Your payment recovery strategy needs a boost from machine learning

Failed payments are a revenue drain that can grow into a bigger headache than you think.

The core problem is you’re trying to fit a square peg through a round hole. Most payment service providers (PSPs) use a brute force retry strategy. But that battering ram won’t solve payment failures issues. In fact, it can make retries less likely to succeed, which inevitably increases customer churn. 

Failed payments need a more strategic approach. Machine learning (ML) – more specifically, Butter’s advanced framework – is designed to analyze and learn from every bit of transaction data. ML helps strengthen a payment recovery strategy, and Butter’s dynamic approach is a great fit for any merchant.

‍

Failed payments are a complex problem for merchants

Lost revenue attributed to payment failure is often 25%+ of a merchant’s ARR. The most significant contributors to payment failure are “soft errors.” These errors often result from issues with PSPs or card issuers, like insufficient funds or suspected fraud.

In most cases, failed payments are retried on a hard-coded interval - in practice this looks something like retrying every failed payment every few days. But because payments fail for different reasons and return different error codes, sticking with a single approach can limit your performance. 

For instance, if you only retry payments every Monday at 11:30 a.m., you may recover some payments, but the odds are you’ll miss a lot more. That’s because the retry doesn’t address the reasons the payment failed in the first place.

Let’s break down two examples of soft errors to see how this applies in practice:

1. A payment fails due to insufficient funds. If a customer doesn’t have funds in their account, retrying immediately may have a high likelihood of failure. Spreading out your retry attempts gives more time for funds to hit the customer’s account, assuring a successful recovery.
2. The customer’s bank has a processing error. If a customer is making an international purchase, their bank may flag it as possible fraud. Retrying during business hours in the customer’s time zone can make the payment more likely to go through.

Even if you are aware of how these failed payments affect your financial health, you need the right resources to handle them. Most PSPs aren't equipped to address these challenges. 

‍

Machine learning models simplify payment recovery

Every payment is unique and requires a unique recovery strategy. Leveraging machine learning delivers superior results to a repetitive approach since there are many complex factors that impact whether or not a payment will be successful, including:

• Transaction amount
• Card type (debit, credit)
• Card issuer (MasterCard, Visa)
• Payment origin (domestic or international)

Unlike manual or rules-based processes, which rely on predefined conditions and fixed sequences for recovery attempts, ML models learn with each transaction. As the model ingests more data, it understands more about your business, increasing the likelihood that each subsequent transaction will succeed.

A payment recovery strategy backed by machine learning recovers more failed payments by analyzing 100+ data points. The results?

• More recovered revenue
• Lower customer churn
• Increased topline ARR
• Improved TAR

However, not all machine learning models are the same. Many models are “one size fits all” – applying the same tactics to each merchant’s strategy. But every company differs, from customer demographics and AOV to payment origin and billing cycles. Effective machine learning models use data that is unique to the merchant, industry, and circumstances of each transaction. That’s what Butter does — and you get the ideal results for your organization. 

‍

Butter stands out from other ML solutions with dynamic learning and experimentation

With Butter’s sophisticated approach to machine learning, context matters. Our strategy combines global and merchant-specific models to capture commonalities across industries as well as cater to the unique needs of individual businesses. 

Another critical differentiation between Butter and other ML solutions is the rate of experimentation for customers. With Butter, each client has multiple models running concurrently - these models undergo constant analysis for what works and doesn’t, leading to a better iteration in the future.

Unlike a static ML model, Butter combines dynamic learning and a team of experts, making each model iteration better than the last. Our data scientists and engineers have poured years into building Butter from the ground up – leading to hundreds of millions of data points gathered to date.

Our constantly expanding data set means retries are optimized throughout the dunning cycle to recover more payments and prevent customer churn. As a result of these performance increases, Butter customers can see 5+% increases to top-line ARR. 

No matter how small or large your business is, Butter can equip you with the tools to improve your TAR and prevent accidental churn. Contact us today to get started with a free payment health assessment!

‍