Monetising an Algorithm
The Story of FICO
There are always ones that get away. Warren Buffett calls them errors of omission. “The mistakes that have been most extreme in Berkshire’s history are mistakes of omission,” he told shareholders at his 2001 annual meeting. “They don’t show up in our figures. They show up in opportunity costs.”
In 2017, he and Charlie Munger reminisced over some of their biggest. Buffett highlighted Google. “We were their customer very early on with Geico. These figures are out of date, but as I remember, we were paying $10 or $11 a click. Any time you are paying somebody 10 or 11 bucks to punch a little thing, where you have no costs at all, that’s a good business… The guys that designed the Google prospectus, they came to see me… I had plenty of ways to ask questions to educate myself. I blew it.”
If I have a career-arching error of omission it may be Fair Isaac Corporation, otherwise known as FICO. The stock has compounded at 20% per year since coming to the market over 35 years ago. This year alone it is up 90%. Its core product, the FICO Score, is integrated into the fabric of consumer finance in America, used by 90% of top lenders. Like Google, the company earns money every time somebody “punches a little thing”, for which it bears no cost. Yet, throughout my investment career, I never owned it. There’s always a reason of course: disruption, antitrust, slowing volumes, artificial intelligence.
This is the story of an investment I never made.
Creating an Algorithm
Fair, Isaac & Co listed on the Nasdaq stock exchange in July 1987. It was a big year for IPOs, and Fair Isaac’s may have got lost in the crowd. In the run-up to the crash later that year, over 500 deals came to market, raising $22.5 billion; Fair Isaac raised just $14.6 million. With revenues of only $18.1 million, it wouldn’t have been a surprise if the company’s stock market debut went unnoticed.
By then, the company was already 30 years old. William Fair and Earl Issac had met at Stanford Research Institute, where they worked on operations research for the military, William as an engineer and Earl as a mathematician. Seeing an opportunity to offer similar operational analysis to the private sector, William founded a non-military practice at the Institute before leaving with Earl to set up an independent consulting firm. They each put $400 into the business and leased an office in San Rafael, California. Lacking computing resources of their own to conduct research, they negotiated a time-share deal with the Standard Oil Company of California (today’s Chevron) to tap into its mainframe during non peak evening hours.1
It took them almost three years but, in 1958, Fair and Isaac identified consumer credit as a process in which they could put their ideas to work. It was the year of the Fresno Drop and credit cards were beginning to catch on. As Bank of America discovered when it flooded the California market with pre-approved credit cards, not all of them paid back as expected. Fair and Isaac (and first employee Earl Follett) realized that they could deploy statistical techniques to measure credit risk. By using historical data already captured by finance companies, they devised a scorecard capable of predicting a person’s creditworthiness based on past behavior.
Fair and Isaac sent letters out to about 50 major credit originators – mainly consumer banks and finance companies – asking for a meeting to explain the value of their approach. Only one institution replied: American Investment Company, of St. Louis, the fourth largest consumer finance company in the country. In partnership, the firms analyzed 13,000 of American Investment’s “good” credit files and 1,000 of its “bad” credit files to determine a model. The result was a system which American Investment projected would enable it to reduce bad debt charge-offs by 20% with only a 3% reduction in total volume.2
The problem was that the system was expensive to produce. As late as the mid-1970s, Fair Isaac employees had to spend days in client back rooms gathering data to build their models. Copying and processing credit records came at a cost. This inflated the price of a scorecard, which, in 1976, stood at $32,000, a headline number that didn’t include the extra costs clients incurred maintaining their data. It didn’t help that bankers were wary. “Lending…had been done judgmentally for centuries,” recalled 1990s CEO, Larry Rosenberger. “And so here were these two guys saying we can help you do this better and as you can imagine there was a fair bit of skepticism about whether that was true or not.”
But Fair and Isaac managed to pick up more clients than just American Investment. They won their first banking customer in 1970 and had some success with department store chains that operated credit programs. In each case, scorecards were customized to lending clients based principally on their internal data.
It took government intervention for things really to get moving. In 1974, Congress passed the Equal Credit Opportunity Act, prohibiting creditors from discriminating against credit applicants on factors such as race. It’s not unusual for regulation to unlock opportunity in financial technology – we’ve talked about that before. In fact, regulatory change is often a more important driver than technology in kickstarting the growth of a financial technology firm.
In this case, algorithmic credit scoring allowed lenders to demonstrate that their credit decisions were not discriminatory. Although Fair Isaac didn’t testify in hearings, one large client, retailer Montgomery Ward, did. Under credit scoring, “no single characteristic will permit an approval nor cause a rejection for credit extension to the applicant,” its assistant corporate credit manager told a Congressional panel.
By the time Fair Isaac listed on the stock market, it had a solid roster of clients. It was also in the early stages of migrating its business model. Its scorecards pulled data from national credit bureaus purchased by its lending customers, but one client – First National Bank of Kansas – proposed using bureau data exclusively, the advantage being that rather than waiting for a consumer to apply for a credit card, they could be pre-screened. Fair Isaac developed a system called PreScore, built on a generic sample of data from each of the three bureaus’ data sets. In addition to its scorecards, Fair Isaac now had a new product line, offering scores based on bureau data.
“For Fair Isaac it was a breakthrough and a business model because we didn’t have to incur the labor of every custom project,” said a former executive. For First National Bank of Kansas City it was also a win: The bank went from third largest bankcard company in Kansas City to ﬁrst with just two promotional campaigns.
Fair Isaac wasn’t the only firm marketing such a score. Atlanta based Management Decision Systems actually signed the first contract with credit bureaus. But Fair Isaac’s long-standing relationship with lenders helped. Its executives “went to the CitiBanks and the AMEXes and Chases of the world and sold them on the idea of scoring the bureau data,” recalls an employee. “These customers, in turn, went to the credit bureaus and said ‘You will code this in, of course, won’t you?’ and the bureaus really didn’t have a lot of choice.” When Experian attempted to back out of contract negotiations at the last minute to focus on an in-house product, client threats to abandon them as a provider forced them back to the table.
Fair Isaac’s scores became known as FICO scores, and they set a standard. To solidify its position in the market, the company exploited competition between the bureaus. Consolidation had whittled their number down to three by the late 1980s which, for Fair Isaac, was optimal – not too many to generate scoring algorithms for, but enough for there to be competitive tension. Scoring models differed across bureaus, reflecting the different composition of their datasets, but Fair Isaac scaled the scores in such a way that the same number was associated with the same risk level no matter which bureau was used. By standardizing the score, Fair Isaac minimized switching costs for users, bolstering its own competitive positioning relative to the bureaus.
Standards normally permeate when they are free, open and usable. We’ve talked about standards before: from the meter rule to the shipping container to the screw thread. What’s different about the FICO score is that it’s not free – to Fair Isaac’s long term advantage.