Our capacity to collect data has outgrown our ability to make sense of it. Promises of a bright future under data science abound. Yet when it comes to efficiently cutting through the noise and generate value right here and now, we are stuck. The financial sector is still relying on manual labor to process and review data.
With any emerging technology — such as financial data aggregation and enrichment — defining the “how” is the tricky part. How do you extract value out of financial data? How do you shape your data strategy? Do you build or buy your data intelligence tools?
Harnessing that power of financial data is a big challenge. It is also an opportunity you can’t miss.
A consumer's transaction history holds data points on how much they earn and spend, when and where. It is a deep dataset that can tell their story for them; past, present and future. But its potential was mostly left untapped — until now.
Data has always been around in finance one way or another to help establish relationships and make decisions. Back when short-term credit extension was a common practice, merchants kept tabs on their customers’ purchases and repayments — a rudimentary form of credit risk analysis. This would eventually lead to the advent of credit reporting in the late 1800s, as agencies started to gather this information, sum it up in credit ratings and distribute it on a larger scale.
Over the past 150 years, financial data came to be at the core of every financial organization’s ability to conduct business. (As well as many organizations outside of finance.)
Transactional data refers to the information recorded from consumers financial transactions - money exchanges between us and the companies we work for or buy things from.
Up to twelve months of transactional data can be sourced directly from the bank, using consent-based data aggregation. This data provides the raw material to gain greater visibility around a consumer's financial position and behavior.
Financial data is deeply embedded in a wide range of data-oriented processes. It defines who we are as customers to the financial institutions we deal with. It shapes how we access credit. It facilitates KYC identity checks and ongoing fraud monitoring.
More importantly, financial data captures all the major milestones in our lives: first car or student loan payment, first job income, first baby, first pension payment. So much so that banks are said to hold the deepest and more personal dataset about our lives. Not Facebook, not Google — good old banks.
Much of this information is recorded in transactional data, which is easily accessible through open banking-type connectivity. Aggregating a consumer's transaction history across all financial institutions offers a unique opportunity to understand their financial profile and behavior. Now, even the smallest fintech startup can build products and processes on top of that data.
Up until now, transactional data has mostly been confined to personal finance management (PFM) features, offering end users visibility over their spending patterns, budgeting tools and such.
But its potential runs far, far deeper — transactional data transforms the way financial companies access, process and utilize data in their business operations. Credit risk analysis, income detection and fraud monitoring still rely heavily on bank and tax statements, credit scores and other credit bureau data. This usually requires time-starved customers to provide copies of their documents, which are then reviewed manually by analysts.
The whole process is slow and cumbersome. What’s worse, traditional data sources are showing limitations. Credit scores, while accurate enough in most cases, struggle with a number of edge cases. Among them, “credit invisible” consumers that have little or no credit history, or potentially good borrowers that cannot be statistically separated from poorer risks. Those consumers are left at the door.
As consumers’ financial profiles — their income sources, the way they manage their money and credit — is rapidly evolving, businesses are turning to transactional data to capture their whole story. Leveraging transactional data in their processes provides them with three decisive benefits:
Transactional data provides a deeper and more comprehensive understanding of a person’s financial position as well as consumer behavior, based on accurate, up to date information.
In just a few clicks, consumers can now share their financial data in digital format. This enables transactional data to be gathered across any number of financial institutions and processed at an extremely fast rate by machine learning and other advanced analytical techniques.
Actionable insights extracted from transactional data are a valuable tool to drive engagement. Understanding customers’ past and current behavior allows businesses to recommend exciting opportunities and help them navigate their future goals.
70% of consumers are willing to provide additional financial information to a lender if it increases their chance for approval or improves their interest rate for a mortgage or car loan.
Source: Experian consumer survey, March 2018.
The pressure is building up to adopt artificial intelligence and machine learning tools in order to leverage transactional data. Financial businesses are left with the age-old question: Should they build or buy? Should they develop their own data capabilities or turn to a tech partner?
The tool for the job is machine learning-powered data enrichment. Accurately making sense of raw transactional data requires specialized work and continuous commitment, which is one of the tools we offer at Flinks. A data categorization engine faces new data on a daily basis: new business names, new languages, and a growing number of categories. Data scientists can’t map out categories once and for all and be done with it.
Using machine learning techniques make sense for a never ending, iterative process. The model needs to be trained and constantly improved, which works best at scale: when machine learning algorithms have access to more data, they are able to learn more intelligently.
Raw transactional data is very messy and noisy. It’s millions of lines of abbreviated and sometimes cryptic transaction descriptions, amounts, and dates — most of which don’t seem to make much sense.
To put it in concrete terms, cleaning a transaction description
would turn something this
[XX012345-TMHRTNS-PYMTXX] into [TIM HORTONS].
