Credit data powers Experian’s software transformation – News

In the course of nearly 30 years of collecting data about consumer spending patterns, Experian plc has amassed detailed information on more than 1 billion people and 25 million U.S. businesses. As one of the “Big Three” credit reporting bureaus, along with TransUnion LLC and Equifax Inc., it collects the information that’s vital for financial institutions making decisions about whom to lend to and how much.

When big-data analytics took the financial world by storm a few years ago, Experian realized it was sitting on a gold mine. In 2019, it began engineering a transformative expansion into software and data analysis, creating a business that now brings in 35% of the company’s $7 billion in annual revenues.

Software is now at the heart of Experian’s growth strategy. “I joined 10 years ago, and the company now is almost unrecognizable,” said Alex Lintner, chief executive officer for Experian Software and Technology.

At the core of the new business is Ascend Platform, an artificial intelligence-driven suite introduced in 2023 that integrates data, analytics, software development tools, fraud prevention, loan analysis and other services. It’s a commercial version of the analysis Experian has been doing for decades.

Currently focused on financial services, healthcare, automotive and digital marketing, Ascend automates many of the manual data science tasks that have been standard in those industries for years. It’s used by 80 large organizations in the U.S. and 8,000 data scientists. Customers and partners have built more than 1,500 packaged solutions on top of it.

Risk balance

Ascend addresses two major needs that are common to financial institutions and purveyors of big-ticket products and services. “They want a model that says what kind of customers they want to gain and how to reduce fraud risk,” Lintner said. “Building the model and going from the analytical to the production environment has been a laborious, time-consuming, tedious and error-prone process. We have found ways to automate that.”

The Big Three credit bureaus are best known for their rating scores that assess a consumer’s creditworthiness, but evaluating the likelihood that a person will pay a loan on time is far more nuanced than a single score.

Patients in a hospital or people buying a car usually pay the bill over time. About 75% of U.S. car buyers lease or finance their vehicles, Lintner said. The likelihood that a person is a good prospect or risk involves complex demographic factors as well as their car-buying habits, credit history and lifestyle characteristics.

Marketers use data from multiple sources to target prospective car buyers with offers while automotive financing firms consider behavioral factors when deciding the terms of a loan. Experian has assembled 200 attributes, or behavioral profiles of buyers. Its customers combine that data with external and internal information sources to build models that guide their decisions. All data is anonymized to comply with regulations.

Automating the grunt work

Experian has automated much of the grunt work of whipping data into shape to make it analytics-worthy. “We cleanse the data and to make sure it’s complete, accurate, in the right formats and fresh,” Lintner said. “Then we have an analytical sandbox environment where data scientists can run their Python, R or SAS code to build a model.”

Experian Assistant, a generative AI tool the company announced last October, aids in coding, enhances model transparency and helps data scientists cycle through multiple iterations quickly with advice for optimal coding and deployment. It recently won the 2025 FinTech Breakthrough Award for Analytics Innovation.

“It doesn’t replace a data scientist; it’s like having an Experian consultant sitting next to you,” said Keith Little, president of Experian Software Solutions. “You can ask questions about the data, and it helps you with code using gen AI that we’ve trained with years of experience.”

Little said Experian initially expected the assistant would improve coding productivity by about 30%, but “we’ve seen clients that are achieving up to 60% gains.”

Fixing model drift

Gen AI also helps after models move into production. Model “drift” is a common problem in which a machine learning model’s performance worsens over time as real-world data diverges from that used in training. Detecting drift has traditionally been difficult, since changes happen gradually and can be hard to spot.

Experian has automated that process, Lintner said. “When a model drifts, we create an alert using gen AI that identifies the root cause and makes suggestions for changes that can restore the model to its effectiveness when it was originally built,” he said.

A fraud sandbox aggregates billions of events worldwide and displays them in a heat map correlated with triggers like email offers. “You can see that you’ve perhaps accepted emails that have been associated with fraud in other organizations, so you should probably not do business with that customer,” Little said.

Challenger modeling, a technique used to test alternative machine learning models against a model in production, has cut the time to identify fraudulent patterns from weeks to a few days, Little said.

Terms of rejection

Another feature of the platform is “reject inferencing,” a credit-scoring technique that estimates the creditworthiness of applicants previously rejected for a loan. Because traditional models only gather data from accepted applicants, reject inferencing tries to reduce bias by inferring what might have happened if the rejected applicants had been approved.

The discipline also looks at customers who rejected a loan offer and later accepted one from another source. “Perhaps they were offered a loan with an eighth of a percent lower interest rate,” Lintner said, “so maybe you should think about the terms that you offer or perhaps your fraud guidelines are too tight.”

In a highly regulated industry, gen AI also helps generate the mountains of paperwork needed for compliance.

“It can take months or years to get models into production; a lot of the reason is compliance,” Little said. Experian trained its gen AI model on the SR 11-7 Guidance on Model Risk Management required by the Federal Reserve and the U.S. Treasury Department’s Office of the Comptroller of the Currency. “We can also train the model on the client’s policies, so the platform will generate the paperwork along with your models to satisfy the OCC,” Little said.

With a platform that combines collaboration, an integrated set of tools and a common user interface, Experian hopes to become the Microsoft Office of credit analytics, Lintner said. “Our intent is to do for the industries we serve what Microsoft has done for the PC,” he said.

Photo: Experian