There have been rapid and exciting developments in the fields of artificial intelligence (AI) and machine learning (ML) for the enterprise. But these developments present an unavoidable question: “Where should investments be made to enhance intelligence in enterprise IT systems?”
AI and ML are key for digital transformation initiatives, but that does not mean that every solution component in the IT stack must be more intelligent. Instead, it is key to judiciously choose the areas from which the most business value can be extracted from AI/ML enhancements.
Before examining which enterprise areas can benefit the most from AI/ML, let’s first consider where the intelligence function resides in each one of us. Obviously, it’s in the brain, which is close to our eyes and ears (our data sources) and holds our memories (our data stores). The brain identifies patterns in our experiences stored in our memories and uses cognitive abilities to give an intelligent response to stimuli. This is a good indicator to consider when building a strategy for adding intelligence to enterprise systems that enable digital transformation.
Intelligence In IT systems
For a long time now, enterprise IT systems have been programmed to perform specific sets of functions like providing user interfaces, storing transaction data in databases, implementing workflows and rules, integrating with other systems and providing reports to end users. Analytics systems use data warehouse/business intelligence technologies. With the steady upsurge of AI/ML, it’s essential to bring these technologies into the enterprise wisely rather than approach these intelligence capabilities as the proverbial hammer while every other solution piece is just a nail.
The AI techniques of the 1970s and 1980s were essentially built on rule-based and case-based reasoning approaches. The emergence of cloud, mobile and the internet of things (IoT) changed how data is generated and thus how it is processed at scale. Newer AI/ML techniques like deep neural networks (DNN) are based on learning from data representations as opposed to task-specific algorithms. ML can be undertaken with structured and unstructured data and with supervised or unsupervised techniques. The supervised and unsupervised techniques identify patterns from the data.
Some use cases of deep learning are:
- Text analytics
- Speech analytics
- Visual recognition
When combined with natural language processing (NLP), conversational platforms using DNN can conduct intelligent conversations and not just answer questions. For example, let’s assume a user asks a conversational platform the following question: “What is the closing price of a stock today?” Next, that individual follows up with the question, “What will that stock’s closing price be tomorrow?” The algorithm of the platform understands the context and relates the second question to the same stock mentioned in the first question and can strive to estimate the closing price of that stock.
Determining Where To Put Data Intelligence In An Enterprise
When introducing AI/ML, enterprises should focus on two important aspects: data sources and data stores.
Data Sources: Data can be sourced directly through user input via mobile or web applications or indirectly through social media feeds or other sources. For an organization on the path of digital transformation, solutions like chatbots, which are based on AI/ML technologies, can play an important role in putting intelligence in the right place for its enterprise strategy. Chatbots provide a humanlike response to user input and leverage DNN to form key components of conversational platforms. The dynamic nature of this technology is seen in smart speakers like Amazon Alexa, Google Home or the Apple HomePod. Other sources of enterprise data may come from system interactions or file transfers across system interfaces, but there are fewer possibilities for AI technologies in those scenarios and more possibilities for good old automation.
Data stores: Data stores are often grouped into operational or analytical data stores. Before the massive wave of AI adoption, enterprises used business intelligence (BI) techniques whenever they needed to extract intelligence from analytical data stores. While BI is still relevant, it can also be complemented by AI to make predictions more accurate using a range of techniques on analytical data stores.
Creating the right mix of AI and BI is another aspect of the enterprise strategy to put intelligence where it belongs. For instance, you can use BI techniques on sales data to make a forecast for upcoming quarters of the financial year. The sales forecasts may be further refined using supervised learning techniques of AI to make the predictions more accurate. Alternatively, AI and BI techniques can be used independently on raw sales data to validate the predictions made by the two different approaches.
Implementing AI/ML Capabilities
AI/ML techniques and tools are different from what is generally used in IT applications. Moreover, since the technology market is changing rapidly — with developments in vendor products and feature sets — both strategic and tactical approaches are needed to build and sustain an effective AI/ML capability.
A strategic approach involves the identification of data sources and data stores. It also includes the selection of a standard AI/ML platform. For instance, an organization might standardize on an AI/ML platform based on Microsoft Azure Machine Learning with HDInsight and Azure SQL Data Warehouse as its data store, using mobile/web apps and chatbots as data sources. The standard AI/ML platform is then supported by a center of excellence (CoE), staffed with resources and the appropriate skill sets to continually take up AI/ML projects for business units in the enterprise and support all digital transformation initiatives.
A tactical approach focuses on leveraging the AI capabilities that are built into vendor products. For instance, Microsoft Office 365 has several AI/ML capabilities that may be leveraged to improve productivity. In those cases, such tactical solutions provide significant improvements to the user experience of the point solution, with little or no additional investment.
The Bottom Line
In summary, when deciding to add AI/ML to enterprise systems, there’s no great value in attempting to make each solution in your IT stack intelligent. Instead, it’s best to arrange the capabilities near data sources and data stores to get the greatest benefit from the technology.