Cleaning data makes it functional by removing unnecessary information. It transforms it into a format that data scientists can work with. This is a balancing act: remove too much information and a description can turn to an empty line; remove too little and your model is bombarded with useless noisy data.
Clean data is readable, but it doesn’t yet have any meaning. The next step is to categorize it, which is basically sorting data in buckets. Here, data scientists use a combination of techniques to understand exactly what a transaction is for.
The meaning of a transaction and ultimately the category it belongs to depend on the specific context of a use case. PFM apps, for instance, need to differentiate between a trip to the gas station to refuel a tank and a similar trip to buy beer. If you are performing fraud screening on loan applicants, not so much — you’d rather know if an account shows irregular activity.
A categorization engine is not a one-size fits all solution — it needs to be tailored to a precise use case.
Building data enrichment capabilities requires a dedicated data science team, time and resources. All of this heavy lifting — cleaning and categorizing data — turns raw transaction history into usable, enriched data. You still have to define and execute a plan to put it to work.
While it might make sense to build your own, your time and resources might be better invested elsewhere. More and more businesses turn to fintechs to integrate data enrichment products to their tech stack, so they can focus on building better processes, products and experiences.
Are you the best company to build this tool?
Are your internal analytical capabilities sufficient to turn the data into actionable insights?
Does building your data enrichment engine in-house gives you a specific competitive advantage?
If you are accessing transactional data through open banking-type connectivity, can you process potentially high volumes of heterogeneous and unstructured data?
Financial data aggregation — the ability to source and combine data across multiple banks — makes consumers’ transaction history and other banking information readily available.
New data enrichment tools enable these businesses to extract insights that can be acted upon. More and more are turning to fintech partners to prepare the data for them so they can focus on highest-value activities.
Most promising areas of impact are business processes involving data-oriented decision making, such as credit risk analysis, income detection and fraud prevention.
Now that we have covered the groundwork, let's dig into how you can concretely put transactional data to work.
Data becomes valuable when it fills information gaps and leads to more accurate and timely decisions. That’s why we created Attributes — a data tool that goes one step further by enabling you to manipulate large sets of clean and categorized data in order to extract actionable insights.
Advanced analytical techniques such as artificial intelligence and machine learning help sort and understand data, but they don’t blindly spit back insights that are relevant for you.
Don’t ask what data can do for you, that’s a common mistake. Instead, to make the most of transactional data’s potential you must first identify and have a clear understanding of your desired outcomes. This will allow you to determine what information is most relevant to reach your business goals.
What you want to achieve and how you intend to use data will inform how you should build your data enrichment capabilities.
Once you have a better understanding of what you need to know about your customers, Attributes processes their transaction history and extracts relevant, actionable insights. It sounds too simple to be true, but that is really the case. Let’s find out how.
Categorized transactional data is placed into a pandas DataFrame — a two-dimensional spreadsheet built for Python programmers. When data is structured, it is easy to manipulate. We group, filter and apply mathematical functions to generate aggregated outputs: data attributes.
This is where you are right now. You might already be using financial data to gather crucial information on your customers in order to feed your credit risk or behavioral models. Then the question becomes whether enriched transactional data can help your analysts work faster or empower your engineers to build better products.
This where you want to go. Your goal is to improve debt performance prediction? Then you need data to gain visibility on borrowers’ cashflow and spending habits, so you know they have the ability to pay.
Data attributes are like building blocks that you can stack together.
Each data attribute reveals an aspect of your clients’ financial profile and consumer behaviour. Total income, average monthly balance, average monthly cash flow, number of NSF fees, number of days with a negative balance. The list goes on and on (and on).
You can put them side by side and create an accurate and holistic view of your clients. You can also combine them together to analyze trends and extract very specific insights.
Attributes is not a black box — in fact, it’s quite the opposite. It lets you define the few things that are really important to you, and then enables you to build your own custom data enrichment tool.
A data enrichment tool is merely a tool — if you understand the job you need to get done, it will help you do it better and faster. Where you had an axe, you now hold a chainsaw. You are still in charge of planning and executing your business strategy.
Transactional data is already part of various workflows in finance, from income verification to credit risk analysis. But manually processing and reviewing data takes time and might introduce errors. Attributes sources transactional data directly from the banks, enabling fully digital workflows.
Data enrichment is not a substitute for human judgement. Attributes’ enrichment process improves the quality of data and makes it ready to use in your specific context. You can focus on what you do best: building your models and handling operations that are too complex or require a human touch.
With Attributes, you can start small to learn from narrow experiments in data enrichment — and apply these learnings to solve other problems. No human is good at biking on day one. When you get the hang of this new technology, then you can start reshaping your business strategy and building better products using Attributes at its full potential.
