Automated Spreading
Product: Automated document recognition tool to scan tax forms and financial statements to translate the information
Platform: Web
Design Task: Reduce the time it takes to spread financial documents
Results: 75% reduction in time to spread financial documents
Learning point: Understanding the existing process is key to reimagining a new process
The Problem
When a financial institution wants to give a loan to a customer, they need to ensure that the customer can pay the loan back. There are many factors that go into the terms of a loan such as credit score. One of the main exercises performed by the institution when vetting a customer is financial analysis. Financial analysis is used to analyze whether a potential customer is stable or profitable enough to warrant a monetary investment. This means looking at tax statements, income statements, balance sheets, cash flows, etc. The pain point during this process is that the numbers in the financials exist in the document, not in the banking system. So, up to this point, users at the bank must read the numbers off the statement and manually type them into the system - a modern day user experience tragedy. This phenomenon is called a “stare and compare” - where the user stares at the document and compares it to what they just typed in. Many intern hours at a bank have gone to waste because banks pay people to merely type in numbers from a document…. until now! Using character recognition and machine learning, we can scan the document, pull out the relevant numbers and text, and allow the user to put the numbers in the correct places in the system. This is a great use of artificial intelligence, however, the computer cannot do EVERYTHING, we need to create a workflow and UI to allow the user to interact with this system!
This is an example of a cash flow statement from Amazon Inc. There are 3 years worth of data detailing the cash, expenses, and other accounting amounts. This is one page of a financial statement that a user would have to manually key in. For large corporations, these financials can be hundreds of pages. We need to help these users and allow them to focus on inputting the numbers to the right spot quickly. This is a paradigm shift from the focus being on inputting the numbers correctly. We can easily save user time and blow their minds.
The Process
For the process of this project, we need to understand what the user can do at each stage of the financial analysis of a document. There are a few concrete requirements of how the system functions that we need to work with.
The user must verify what type of page each page is in the financial document, this helps the character recognition understand how to read the page
The system makes an accurate guess of what that page is, but the user must confirm
The user must see the line item on the document and the number that is associated to it so they can have all the relevant information to assign it’s place in the system
The challenge behind this product is having the computer do all of the reading of text and numbers and some of the actual placing of the numbers in the right spots. That is the big piece we need the user to verify so we have to include confidence indicators to say “the computer is sure of this” or “the computer isn’t fully sure but htis is a good guess” and lastly to say “the computer could not identify what this is”. This is super important because we want to show the POWER of the artificial intelligence but still convey to the user that there is still some human interaction needed.
The secret with the machine learning aspect is that as the user assigns page types and categorizes numbers into places in the system, the automated spreading becomes more automated because it is learning from each user move. So when you spread your first financial statement, you may have to undertake a bit of leg work to assign pages and categorize numbers, but on your 10th, 20th, 100th document, the computer will have learned and the user interaction turns to just verifying what the computer has done. This saves time and allows the user to do what their job is - being a financial analyst, not a human keyboard.
This is a journey map that was made to talk about our user personas, our goals, and our questions. It also includes a journey map of the steps performed in automated spreading at each step of the way - from uploading the document to seeing the completed financial analysis in the system.
Additional brainstorming done on the possible user journeys. This includes pain points, where in the system the user could start this workflow, which users would prefer to start in different places, etc. Along with a pros and cons section to help weigh the design decisions.
The Solution
The automated spreading solution looks amazing! We are able to accomplish many things in just a few screens. Once a file is uploaded, it appears on the automated spreading dashboard. The user can interact with the document cards to view what files are eligible for autospreading, fix document errors if they occur, or spread the document. The computer will read in the document, guess what the page types are, guess where on the document the financial data is, and then finally guess what category the data should be mapped to. Incredibly efficient and a powerful tool for the financial analysts! The best part about this design is that we can see the document WHILE we are working with it - no need to open the PDF in another window or print it out! Let’s walk through some highlights.
This screen shows the automated spreading dashboard. This is where the user can go to see what documents are eligible for autospread. The user can filter by document type, hide documents, and delete documents. The user is able to spread a single document document or we have the multi-select ability to string together multiple documents and spread them as one. This is amazing because there are cases where multiple files make up one financial statement.
You can also see the different states of the document cards. There are cards that are ready for autospread - they have a blue button to jump into the autospread UI. There is an error card - this one shows that the password was not correct for a password protected document. There are other errors like file type, connection issues, etc. Lastly there is a document card with a loading state. In the product, this image is animated to spin to show processing power. The spinning image is the logo for the nCino IQ portfolio. Once the document is scanned (usually just a few seconds) the card will be ready for autospread!
This is the page identification screen where the user can verify all the data within the document. On the left side, there is a page summary which shows what the computer determined the page to be - we indicate confidence quickly with the green vs. yellow color. This way the user can quickly scan to see where they need to intervene.
Once they have chosen a page, the user can automap the values - where the computer automatically categorizes line items. They can change the page type if the computer was wrong or wasn’t able to determine what it was. Lastly, they can set the page format depending on how the statement appears, financial statements come in all shapes and sizes so the ability to set the page format is key to covering all the document use cases.
On the right side, the user can set the attributes for the financial period that they are attempting to enter into the system. Again, with all of this being in ONE interface with the document displayed, the user can be faster and more efficient at getting these financials into the system.
After the user completes the first step of verifying how the computer has read the document, the user can now focus on the financials within the document and making sure they are categorized correctly in the system. On this step, the page summary is abbreviated to allow more affordance for the financial mapping grid. The user can navigate thru the document by using that page panel, scrolling through, or arrowing through the document.
Displayed on the document we have MORE indicators. This lets the user know the status of each line item. The green checks means that the computer successfully categorized this line item. The blue “person” icon means that the user has manually made this categorization within the grid. Lastly, the yellow warning icon tells the user that this line item has not been mapped to a category. This allows the user to quickly know what has been done and what needs attention.
On the ride side, there is an excel like experience where the user can make sure that the document is accurately reflected. The user can make sure the line items, numbers, and categories fit to their liking and their financial mapping standards. This is another powerhouse page with a TON of functionality so that the user can do everything they need with this document in one stop, and be sure that the data is accurately transferred into the system.