Artificial intelligence, or AI, can be defined as “the ability of a machine to perform at the level of a human expert.” Recently, we saw a prime example of this when Google-owned DeepMind’s AlphaGo program beat the reigning world Go champion, Lee Sodol, 4-1 in what many consider to be the most difficult game in the world. What is most compelling is that the computer actually trained itself to play the game – it was not trained by humans.
Hollywood movies have, perhaps, done AI a disservice presenting AI as the brains behind megalomaniac robots creating havoc for humans. But the reality is that as computer systems become more powerful and complex, the ability to apply machine learning – machines “learning” from past experience to provide increasingly more accurate results – will continue to expand.
In fact, artificial intelligence programs have already become a major part of how businesses operate. Companies are using these programs to automate an enormous variety of data management and analytical tasks.
Assessing Human Behavior and Emotions
Accurately assessing human emotions and behavior is extremely difficult. Some people never seem to quite get the hang of judging the emotional state of others. However, there are now AI systems that can evaluate the emotional state of a person by analyzing how they speak. These systems break human speech into identifiable features and, once combined into vectors, are inserted in machine learning algorithms. Those features can be associated with particular emotional states; by studying them, the AI of voice analytics can compile an accurate assessment of a person’s emotional state.
Predicting Future Behaviors and Actions
Predictive voice analytics programs take emotional behavior assessment even further by predicting future actions of customers who interacted by phone. After the voice features are those of previous customers. As the machine learning algorithms collect more data on the results of calls with customers, they can self-adjust their criteria for linking behaviors to outcomes. This means that, over time, the algorithms will become more and more accurate at predicting future behaviors.
While the “achievements” of machine learning are considerable, right now most artificial intelligence systems cannot make inferences or introduce new ideas – they simply cross-reference existing data. As we look ahead, true machine intelligence will do more than just use existing data to come up with answers, they will actually be able to understand how and why specific events occur, and proactively make recommendations based on that understanding.
RankMiner, an early-stage St. Petersburg-based predictive analytics company, is in the forefront of applying artificial intelligence systems as described above to leverage voice-based emotions and behaviors. Their patented software systematically analyzes interactions between call center agents and customers, which helps companies to identify their best and worst customers and evaluate and improve agent performance.
Call centers have achieved productivity gains of 20% or more by using RankMiner software to analyze every call (not just a small random sample), therefore being able to target struggling agents for additional training before they negatively impact countless customers.
The call center can also pair their top performers with the prospects most likely to buy, thereby closing more sales and wasting less time. RankMiner can also systematically identify which of the top-performing agents are likely to leave the company – so that action can be taken pre-emptively.
By identifying which prospects will say YES, which customers are at risk of leaving, and which “receivables” will pay (or not), call centers can generate substantial productivity increases.
The key advantage of predictive analytics artificial intelligence software is that it identifies patterns in seconds or minutes , rather than the days or weeks more traditional “human” methods can take to come up with a conclusion – a conclusion that may not necessarily be the correct one.
As mentioned earlier, the field of artificial intelligence is moving forward at great speed based on both the increased power of computer systems and human ingenuity discovering more and better ways to use AI.
RankMiner is not standing still.
They are expanding their capability to build models that are able to solve more and more complex problems. Functionality, such as time series analysis algorithms, will allow RankMiner to shift the focus of predictions on individual calls to predicting human behaviors as it evolves over time.
This will allow for even more accurate and more complex predictions for individuals. Other modeling expansion will include advanced feature analysis and selection and multi-target classification to further improve prediction performance.
The enhancements contemplated above will allow RankMiner to begin the process of automating model creation and fully automating model enhancement. What this portends is that the user of the system will be able to create predictive models from scratch without touching, or indeed even requiring the user to have any knowledge of software programming.
This will be the first step in creating a “self-service” product wherein non-data scientists will be able to build powerful machine learning based models, without requiring expert domain experience.
And finally, RankMiner will begin to incorporate reinforcement learning techniques into the modeling improvement process. In time, this concept may revolutionize the predictive analytics industry. In essence, RankMiner will build an artificial intelligence that will “learn to learn” what needs to be done to improve predictive performance based on past behavior, and the way the operational environment has responded to that behavior.
Could it be that Hollywood was not so far off?