By using data enrichment tools, financial organizations can capitalize on opportunities to improve their data-oriented processes.
Emerging data enrichment tools don’t necessarily require you to build entirely new processes around them. Our own data enrichment tool, Attributes, has been designed in part to accelerate business operations by digitizing and automating repetitive tasks — those usually done through manual labor, such as collecting and processing data. Processes relying heavily on data for decision-making are especially ripe for this type of incremental innovation.
Below are some of the most effective use cases where Attributes has been deployed to create meaningful outcomes.
Streamline your credit risk analysis
Among the most important application of data enrichment is identifying the creditworthiness of potential customers. Gathering information often involves credit reporting agencies and documents that the potential clients must provide themselves.
Data attributes provide an up-to-date image of a customer’s financial position, capturing changes that might not yet be reflected in their traditional credit score. Its fully digital workflow makes it easy to feed existing models for instant decisions or equip credit risk analysts with a holistic financial profile of their client.
New Canadians represent a growing but underserved market of people with diverse financial profiles. Find out how Attributes enables you to gain risk related information where none is available.
As credit risk analysts start working with a data enrichment tool that does all the heavy lifting for them, their role will shift. It will become very important to improve their analytical skills so they are able to leverage insights efficiently.
Average Monthly Employer Income$5,000
Sum of Employer Income (Total)$50,000
Count of Employer Income Deposits10
Trend of Employer Income100%
Average Monthly Non-Employer Income$500
Trend of Non-Employer Income100%
Average Monthly Free Cash Flow(+) $500
Number of Days with a Negative Balance5
Average Amount of Monthly Loan Payments1
Average Amount of Monthly Bill Payments$350
Minimum Balance(-) $1,200
Recent Activity Analysis120%
Count of NSF Fees1
Age of the Account201 days
Count of Stop-Payment Fees
Average Monthly Micro-Loan Payments$250
Fraud management is an enormously important area for financial organizations, with impacts ranging from physical and digital security up to regulatory compliance and trustworthiness. It is a major cost centre for most institutions, so minimizing the costs of maintaining efficient processes is becoming a key focus.
To fight fraud, organizations rely heavily on data in Know Your Customer or monitoring processes. It can lead to labor-intensive and time-consuming operations, especially if manual work is involved. But once fraud risks have been identified and prioritized, data enrichment becomes the perfect tool to automate fraud detection.
Attributes allows businesses to analyze their customers’ transaction history to automatically detect irregular behaviour and fraud patterns.
Fully digital workflows make it easier to store data. Keep the data attributes securely stored for future reference. You can also aggregate that data to gain visibility on whole segments of your customers or perform other analyses.
Automating fraud detection doesn’t mean taking your hands off the wheel. Attributes returns both account-level and transaction-level information, increasing the chance of detecting fraud while limiting false positives.
Using data attributes as a flag system to understand your customers’ activities, you remain in charge of deciding what counts as regular and irregular. Your risk experts can focus their work on the accounts that fall below this threshold for more efficient fraud investigations.
Account Age160 days
Count of Active Days100 days
Unusual Recent Account Activity175%
Unusual Recent Debit Activity250%
Unusual Recent Credit Activity50%
Sum of Recent Loan Deposits$0
Average Monthly Recurring Payments$0
Average Monthly Employer Income$500
Recent Balance Trend25%
Recent to Historical Usage Ratio200%
Verifying a customer’s real income is remaining one of the biggest challenges for banks and lenders. This is undermining business, since income is often one of the determining criteria to offer a loan, besides credit score. Customers are still being asked to input information and provide documents — which might result in incomplete data, bad customer experience or cumbersome manual processes. Without a digital workflow and proper data tools to understand a customer’s current income streams, decisions are hardly data-driven.
Attributes has been designed to differentiate employment and non employment income and outline trends in both, allowing you to understand their current financial situation and track their trajectory.
A major role of income detection is to help assess a customer’s affordability and ability to repay. Attributes also returns a series of data attributes giving visibility over the other loans a customer has already incurred as well as their free cash flow.
The information gathered through income detection can be used to reveal trends in a customer’s spending patterns. Paying close attention to those trends allow you to detect significant changes in life events and take action sooner.
2 Months Ago$2,000
3 Months Ago$1,850
2 Months Ago3
3 Months Ago4
Trend of Employer Income75%
Average Monthly Non-Employer Income$2,000
Average Monthly Government Income$1,000
2 Months Ago$1,000
3 Months Ago$1,000
Average Monthly Total Free Cash Flow(-) $500
2 Months Ago$6,000
3 Months Ago$3,500
Trend of Total Deposits80%