Ian Harding - Featured - Cash Flow

Case Study – Cash Flow

Role: Director of Design and UX

Building Better Businesses Proof of Concept

This project was a Proof of Concept (POC) to see if we could improve Canadian small and midsize business (SMB) loan application success rates by improving business owners’ knowledge and awareness of their finances and understanding the importance of maintaining a healthy cash flow.

The problem was first brought to our attention by subject matter experts (SME) who had recognized that when business owners entered into their third and fourth year of operation, many were struggling to obtain new loans or loan increases because of the total debt exposure they had previously accumulated – their risk profiles were now too high.

This meant that rejections to seemingly small loan requests were causing CSAT scores to decline and in many cases, clients were closing accounts altogether.

How might we

Deliver a simple and intuitive tool for business owners to better understand the health of their finances in order to make informed loan decisions?

My team had 8 weeks to come up with a solution. We divided the project into 4 two-week sprints and coordinated tasks using Jira. During our sprints, we conducted stakeholder discovery sessions, ideation jam sessions, and generated ideas using AI analysis of current customer data.

We also conducted user testing to obtain data that would inform our interface designs and spoke with real business owners to approve, disprove, and discover new ideas and solutions.

Guided by our SME stakeholders, the team generated more than a dozen solutions against our primary objectives:

  1. Support customers through the development of a comprehensive financial model
  2. Visualize transaction data and see the health of their cash flow
  3. Allow customers to generate accurate loan application documentation

Sprint 1

Research & Discovery

In our initial research with stakeholders and data/analytics teams, we learned that business owners were providing poor or insufficient financial data with their loan applications. Why was the provided data poor and what were people doing wrong?

Did they not understand what they were being asked to provide?

Some first round solutions gave business owners an aggregated view of their financial landscape by allowing them to connect their financial accounts into a single view. Our solution would then use artificial intelligence to extract relevant data and apply it to loan application documentation for submission.

Using Flinks as a technology partner, the team explored connecting bank accounts for data ingestion, data extraction from documentation (uploaded either by file or photo capture) and using transcription data captured directly from human voice inputs.

During our interview sessions, we confirmed that business owners had poor financial literacy and didn’t fully understand what financial models were. They were too focused on other areas of their businesses and relied heavily on the support of friends and advisors to produce the documentation banks required.

One key discovery was that most were very tech savvy and utilized either software or AI to support their business’ accounting needs.

“I regularly use ChatGPT to see how I’m doing… I’m not afraid of giving it my personal information because I know my personal information is already out there, so I might as well use the tools that help me.”—Female, 36

When the interviews had finished and the feedback was analyzed, we concluded that our initial ideas were too complex for a majority of our defined audience and that a lack of financial literacy was the real problem. When we dug into which areas of the financial model were the most problematic, it was determined that cash flow forecasting was the primary culprit of loan rejections.

This outcome highlights the importance of regularly seeking feedback and asking questions. By producing paper prototypes and quickly reviewing them with stakeholders, we were able to produce a solution that best aligned with the problem.

Sprint 2

Definition & Structure

Our product requirements document (PRD) for this project broke down the solution into two primary scope categories consisting of  ‘must haves’ and ‘nice to haves’.

The must have functionality for our MVP included connecting bank accounts, viewing connected accounts, viewing transaction data, viewing transaction categories, document upload, and document export. Nice to have features included loan insights, grant offerings, and advisor booking.

The team explored using generative AI tools to produce early-stage visual wireframes and rapid prototypes. Using AI to generate these artifacts was exciting to test and although AI can be a great tool for producing text-based outputs such as personas, user journeys and page content, it still has a long way to go when it comes to creating visual outputs that satisfy the needs of today’s product designers.

Shoehorning AI chat interfaces into traditional design interfaces is a disjointed experience that doesn’t align with how humans solve problems. Only the products that are able to seamlessly integrate AI solutions into existing workflows will be the ones that lead the path forward to producing high quality artifacts at speed.

Sprint 3

Testing & Validation

In a fast-paced environment like a startup, teams will often omit user testing due to having higher priority objectives, tight deadlines, or simply because they don’t think about it (or care).

Regardless of the team size or skill sets, user testing should be performed at every stage of a project. 

Tests can be performed quickly and easily using a small number of participants, often friends and family, and can provide significant value and insight into ensuring solutions are understood and usable.

For our first round of testing, participants were asked to evaluate various designs and rate them against content hierarchy, language and messaging, interface functionality, and overall complexity.

Important note: We recognized that the data being displayed was merely a snapshot of a singular moment in time and required historical data to perform effectively. Since it did not take into consideration extraneous values and understanding that expenses are typically withdrawn early in the month whereas income typically comes in near the end, it was important that our use of AI accounted for this and supported the output.

For example, you could have 4-5 weeks of negative cash flow, but in the last couple of weeks you get an influx of income that suddenly swings the cash flow into a “healthy” state, even though the trend is contrary. So when viewing mid-way through the month, it will look unhealthy until the month is finalized and income balances it out to be healthy.

Audience criteria:

  • Canadian small business owners (1-50 employees)
  • Had a loan rejection in the past two years (<$50,000)
  • Actively used accounting software (QuickBooks, Xero, etc.)
  • Deemed themselves tech savvy, but not financially literate

Test 1

Evaluate participant’s understanding of the content, language, data, and overall complexity.

Assumptions:

  • Background colours for income/expenses helped people better associate the data within the chart
  • Health callout gives a target to works towards
  • Time frame filter was easily recognized
  • Chart displayed just the right amount of data
  • Less is more (minimal labels and support)
  • Users understood what an expense ratio was

Results

  • Users appreciate the visual connection using colour
  • Having the advisor callout directly connected to the health was positively received, but the advisor support question felt inappropriate due to users already using the app and feeling it was important
  • Users misinterpreted time frame filter as being “millions” of dollar)
  • Users wanted more data (ie. see weekly values by tapping bars)
  • Users desired supporting headlines such as “cash flow health”
  • Users assumed that the higher the expense ratio number, the better

“Without labels, I can’t tell what I’m looking at. Sometimes having fewer elements on the screen doesn’t make it simpler, it makes it harder for me to understand.”—Female, 39

Test 2

Evaluate participant’s preference for layout, elements, colour, and iconography.

Assumptions:

  • Balance sheet layout would be easier to interpret
  • Use bright colours to evoke emotion and prompt action (if required)
  • Condensed time frame filter was more of a setting than an active tool
  • Removing all extra content would make the data easier to read
  • Top-down hierarchy reduces cognitive load

Results:

  • Simple math equation was preferred over expense ratio
  • User of colour help direct action and evoke emotion (ie. green felt positive and encouraging)
  • Drop-down menu was simpler and more intuitive. Users wanted to “set it and forget it”
  • The larger the chart, the better
  • Top-to-bottom was preferred over left-to-right eye scanning

“I guess an expense ratio makes sense, but do the math for me so I don’t have to. I want to know right away what the result is.”—Female,36

Sprint 4

Conclusion

Sadly, not every project gets to see the finish line.

Due to an unexpected organizational restructure, this particular project was not completed during my tenure. However, the team saw our product idea as valuable and impactful for Canadian SMBs and were fully invested in the process and delivery every step of the way.

We were proud of what we were able to accomplish in such a short time and felt positive about the results, especially knowing that future projects would benefit from the utilization and application of our experiences, discoveries and lessons learned.

All designs outlined above are the exclusive intellectual property of ATB Financial and may not be saved, copied, shared, or reproduced in any form.