.
One startup aggressively using advanced machine learning to comb through vast sources of alternative data to predict an individual’s creditworthiness is Lenddo. Lenddo’s mission is to improve financial inclusion. They do this with LenddoScore, which is a patented scoring method which combines traditional underwriting with additional unstructured data. Lenddo claims that their system has allowed their partners to approve up to 50 percent more applications.
Lenddo does an analysis on a potential applicants’ digital footprint via their smartphone. They claim to have over 12,000 variables. Lenddo leverages big data via: social networks, browser histories, E-commerce transactions, geolocation data and other smartphone info. The AI algorithm crunches this data and churns out predictive features, producing a credit score as the final product.
Lenddo state that they do not share personal data with lenders. They only share the final result (credit score) of their analysis to protect individuals’ privacy. Furthermore, personal data is only taken with the users’ consent.
Data privacy concerns have become a popular issue in the world today. Now that tech companies have the highest market capitalisations, data is seen as the most valuable commodity on earth. With data comes responsibility. Facebook have had numerous privacy scandals, regulations can’t seem to keep up, and there have been countless cyber-hacks on companies all around the world releasing personal information. It is all too easy to accept the T’s & C’s and turn a blind eye to what is really happening. As people begin to realise the impact their data can have, AI analyses on big data may pose a problem.
Does it seem too invasive to analyse people’s public social media data? Do you feel comfortable knowing that AI chatbots save and analyse texts to build your behavioural profile? Is the line drawn at analysing location data, audio or transaction histories? Insurance and loan companies (as well as many other industries) can provide more personalised offerings by obtaining these extra insights about your behaviour. But the question of whether these methods are too invasive still remains.
The banking industry is facing a FinTech revolution as challenger banks fight for market share.
Startups are beginning to disrupt the insurance industry. The “InsurTech” space has seen a number of innovations. Shift Technology are able to process claims in minutes, instead of weeks, thanks to AI. There are also companies that offer car insurance by the hour and even by the minute. Additionally, Trov allows you to insure individual items for flexible amounts of time, all done in minutes via their app.
Will the credit industry face a similar situation with the rise of startups utilising AI, big data and behavioural analytics? Will regulations stifle these innovations? What do you think?..