As I make my way into the kitchen to make my first coffee of the day, I ask Alexa to play Gabriella Quevedo. A moment later, gentle guitar music fills the room.
In that fraction of a second, the software in my Bose smart-speaker detected the wake word Alexa, recorded my voice and sent it over the internet to Amazon’s servers, where Amazon Voice Service converted my speech to text, interpreted my request, correctly identified what I’d asked for and contrived for Spotify to play Gabriella’s music for me, on my speaker, in my kitchen.
It’s nothing short of miraculous that Alexa gets this one right. There are an infinite number of ways to form a sentence and I’ve yet to meet another human being who can understand my pronunciation of Quevedo, let alone intuit what I’m on about.
Alexa pulls off this amazing feat through the magic of AI. Well, OK, it’s not magic. It’s a combination of Natural Language Processing and Natural Language Understanding (inferring not just what I said, but what I meant). But as Arthur C. Clarke said, “Any sufficiently advanced technology is indistinguishable from magic.”
The magic of AI has become ubiquitous. It targets us with ads in our Facebook feeds. It powers Alexa and Siri. It recommends films for us to watch on Netflix and music to enjoy on Spotify. It helps drive our Teslas, optimises the routing of our Ubers and supervises our Amazon deliveries.
Clever automation no longer surprises us. We’ve come to expect a level of intelligence. From mobile banking to developing vaccines, AI is powering new ways of working and new business models that would previously have been seen as too complex to contemplate.
The flavour of AI at the heart of almost all commercial applications is Machine Learning. This is the process whereby computers are trained on multiple sets of data in order to improve their decision making, in much the same way we improve our own decisions.
This training may be done using specially created data sets or through iterative exposure to real world examples. It may be supervised or unsupervised:
Whatever the approach, the end result is a computer system that can do four things really well – faster and better than any human:
Over the last three decades, our business processes have become increasingly automated, starting with simple workflow and evolving through robotic process automation to the point where machine learning can be used to optimise processes in ways that far exceed our own abilities.
This latest form of AI-enabled process automation is referred to as Intelligent Automation. What sets it apart from simpler forms of automation is the way information is used to optimise the process and make decisions earlier than ever before. To give one simple example (a project carried out by one of our member firms):
More generally, wherever there is a large amount of complex data that can be used to inform high-value decisions, there is an opportunity to use machine learning to improve those decisions and drive value for the business and its customers. For example (again from one of our member firms):
More advanced forms of machine learning, such as neural networks (collections of algorithms modelled loosely on the human brain), can be used to automate processes that rely on recognition of images or patterns. These are particularly useful in applications that involve natural language processing or image recognition. For example:
Even more advanced forms of machine learning involve the use of layers of neural networks each extracting a higher level of insight. This is known as deep learning. These techniques are applied in a range of applications, from the machine vision used in self-driving cars, to drug design, to the famous AlphaGo board playing AI. They are also being used in a business context, where they provide the ability to process ever more complex data with increasing accuracy. For example:
The short answer is wherever you have processes that involve large amounts of complex data that can be used to drive critical, high-value decisions.
If you can see areas of your business where you have high volumes of data and high-value decisions then there is most likely an opportunity to generate value from data and machine learning. A good example would be the use of real time claims data to inform dynamic policy pricing in the insurance industry.
A good starting point, when looking for the best opportunities to apply AI, is to ask the following questions:
One. Where do we have people using logic or judgement to make decisions?
Two. Are these decisions high value or high impact for us or our customers?
Three. Do these decisions involve the use of data from multiple sources?
Four. Do other sources of data have potential to improve these decisions
Five. Is there a high cost to mistakes or bad decisions?
Six. Are the processes high volume or repetitive?
Seven. Do we have bottlenecks in terms of capacity or skills?
Many successful AI projects start with a data study or proof-of-value experiment to test the underlying hypothesis and build the first iteration of the data model and algorithms. Indeed, one reason why so many AI projects flounder in their early stages is a failure to get to grips with data that is largely siloed and unstructured.
The skills and experience needed to do this are in very short supply. A combination of machine learning and data science expertise is a prerequisite for successful AI projects and people with these skills command a hefty premium in the job market. There is also a very real risk of reinventing the wheel if the team are not experienced with the extensive libraries of datasets, models and algorithms already in use.
For this reason, we strongly advise our clients to engage with niche firms who specialise in Applied AI (data and machine learning) and Intelligent Automation and who have these skills and experiences in depth, across many industries and use-cases. Many of these firms also have proven platforms on which to build solutions, rapidly and reliably.
There is a paradox inherent in artificial intelligence: once a difficult AI challenge has been mastered, it is no longer considered an AI challenge. Over the last few years, we have become so accustomed to the use of AI in our lives that we no longer regard it as exceptional.
In much the same way, AI is becoming embedded in our business systems and processes both directly and indirectly, because so many of our processes are now carried out by SaaS software in the Cloud, the natural home of data and machine learning.
Nonetheless, AI remains something of an enigma to many executives and the overwhelming majority of in-house AI projects fail to deliver. The solution is to reach out to the experts for help and advice.
If you would like to learn more about the examples in this article or the expert firms involved, please do contact us at innovation@clustre.net