Paying Big for Cheap Prediction
Among the activities at the recent Cisco Collaboration Summit was a thought-provoking keynote delivered by James Cham of Bloomberg Beta, an early-stage venture fund that invests in data-centric and machine learning-related startups. In his address, Cham made the point that the ability to do complex math "cheaply" has spurred the recent technology advancements in machine learning. Cheaper math enables machine learning algorithms to run faster and cheaper, and this in turn enables the cheaper predictions machine learning delivers.
You might ask, "Why is cheap prediction important, and how does it impact our industry?"
It turns out that machine learning is really a prediction mechanism. It tries to predict if a person has cancer or not; whether a self-driving car should stop, speed up, turn, or keep going straight; the likelihood that someone using a Web browser will click on a particular ad; which contact center agent is most likely to close a sale; the intent of what someone said or typed; and so on.
Being able to deliver accurate predictions is a hot capability, even in the communications space. Since January, major players in our industry have acquired machine learning companies. The implication is that they have done so to purchase prediction technology:
- In January, Avaya acquired Spoken, and gained its IntelligentWire technology. IntelligentWire does speech-to-text and sentiment analysis to identify patterns in conversations in order to predict next best actions or when an agent conversation may in trouble.
- In February, Genesys purchased Altocloud. The Altocloud technology monitors real-time digital behavior to predict what a customer may want to do next, which customers may buy or abandon a site, or when to do a customer intervention (such as popping up a special deal).
- On May 1, Cisco announced that it was purchasing Accompany. The Accompany technology searches the Web and predicts company org charts and the characteristics of individuals within an organization, with the aim of helping users make and keep the "right" external relationships.
- Just last week, 8x8 acquired MarianaIQ. The MarianaIQ technology searches social media channels and rationalizes them so that it can specifically identify which channels an individual uses. Think of it as contact center omnichannel, but in reverse. This information is then used with product and personal profile data to proactively market to these individuals using their own social media channels.
As economists Ajay Agrawal, Joshua Gans, and Avi Goldfarb stated in their book, Prediction Machines: The Simple Economics of Artificial Intelligence, prediction is a foundational input to almost every business process. Better prediction means better information, which enables better decision making. They contend that reframing artificial intelligence and machine learning as a shift from expensive prediction to cheap prediction, or "from scarce to abundant is invaluable for thinking about how it will affect your business." Machine learning and the prediction it enables has gone from expensive to cheap, and we are seeing it emerge, mostly in quiet, behind-the-scenes ways, to change how we work and live.
To understand the impact that going from expensive to cheap has on society and individuals, we need look no farther than our use of artificial light. According to Yale Economics professor William Nordhaus, the cost of light in the year 1800 was over 400 times what we pay for light today. As the cost of light has fallen over the years, we no longer even think about whether to turn the light on or even turn it off.
The point is that as prediction becomes cheaper and cheaper, it will permeate into every aspect of our work, social, and personal lives where prediction can play a role, and to some extent we already see this occurring. Every time we search the Web, shop online, make a credit card purchase, tweet, change our smart thermostat, watch a program on cable TV, make a cell phone call, call a contact center, etc., our data either is or soon will be processed by a machine learning algorithm to enable better predictions of actions, behavior, and outcomes.
So, where does all of this lead us?
The economics of something becoming cheap usually leads to other things increasing in value. With cheap prediction, the value of good judgement and reasoning -- something machine learning systems aren't good at doing -- becomes more valuable. In my previous career, when I first began simulating petrochemical processes using complex thermodynamic, heat, and mass transfer algorithms, we'd often say, "The computer gave the answer it was programmed to give, but what does it matter if we blew up the plant."
Judgement and discretion will certainly play a big role in how machine learning's cheap predictions are put to use. And, this judgement and discretion is what will allow us to tweak our machine learning algorithms to achieve better and better predictions.
Avaya, Genesys, Cisco, and 8x8 have all paid big money for cheap prediction because it can help inform business decision makers to create high-value outcomes. So far, these "cheap predictions" are mostly centered in the contact center (Avaya, Genesys, and 8x8) and on the collaboration experience (Cisco). Microsoft is clearly in this game as well, as it continuously deploys more machine learning as part of its Office 365 and Teams offerings.
Cheap prediction is in demand and can cost serious money for those rolling it out. However, the capabilities it enables can and is influencing how each of us does business and interacts with the many organizations and people in our circles of influence.