From Uber, to Tinder, to Upwork, how Matching Algorithms Determine Our Lives

Google+ Pinterest LinkedIn Tumblr +

Matching algorithms have become prevalent in every aspect of our daily lives. From finding a date (Tinder), to finding housing (StreetEasy), to ordering a dog walker (Wag), to grabbing a ride (Uber), algorithms help us make decisions that simplify our daily existence.

Matching algorithms can be so disruptive that Alvin Roth received the economics Nobel Prize in 2012 for matching medical students with residency programs. In fact, at the heart of many of the biggest tech innovators lies a matching algorithm: two equally sized elements are paired based on an ordering of preferences (either set by each, or determined by the algorithm writer). This concept has applications in myriad areas of life — after all, many of the conundrums we humans face are simply a matter of finding the perfect match. But how effective are these algorithms, and are they really enabling better pairing than random fate was?

Data Says Matching Algorithms Are Not Always as Useful As We May Think

Research demonstrates that matching algorithms on dating sites are negligibly better at matching people than just random pairing would be. When you control for factors such as location, age, education, gender and other demographic features, the algorithms are essentially shots in the dark at suggesting potential mates.

One reason for this is that matching algorithms are based on the assumption that a good pair is the result of the input criteria fitting the characteristics of the match. However, there is no credible research that points to any factor such as similar tastes or interests being a good predictor for a successful marriage or long term coupling. While dating sites certainly widen the pool of potential mates and enable singles to meet people they may not otherwise have encountered, the scientific validity of the matching algorithm that many sites tout is questionable at best.

Matching algorithms have also disappointed in areas less nebulous than attraction. For instance, when Uber released its ‘Pool’ feature, riders found that 20 minute trips could suddenly take 40, with drivers doing multiple trips for other passengers or being stalled by late or confused passengers. While the system technically worked, its practicality was not well thought out. Similar issues have arisen in marketplaces like Upwork or Clarity, where the algorithm relies on simple tagging systems (e.g. a person has the tags “healthcare” and “technology” because they worked as a nurse, and later in technology). These tag-based systems don’t correctly identify intent or experience — if one were to search for ‘Technology in Healthcare’, they could get matched with a person who has worked in both industries, but has no knowledge of technology in healthcare.

While these services expand the pool of access that we have, they don’t always select the best options from this pool. Because of this, we’ve seen matching algorithm-based business models be both popular and frustrating — users recognize the usefulness of ordering an Uber Pool (for price and environmental reasons), but are frustrated when their trip takes twice as long as it should have.

As the CEO of a leading Robotics Process Automation (RPA) company explained, “Uber, Tinder, Upwork — all of these companies are giving consumers a sort of first generation look at the power of automation and matching algorithms. But they’re nowhere near mature. In the next few years, we’ll see these services become more and more precise and accurate with their matching.”

Matching Algorithms Are Still an Untapped Opportunity for the Business World

Many problems that businesses face today result from issues of matching. From recruiting, to sales, to product iterations, to finding office space — we spend an inordinate amount of time attempting to match aspects of business into various slots. Businesses that leverage effective matching algorithms to solve some of these issues will face a massive total addressable market.

In recruiting for example, according to a survey of 21 NewtonX former hiring managers at companies with over $500M in revenue, 76% stated that they did not use AI in recruiting, but 20% expressed that they would be open to trying it. An AI-driven recruiting tool called Woo said in a recent interview that by using a matching algorithm they “cut down the process by at least half, saving the recruiter around 50% of time taken to hire”. Tools like this that use matching algorithms are slowly gaining access to the $200B recruiting market.

Customer service is similarly in need of matching algorithms: the market for automated ticket triage alone is estimated to grow to as much as $6.5B by 2022. A recent NewtonX customer service survey of C-level executives at some of the biggest SaaS companies in the world found that while 13% believed investment in AI would be a top trend in 2018, none of the companies surveyed currently had AI-driven ticket triage. According to the VP of Product at a company that makes AI-driven automated ticket triage software, “Everyone wants this service. It’s not a very difficult thing to sell. The only barrier to adoption is getting prospects to understand how automated ticket triage works and showing them that products like these exist in the first place. Many of our clients didn’t even know automation in customer service at this level could actually work.”

Products that effectively leverage matching algorithms to minimize manual labor in these two key markets, as well as in others, will not only be successful in coming years, but will also become the norm in various key aspects of running a business.

The Next Frontier of Matching Algorithms: Using AI

Some areas are so complex that simple matching algorithms cannot provide a sufficient match. NewtonX, for example, employs a model to match clients who have extremely narrow needs (e.g., Virtual Reality market in agriculture in Africa) with experts faster and more accurately than any other company. In order to do so, NewtonX has built a matching algorithm that can identify extremely precise matches for experts both on and off of the proprietary system. This expands the number of potential experts, which increases the likelihood of an extremely high match percentage. In order to match the topics to experts with precision, NewtonX deploys data science including neural networks and feedback loops in order to develop the Knowledge graph that enables such complex matches.

The NewtonX platform thus achieves the three primary goals of utilizing a matching algorithm:

  1. Improved speed
  2. Increased accuracy
  3. Expanded pool  

Companies that rely on discovery will increasingly implement similar strategies to offer clients greater access and an improved user experience. As a senior level engineer at Uber for the dispatch algorithm put it, “Any product that is more convenient and delivers service faster than other brands will end up with a major strategic advantage. Consumers want things to be fast and easy. That’s it.”

Matching algorithms have expanded and refined our pool of options — from who we date, to who we hire, to where we get information. The next frontier of matching algorithms will offer increasingly precise matches using AI, and will choose these matches from an ever widening landscape of options.


About Author

Comments are closed.