Artificial Intelligence in Fintech

The Challenge

In the mergers & acquisitions business, investment banks need to upload and share highly sensitive information with potential investors quickly and accurately.

Depending on the phase of the deal, specific information inside of these documents needs to be redacted. Traditionally this has been done using sharpie markers and paper, however, this process is time-consuming and is prone to human error.

Intralinks is in the process of building an application that utilizes machine learning to predict specific words, phrases, or numbers that need to be redacted. Given how risk-averse our banking customers are, the biggest challenge is building trust that the predictions are accurate and appropriate.

I served as the lead researcher for both the early stage product discovery and initial evaluative studies on the product MVP. This product is set to release a beta in November 2020, and I’m serving as the research lead on the early adoption feedback program.

My Project Contributions

Research Plan & Protocol
0%
Survey Design & Analysis
0%
Study Moderator
0%
Analysis & Report
0%
Can I trust that Intralinks redacted all the correct information from these highly confidential documents?
The product survey was built and distributed in Qualtrics.

Product Market Fit

Before engaging the engineering to build an automated content redaction tool, we reached out to clients to better understand their current process. After surveying 100 Intralinks customers across the globe, we learned:

  • 53% of customers use manual online tools such as Adobe Acrobat, followed by pen and paper (18%). This indicated a market opportunity to move redaction capabilities into the data room.
  • Predicting sensitive data was required for 33% of our participants. The approval workflow is more sought after feature (59%).
  • Our hypothesis for what customers find the most important did not match the survey results.

Understanding the Customer Journey

After gaining high-level insights into what tools are used to redact content today, the team was eager to develop a stronger understanding of the end to end deal preparation process.

In a series of six customer interviews, we collected multiple data points to create a journey map. Key points included:

  • Most participants redact content for competition reasons. Solving the personally identifiable information (PII) use case will not solve all of the customer’s needs
  • Their current process involves a document approval process between the junior legal analyst and partner before documents are uploaded into a data room
  • Some customers expressed a desire to save multiple versions of redacted documents. As a deal progresses and final buyers are close to being selected, more sensitive information can be requested.
Customer conversations impacted changes to the journey map created by the design team.
Sample Rainbow chart created in Microsoft Excel

Data collection is a team sport

Intralinks has a small team of researchers. We practice a democratized research model. Every project encourages full team collaboration, ensuring that key findings make their way into the product development lifecycle.

To improve efficiency and gain alignment on collective findings, I ran team debriefs after each session. We used a rainbow chart to track trends in data points across all six customer interviews.

Key Design Finding

Participants expected to be able to apply changes to content redaction directly onto the document they were reviewing. The interface that was tested required the user to open a modal to make changes across all documents.

Recommendation

Consider providing customers with the ability to edit content types in the context of where the review is taking place.

Improvements in Action

Due to technical constraints, the team was not able to allow bulk editing directly on the document. However, based on the findings from MVP validation, the designer quickly pivoted and embedded contextual help to guide the analyst to the edit feature.

We will continue to monitor this finder during early adoption to determine if improvements to this flow need to be moved higher up the backlog.

Study Materials

Full Results

Machine Learning Concerns

When it comes to embedding artificial intelligence into financial services products, it is critical to take customers’ apprehension about automating highly sensitive tasks into consideration. Customers will gain confidence in your product over time as the accuracy of results improves.

Even if there is a tool that allows us to redact automatically, I’ll still need to check everything manually