DS News

60 tion is key to the swift and successful implemen- tation of AI solutions. THE BIRTH OF AI & ML e concept of AI was introduced in the 1950s with the development of such important ideas as neural networks, but a failure to meet its initial promise meant funding for AI dried up in the 1970s. AI underwent a series of booms and winters in the intervening years, but unrealistic expectations and the rise of desktop computers meant it did not make a genuine comeback until the early 1990s, when the obstacle of insufficient computing power was overcome and AI began to enter various disciplines. e emergence of big data in the first decade of the 21st century opened the floodgates for AI development. Today, AI— especially the subset called machine learning—is being adopted rapidly across many industries. e rate of adoption is accelerating alongside the com- plexity of business problems being tackled by AI. ML is particularly relevant for businesses, and, indeed, most headline-catching AI solu- tions are based on ML. It is based on the concept that a computer system can train itself to perform certain tasks faster and better than engineers could program it. e ability to access and process large data sets—better known as big data—made it possible to implement the concept. Once big data became widely available for businesses, either directly or through their technology partners, many ML-based systems were created and deployed. THE BUSINESS BENEFITS To implement an ML solution for a busi- ness problem, the training data and data model must first be identified. e learning algorithm processes the training data and produces an ML model, which is used to make predictions based on new data. Simple problems such as clustering or anomaly detection do not require sophisticated algorithms. In those cases, so-called unsuper- vised machine learning can be used. is method facilitates simplified data pre-processing that requires no human input and generates satisfac- tory results based on small datasets. e results generated by unsupervised ML usually do require further human interpretation, however. is is because certain patterns or relations between data points may become evident, but a human brain is needed to determine what these patterns mean. For example, unsupervised ML can be used to group properties into neighborhoods or to seg- ment borrowers into groups with similar demo- graphics, but the technique has limited capabili- ties to solve more complex business problems. Supervised ML can be used to train ML models to provide insights into more sophisti- cated business applications, such as advanced classification or categorization. A borrower's propensity for default or the probability that a loan will default are good examples of classifica- tion problems relevant to the default servicing industry. To train ML models to provide such insights, data is pre-processed and labeled by humans before the training process can begin. Borrowers who have defaulted, for example, would be clearly labeled for a learning algo- rithm. With this defined combination of inputs and outputs, an algorithm can train an ML model to recognize loans that are more likely to default or become delinquent. e challenge here is that substantial human input is required to link known results to his- torical data and to provide a sufficient quantity of high-quality data. Furthermore, human pre-processing of the training data the learning algorithm consumes, usually represents around 80 percent of the time required to deploy an ML solution fully. e availability of relevant data is critical. Nonetheless, the benefits of being able to find the answers to far more complex problems are worth the additional effort. DATA IS KEY e levels of human input and sufficient quality data required are partially responsible for the slow adoption of ML systems throughout the default property servicing industry. is presents a huge opportunity for the first players who successfully implement ML-based solutions that disrupt the industry. is is already evident in other industries, where the first businesses to exploit AI are pulling ahead of their competitors. For example, Yelp uses ML for the picture- classification technology it includes in reviews and Pinterest uses it for everything from content discovery to spam moderation. e advantages enjoyed by the companies who are leveraging ML is only accelerating because ML-based AI systems will self-improve as more data becomes available. As AI becomes more powerful, it can start modifying itself to make itself smarter. As it improves its capabilities, it gets better at mak- ing itself smarter, so its intelligence soon grows exponentially. Hence, those early adopters will become more difficult to catch than the early adopters of earlier technologies. As we have discussed, the availability of data and the data model are the keys to success when it comes to implementing ML solutions. Most businesses gather vast amounts of data, but pro- cessing and analysis to build ML models are not among their areas of competency. It may also be difficult to identify the data that is relevant to their business. In some cases, particularly within the default servicing industry, the most impor- tant data points might not even be collected. Data associated with activities and processes related to an asset lifecycle is dispersed through many specialized service providers. A loan on a property might be under the loss mitigation process, for example, but a servicer who carries out inspections on that property might not even have that information available in their system. As a result, they may carry out unnecessary or incorrect inspections. ankfully, the gradual integration of the industry's computer systems opens access to data points previously locked in specific verticals. Now that advancements in computer science have enabled AI—particularly ML—to enter the realm of real business applications, compa- nies who want to remain competitive in today's landscape must adopt and implement these technologies. As previously discussed the most expedient way to do this is to leverage the ex- pertise that already exists in property servicing platforms/integration platforms that are focused on maximizing AI/ML value by reducing costs and preventing losses. AI and ML bring a new dynamic to property management, using the concept that systems can learn from data to identify patterns, help make decisions, and identify anomalies to enable management by exception— thus cutting cost and increasing efficiency.

• Cover