The Truth Behind Merchant Underwriting Automation Failure


July 3, 2018


Real-time Approvals for the Modern Merchant

Offering same-day merchant approvals, or even doing so within minutes, has become table stakes in merchant acquiring. Whether you’re a highly digital technology company or the most traditional acquiring bank, providing incoming merchants with an answer on whether their application was approved or not has to happen between minutes or a day to be competitive in today’s digital world.

Merchants now expect real-time approvals, just as they expect instant payments and the ability to see their merchant dashboard online anytime. Without it, merchants see the acquirer as outdated and unable to deal with them as instantly as their business now requires. Besides, every other payment facilitator, ISO, or acquirer can approve them and get them set up and processing payments by tomorrow. So why would they wait?

Automation for Faster Underwriting

To get faster and satisfy this need, many are turning to the slowest part of the entire merchant application and boarding process for efficiency improvements – underwriting. Underwriting automation has become a common solution to this problem, but it can come with issues of its own. Automating the underwriting process can speed through some of the work traditionally executed by diligent underwriting staff, losing details in the process. The concern, of course, is that automation won’t be able to handle the complexity of the underwriting process, decision-making and exceptions.

There is a way to automate underwriting without increasing the chance of accidentally accepting merchants beyond the organization’s risk tolerance, or rejecting acceptable ones. Many haven’t achieved this because their risk model isn’t suitably prepared for automation. The core, common issue lies in risk scoring.

The Pitfalls of Binary Risk Scoring

Most companies have complex ways of scoring merchant risk, but it isn’t always congruent even within one underwriting department. After the underwriter has assessed each aspect of the merchant, they usually assign them as ‘accepted’ or ‘declined’ – a binary decision with only two options. If it later proves that the merchant was a poor choice to accept, their file gets opened and and underwriter must pour through all the decisions that were made up until the point when they were accepted to see what went wrong.

Translating this approach, a binary ‘yes or no’ decision, is easy for a software system to adopt. Boolean logic (accepted = 1, rejected = 0) is relatively simple to set up for automation. But there are challenges with this solution as well:

  • Gathering underwriting information still takes a long time. Underwriters must still obtain all the KYC, AML, and other information necessary for underwriting. It takes a long time to do so, but these are necessary inputs for the automation to occur.
  • Binary scoring can lead to mistakes and increased risk. Having such a hard line on whether the merchant is accepted (1) or rejected (0) can lead to risky merchants accidentally being accepted or non-risky merchants accidentally being rejected. This is because there are no weights to the information. If there are lots of acceptable data points about the merchant and only one negative, the system will accept the merchant. A human underwriter, however, might have identified that one negative data point as a clear indication of fraud and chosen to reject the merchant.
  • Binary scoring restricts business insights and growth. Retrospectively viewing underwriting data from a binary scoring and decision model shows a muddy picture. It is clear how many merchants were accepted and rejected, and how many of them were successful or not, but it is not clear why some accepted merchants were unsuccessful or whether there is growth opportunity in previously rejected merchants. To decide whether to revisit previously rejected merchants that might be a good fit now, each application would need to be reviewed all over again.

Revolutionized Risk Scoring

  • So how can underwriting automation be successful? It should be able to:
  • Execute with the complexities of your risk model
  • Flag individual data points
  • Weight some data points more heavily than others in its decision
  • Provide valuable business insights
  • Show you opportunities for growth

To do so, take a scientific approach. We recommend scoring each data point in the underwriting process from 0-100, not 0-1. This gives a better picture of just how risky each merchant is and why. Providing weights on each data point indicates which is more important than others in your decision. Finally, learn over time. Don’t set up automation and expect it to work perfectly from day one. Just like a new team member, give the system a small percentage of applications and get an underwriter to review its work and recommended decisions before allowing the decisions to be implemented. You will learn how to tweak your model to so the system can give you the results you want before allowing it to automatically accept or decline applications.

Fully functioning underwriting automation shouldn’t be possible from day one, unless your risk model is already set up for it. However, risk models and scoring are typically set up for people to use, not machines.


Approving merchants in real-time is a competitive necessity in merchant acquiring today. Underwriting automation is the solution, but few underwriting teams are set up for automation success. Before embarking on this endeavor, reevaluate and revolutionize your risk scoring model to increase complexity, visibility, and correct decisions both now, and after implementing an automated system. Getting faster at underwriting shouldn’t automatically increase risk assumption.

Download the discussion paper to learn more about how to set up your risk model for success.