Systems of hardware and software have always excelled at repetitive tasks, but modern cloud services make it possible to process absolutely titanic volumes of data quickly and affordably. With this capability comes the ability to derive meaningful information from new and historical data – including predictions – in near real-time, allowing your systems to behave in a way that can seem almost magical to their users.
The buzzword that applies here is machine learning (ML), and you’ll see it in many contexts these days. Some of the hype is merited and some are not, and to distinguish the two you’ll need to understand what ML is in general, and what each of the two varieties of it are good for.
In general, ML is pattern detection. By examining a large body of data, ML tools can spot patterns with greater sensitivity than any human. For example, an ML model (which is what we call an ML system that has been trained with data) might determine that people who buy fishing equipment in an online shop also tend to buy camping gear (or something more obscure, such that buyers of fishing things tend to enjoy a particular television series).
Conversely, they can also spot anomalies, which are exceptions to identified patterns. A model might “know” that people who practically never use their credit cards to buy toys or clothing for children are unlikely to make a payment to a school or childcare provider – and suggest that such an uncharacteristic transaction should be investigated as possible fraud.
ML models are trained using two general kinds of algorithms: Supervised and unsupervised. The difference has to do with how much human involvement there is.
Supervised algorithms generally require a set of training data, which a human (or team of humans) has labelled. Image recognition is an easy-to-grasp example. A set of training data might be in the form of photographs of animals, all identified in advance by human labellers. Some images would be labelled as dogs; others would be labelled as cats; still, others would be identified as horses. By examining the set of labelled images, the ML model would learn what distinguishes dogs, cats, and horses, even though they are all four-legged mammals that exist in various breeds and colours.
Unsupervised algorithms don’t require human preparation and instead rely on spotting trends in (usually numeric) data. For example, an unsupervised algorithm might take historical information about real-estate sales in an area and use it to predict the sale price of a house given its location, size, and year of construction. The algorithm would spot historical trends and apply them to new inputs.
Contact centre applications of machine learning are numerous. ML underlies spoken language comprehension, in which a caller is able to just say what he or she wants to do, and have the call routed appropriately. Similarly, ML models can be used to predict what a caller wants to talk about – maybe for example, by correlating a call’s geographic origin with information about recent hailstorms or bushfires. ML facilitates smarter, faster contact centres that don’t rely on old-fashioned DTMF menus.
CloudWave has experience with ML models and their applications in modern contact centres. Tell us about your data, and we will suggest how to turn it into an ML model that will give your customers a more magical experience.
댓글