A New York startup company spun off from Citigroup is about to launch a service that can offer consumers recommendations for restaurants and other retail outlets, based on an analysis of billions of credit card transactions.
Jaidev Shergill, CEO and founder of the Bundle.com personal finance service, described this new service at the GigaOm Big Data conference, held this week in New York. He argued that a service based on people's recorded buying habits would offer superior recommendations to those from opinion-driven services offered today.
The company plans to launch "a more holistic" recommendation engine for dining and other retail establishments, he told the audience. "We believe that [financial transaction] data can actually result in better recommendations and better decision-making for people," he said.
Core to these recommendations will be analysis of the anonymised transactions from 25 million Citi credit card holders, who generate about 1 billion transactions a year. They are cross-indexed with US Census bureau data and other third-party demographic information.
User-generated recommendations, like those found on Yelp or Citysearch, are inherently flawed by their subjectivity, Shergill argued. Such online recommendations - for everything from books to new cars - are generated by feedback by users, who may provide a written summary along with a ranking of some sort.
"The one thing I think that leads to poorer decisions is the amount of subjectivity that has come into the recommendations. It actually results in worse decisions," he said. He noted that most restaurants within services like Yelp usually average between 3.5- and 4.5-star ratings but users will not know if any given recommendations came from, say, a friend of the restaurant owner.
In the case of restaurants, credit card transactions could offer a better glimpse into the value of an eatery, for they track how many times a patron returns and how much they spend, he argued. An establishment with lots of repeat traffic is probably a good bet.
Moreover, comparing restaurants' average bills can point to locations that are better value. Two restaurants near one another may each have a loyal following, but the one with the lower average bill may be "easier on the wallet," Shergill said.
Shergill displayed a sample map of dining establishments in New York City. "We basically mapped out which restaurants people went to. They serve as nodes of where people went and where else they went to," he said.
In addition to providing recommendations for customers, the service could also provide intelligence for the retail establishments themselves. For instance, it could provide a breakdown of where a merchant's customers live by zip code. It could also provide the average income of residents in that zip code, which could be a valuable marketing aid for the establishment.
He noted that, in the New York City sample, residents of the Upper West Side of Manhattan spend the most on dining out, with the top 10 percent of these foodies spending about $2,200 (£1,374) a month.
However novel the idea of transaction-based recommendations, not everyone in the audience was sold on the idea.
"Credit-card purchases do form an objective spending record, but they're neither comprehensive - each of us also uses cash, debit, electronic transfers, even PayPal - nor 100 percent reliable for predictive purposes," said conference attendee Seth Grimes, a text analytics industry analyst for Alta Plana. Grimes said that financial data could be married with other forms of input, such as web traffic analysis, surveys, social chatter and, yes, user reviews.
"Bring multiple data sources and complementary approaches to bear on your predictive analytics problems. Consumers and product and service providers alike will get more useful recommendations and insights," he said.