# Overview of how AI can be used in Open Finance
source: https://developer.mastercard.com/blog/ai-in-of-overview/index.md

Authors: Natesh Babu Arunachalam. 9 out of 10 US consumers are currently using technology to manage their finances. This could be a mobile app offered by their bank to pay bills, budgeting apps that import their transactions from the consumers bank, or even an online brokerage account to handle their investments. Linking your bank accounts to financial services is made easy due to the rise of open finance which in turn generates billions of transactions per day and this data can be used to train AI models to generate valuable insights for consumers. What is Open Finance? Open Finance is a broad term that is used to describe the practice of allowing consumers to share their banking transaction data with third parties who are providing some value-added service such as an expense tracker app. This sharing of data is done using secure APIs. For example, imagine a consumer wants to understand their spending habits and signs up for an expense tracker app. The consumer could input their transactions manually, or they could link their accounts with the app using an open finance API provider like Mastercard Open Finance (US) API. Core techniques that underpin the models Named Entity Recognition Named Entity Recognition (NER) is a subset of Natural Language Processing (NLP) that aims to extract relevant information from text data. Since all Open Finance transaction data records include text, this is an invaluable Machine Learning technique in Open Finance. NER is often used to extract and analyze a consumer's bank transaction data. If you look at your most recent transaction it would look something like this: May 16, 2023 -- Uber Trip - $12.25. Please note this is just an example and all banks display transactions differently. The text of "Uber Trip" in this transaction is the transaction description. What NER can do is look at a consumers transaction data in text form and identify the merchants so this information can be used to enable intelligent decisioning. For example: May 15th, 2023 -- Acme Payroll - $1,500 May 16th, 2023 -- Uber Trip - $12.25 May 17th, 2023 -- Chipotle - $6.19 May 18th -- Internet Bill - $100.00 NER can take the relevant information in this blob of text and organize it as such: Named entity recognition is useful in Open Finance for extracting and organizing text-based transaction data. Transaction Categorization Transaction Categorization is a text classification issue, and Machine Learning is famous for solving classification problems! NLP is the technique that powers transaction categorization, each transaction description text can be used to identify what type of transaction it is. For example, the transaction description of "Uber Trip" can be classified as a "Transportation" expense. Look at the table below for some examples: The machine learning techniques used to solve this problem would include Natural Language Processing (NLP) based classification models. These techniques would be used to take the transaction description text and process it so it can be classified as one of the defined categories. This is not a trivial problem to solve due to the amount of variation that can exist in the text. For example, the NLP engine would need to take the text "SAINSBURY SMKT" and understand that SMKT is referring to a supermarket. At the same time, this text could have been "SAINSBURY SUPERMARKET" or "SAINSBURY SUPERMK". Once NLP can parse and understand the transaction description text, an NLP based large language model is used to map a transaction against the correct category. Transaction categorization can be used for: Budgeting apps where the user would want to track their expenses in specific categories; Commercial travel and expenses (T\&E) app that automatically categorizes expenditures. Use Cases of AI in Open Finance Smarter Payments Most US consumers will have multiple monthly bills to pay with different due dates. Even the most on top of their finance's consumers can forget to pay a bill every now and then, which can rack up late fees. Another situation consumers can run into is paying a bill at the wrong time of month, which can result in an overdraft. Smarter payment systems can analyze a consumers cashflow, spending, and bills to figure out which bills to pay and when during the month. For example, assume a consumer gets paid on the 1st and 15th of a month. The paycheck that lands on the 1st will need to cover expenses from the 1st to the 15th. Assume the consumer gets paid $1,500 and must immediately pay $1,000 for rent on the 1st. The consumer would be left with $500. Now with that $500, look at the following bills schedule. To avoid late fees, a smarter payment platform would analyze the consumers cashflow and understand that on the 1st and 15th the consumer is paid. A Risk classifier built using ensemble tree based models is used to give a confidence score to figure out what is the best date to pay a bill. Merchants who have customers with autopay setup can benefit from a smarter payment system as well. If a merchant is about to execute an autopay via a pull payment, it can be beneficial to check if it is a good time to make the pull payment request. If a transaction fails due to the customer having insufficient funds, this could cost the merchant fees and increase operating costs since the failed transaction might require support. A smarter payment system would allow the merchant to get a confidence score powered by a risk-model classifier to know if it would be better to wait a few days before requesting the pull payment. Smarter Payments can be useful for: Applications that help consumers figure out the most optimal order to deploy their capital to their bills; Allow merchants to know what the optimal day of the month is to request a pull payment from their customers to avoid failed transactions due to insufficient funds. Cashflow Underwriting About 25% of us consumers have a credit score of fair or poor. These can include individuals that have recently declared bankruptcy or are new to the US. There are also individuals who do not have a credit score since they never opened a line of credit before. A low or non-existent credit score does not necessarily mean that a consumer will not pay back their line of credit if extended one. Cashflow underwriting can be a great alternative way to measure someone's creditworthiness. By looking at a consumer's bank transactions for a given period, you can get a full picture of their income, spending, and expenses. This will give a holistic picture of the consumers' finances vs just looking at their credit score. For example, a consumer who is taking in $3,000 a month, and spending $2,900 in a month might be paying all their lines of credit on time. However, they can be a risky borrower since their disposable income is low, and there is a risk of overspending. This can be done using similar risk classifier techniques that are used to power smarter payments. Cashflow underwriting can be useful to: Analyze the creditworthiness of consumers with low or non-existing credit scores; Check if the customer is spending on goods and services that make them a riskier borrower, such as gambling. Open Finance (US) and AI Mastercard has a suite of Open Finance (US) APIs that are powered by AI and Machine Learning. Check back here soon to see the next in our series on AI in Open Finance where we talk about the challenges of training models with Open Finance data. Although AI is beginning to make headlines, Mastercard is not new to the field. In fact, Mastercard has been using AI for the better part of a decade and AI is embedded into a whole range of Mastercard's products. Mastercard is committed to the responsible use of AI in our products and will remain dedicated to customer data privacy.