A Portland, Ore.-area firm with data on tens of millions of credit and debit card transactions flowing into its system each day published a new report this week that reveals a glimpse of how the COVID-19 pandemic is changing U.S. consumer spending.
Spending on beer, wine and liquor rose 60% the week of March 16 over the same week in 2019. In Washington, the average transaction size at Costco was up 16% across all demographics and 35% among baby boomers as the outbreak intensified during the week of March 9.
And despite all the news about toilet paper and sanitizer hoarding, there was another product category where spending really spiked: booze (see above).
Other data points: video game spend is up, while travel and lodging are down due to the social distancing measures put in place across the country.
Founded in 2010, Facteus just switched its name from ARM Insight. The name change reflects a gradual evolution for the firm, which has begun to focus on the emerging realm of synthetic data.
Synthetic data is just what it sounds like: it’s fake. A growing number of synthetic data firms including Facteus use machine learning techniques to transform real-world information revealing sensitive details of bank transactions, clinical drug trial response, or travel locations into new data versions that they say cannot be traced back to identifiable people. Alphabet’s controversial Sidewalk Labs uses a synthetic data approach.
Facteus has traditionally served banking clients. The company provides analytics software and helps customers monetize consumer transaction data by spinning it into insights that show where people spent before buying lunch at Burger King, or what they bought after playing nine holes of golf, for example.
Now the firm focuses on providing synthetic data services to its 1,000 clients, many of them payment processors, credit and debit card providers, credit unions and large banking clients. Those firms want to turn purchase transaction data into something they can share with partners or use internally to detect fraud without risk of exposing personal identities of consumers.
Part of the promise of synthetic data is that it offers privacy protection that improves upon traditional methods. Synthetic data approaches are especially appealing in the highly-regulated realms of financial, health and clinical data, where fake versions of real-world data can still satisfy a lot of data needs.
Traditional data privacy techniques take raw information featuring names, addresses, ID numbers and other personally-identifiable elements and strip it of those identifying factors. But in many cases, classic de-identification methods are not fail-safe; data can be reverse-engineered and linked back to individuals.
Or, in the case of banks using their own raw data to perform fraud analysis and security measures, employees could defy protocol and obtain personal information.
By producing a synthetic version of an original data set, the same security measures can be performed without exposing personally-identifiable data.
Synthetic data firms take a variety of routes to get to their end result. Facteus provides clients with a server installed behind client firewalls. The server is loaded with Facteus software configured for each specific use case, but not connected to the internet. A client might need it customized so it can ingest electronic payroll data or cross-border transactions, for example. Once the synthetic data process is performed, clients can export the new data into a secure data cloud or upload it to an FTP server.
The system’s machine learning algorithm detects all identifiable or sensitive business information and completely removes it. The synthetic data process comes in when that deidentified data is altered by adding what data scientists call “data noise.”
“We never get the real transaction itself,” said Facteus CEO Randy Koch.
Coronavirus Live Updates: The latest COVID-19 developments in Seattle and the world of tech
Rather than ending up with a dataset that still includes real details of actual transactions, numerical details of that data are altered by adding noise. So, a purchase of a latte and snack at a coffee shop for $5.60 at 3:18 pm might be changed to show a similar purchase for $5.72 at 3:34 pm.
Even when making those mathematical changes, the data “still maintains its statistical relevance,” said Steve Shaw, SVP of marketing at Facteus.
Facteus also provides synthetic data to companies who might use it to determine whether a location makes sense to build a new store based on consumer spending in the area. In that case, the company gives its banking clients a cut of the data sale.
Another application of synthetic data is for artificial intelligence algorithms, which need mounds of data. But in the case of autonomous vehicle systems, for instance, there just isn’t enough real-world data that exists. Some data companies produce synthetic training data from scratch for AI developers.
A privately-held company, Facteus has been profitable since 2016 and has not raised venture capital funding. “It’s hard to start new companies especially in the data world especially if you’re not going to raise a bunch of private equity and VC money,” Koch said.
Koch was previously president of CommerceGuard, a joint venture between Samsung, GE, Siemens, and Mitsubishi), a global technology and data company, with operations in the US, Asia, Europe, and the Middle East.
The company has more than 30 full-time employees. Its economic impact report is the first of a planned series of weekly reports.
