The best of machine learning and deep learning models depend on a large amount of processing power so that a series of calculations are done in as little as microseconds or nanoseconds. However, there isn’t enough processing power across enterprises to implement ideal artificial intelligence techniques, which poses a challenge for advanced implementations. To tackle this, cloud computing and massively parallel processing systems are needed for enterprises to deploy AI systems that are scalable in accordance with the rising volume of data.
A report suggests that by 2020, 85% of CIOs will be piloting enterprise AI projects with a combination of buying, building and outsourcing initiatives. As the subject of democratisation of AI and deep learning technologies keep popping up in a variety of industries, the demand for AI professionals has become very high. The problem is that there’s a clear shortage of skills in data science, robotics and AI engineering. Even though there are about 300,000 AI professionals across the world, the number of available job roles for such professionals run in millions, according to estimates. In fact, staffing skills is the number one challenge for 54% of CIOs looking to adopt AI, says one report.
Data Plays A Crucial Role For AI To Thrive
It goes without saying that the accessibility to ‘right data’ plays a crucial role in creating the right AI model. With the tremendous volume and velocity of data in the enterprise, one of the biggest challenges is to be able to make sense of it all to drive profitable business decisions. Too much data can take the focus away from actionability and can cause data paralysis. It is important to capture data and correct the noise to have the right data strategy so AI models can be trained in the most efficient manner. However, datasets that are relevant for AI applications to learn are often rare. The most powerful AI machines are the ones that are trained on supervised learning. This training relies on labelled data – data that is organised to make it work for machine learning, and such labelled data is limited in the enterprise.
To take advantage of AI requires that enterprises integrate all the relevant data. AI systems cannot derive ad hoc insights unless the data is quantified and inserted in a data pipeline which is then connected into the model itself. In any organisation, there can be thousands of disparate streams of data or metrics. To make data streams converge at a single data warehouse or data lake is a big challenge. Gartner research has found that organisations report poor data quality may cause an average of $15 million per year in losses. The situation may actually become worse as data comprising of different formats become increasingly complex — a challenge faced by organisations of all sizes. Businesses with many units and operations spread out in varying geographies may experience more severe data quality issues.
One of the other issues is an assumption that AI does not need governance or being supervised: the assumption that it is all well-programmed and will work forever. But any change in a business model, company’s policy or government regulation can impact artificial intelligence systems and therefore a governance layer is always needed. Organisations need to monitor the outcomes of AI-driven projects and refine them to drive improved outcomes over time.
AI Innovation Does Not Mean Infringement Of Privacy
Another challenge that businesses face has to do with user privacy. According to PwC research, people have concerns regarding data privacy to share data even for a better experience. A vast majority of respondents (93%) reported their hesitance to share personal data such as medical records. At a time when large tech corporations have come under fire for violations of data privacy, it is therefore imperative for businesses to be compliant with the regulations. Privacy laws can hamper the way AI models are trained. Regulations like GDPR have clauses which oblige companies to provide either detailed explanations of individual algorithmic decisions or general information about how the algorithms make decisions, which can make it difficult for companies to adopt AI.
The Opportunity: Customer Experience & Contextual Recommendations Will Drive AI Growth
With all the efficiencies that exist today, enterprises are looking for different ways in which processes can be made more efficient, costs can be reduced and decision making can be enhanced. This is where AI can be of great help. Today, exponentially more computing power has become available to train larger and more complex AI models. With the ability to run millions of simulations rapidly, or analyse large volumes of historical data, AI systems can detect non-obvious variables that have the predictive capability and leverage them to acquire a competitive advantage.
Enhancement of customer experience is one of the biggest opportunities which businesses aim to achieve using AI systems. The highly repetitive customer interactions can be better handled by automated systems using natural-language processing. For an e-commerce business, chatbots, either through voice or email can resolve the extremely large volume of customer requests. Analysing unstructured data from multiple sources across social media and web channels, AI systems can present actionable insights on the data in real-time.
Businesses need predictability of functions and in this context, AI-powered recommendation systems have a prominent role to play in driving revenue growth. For example, AI can add value for sales through contextual recommendations using data in customer interactions across different channels. AI-driven processes can also be applied to a number of related areas at the same time. Consider the departments of an e-commerce company where an AI system trained on customer data and preferences from one department, can also be applied to other departments in making the right decisions. The goal is to optimise process efficiency by eliminating pain points using artificial intelligence.
