A friend of mine runs the development of risk management systems at a top Wall Street firm and recently I got some insights into his work. Fascinating stuff: he develops the models and systems that hedge the risk of the firm’s investments based on thousands of inputs and sophisticated business rules that manage their financial risk in real-time. To him, a few million dollars is a rounding error, since the models hedge risk on billions of dollars a day. Essentially, it’s a giant algorithm. A model.
I believe algorithms will one day rule the world. At least the business world. Let’s take customer service. It’s always nice when you get a little extra customer service at a store, a restaurant, a hotel. Anywhere. And these days, with the inter-linkages of customer touch points and CRM systems that track your preference for, say, fluffy vs. flat pillows, we’ll be seeing more algorithmic driven service. That phrase sounds like a paradox because isn’t service, by definition, delivered by humans? Well, in the age of marketing automation, with pre-defined rules that trigger auto messages it may not be the nice lady at the counter who sends you that lovely follow-up note. And when it comes to the business of customer service, we should distinguish the delivery mechanism from the rules that govern it. The business rules will increasingly be based on data models.
American Express is a good example. They proactively call people when their algorithm notices a purchase that is a statistical outlier — an “unusual transaction” they say. How good is their algorithm? Someone I know always rounds up every restaurant charge on his Amex card to the nearest dollar. Instead of $56.81 restaurant tab (including the tip), he’ll add 19 cents to make the bill $57.00. The rounding makes it easier for him to review charges on his Amex bill; if he sees a restaurant charge that includes anything with cents he knows it’s wrong. Recently, in a rushed moment, he forgot his usual pattern and did not round up the bill. The next day, American Express called him to inquire about an unusual transaction. Their algorithm recognized his purchase was at a restaurant and that he did not round up the bill as usual. To him, it was a proactive security measure and really good customer service. Behind the scenes, it was a really smart model.
Algorithms are everywhere. When you call an 800# to inquire about a product and get extra speedy service, it may be due to a program that recognizes your phone number, associates it with a zip code and other data that predicts you are more likely to buy. When you check out in a supermarket and receive coupons printed on your receipt, the offers are based on massive data sets of aggregated purchase history run through a model that predicts which coupons are most likely to generate future purchases. Google has a $20 billion dollar business based on a model so elusive that an entire industry (search engine optimization) is built around figuring it out. Amazon.com too, but that’s an obvious one. The notable thing about Amazon’s product recommendations is that they are determined in real time based on your online browsing patterns. Click on a Kate Bush album and you’ll see other albums from Kate Bush or artists similar to her — not just in the emails you get afterwards, which has been around for a while, but during the same web site visit.
For airlines, pricing of airfares is based on an algorithm with a complex set of rules that optimize revenue yields for routes. Prices – and price elasticity – vary by location, number of flights in the route, days before departure, booking levels, seasonality and of course competition. Seems like the airlines need an algorithm to manage their algorithms.
These kinds of modeling applications have long been the purview of enterprises with deep resources: in-house statisticians, data scientists, programmers and quality assurance — or funds to outsource the development of custom models. But now it is it is easier to run models over large data sets than ever before. There are many pre-built statistical models available for purchase that can solve for a wide variety of use cases out of the box. And when custom models are required, programmers can create data analysis faster, with fewer lines of code, especially with programming languages like R which is designed specifically for data analysis. It’s estimated that R is used by over 2 million analysts world wide. That’s a lot of people developing code predicting your inclination for chunky peanut butter. And now, in tech circles the term “big data” is almost a cliche at this point.
I believe that cloud computing, accessible statistical analysis and the age of apps will result in the democratization of models — an era when small businesses have easy access to algorithms that power their operations.
We’ll see a day when pizza shops predict with accuracy which toppings to include on their pies based on the hour of the day (and the optimal order of delivery drop-offs to reduce the costs of gas and speed up delivery times). We’ll see a day when small online merchants optimize their merchandise and promotions in real-time just like Amazon.com. Powering the shift of algorithms down market will be service providers focused specifically on that goal. Just as Constant Content did for email, Hubspot did for online advertising and social media, Google did for analytics and MarketTools did for online market research. All of these companies brought high end solutions down market to mom and pop businesses. The same will be done with algorithms.
And since 90% of all businesses are small companies, algorithms will rule the world. Or will they? Eventually, as algorithms become pervasive and we become an Economy of Algoritihms, the efficiency and effectiveness of models may result in their over use. We may become overly reliant on the algorithms, because they will drive commerce, customer service and marketing. But no algorithm will be able to detect the smile on a customer’s face or the furrowed brow of the annoyed prospect who waited too long in line.
It reminds me of the Star Trek episode where Captain Kirk and his crew are surrounded by enemy aliens who are “more advanced” than humans and have the crew out numbered. With one look of the eye and without saying a word, Kirk gives a nuanced glance to his crew…phasers are shot, a fight ensues and you know the rest. Victory for the Starship Enterprise. When algorithms rule the world, ultimately, there will be a premium placed on wise human judgement. Sure, the nice lady at the counter will know that you like fluffy pillows by looking in the hotel’s CRM system. But it will be her, and her alone, who decides to offer you an extra cookie after that long flight.