The AI Advantage
Businesses that capitalize on artificial intelligence (AI) and cognitive computing realize real results. Those that embrace AI see annual growth rates that are 1.9% higher than competitors and they expect AI to boost profitability by an average of 38%. AI opens the door to new business models, streamlined processes, and problem solving in previously infeasible ways. The artificial intelligence era has arrived.
AI is weaving its way into all aspects of everyday life, from our mobile phones to home assistance (Alexa and Google Home). The AI Advantage comes from elevating human ability to use otherwise unmanageable quantities of data as well as freeing them to focus on higher value judgement driven tasks.
In an increasingly connected world, there is an explosion in data generated and gathered. This is more data than any group of people can feasibly manage. AI is needed to parse through the noise to gain insight and harvest the value.
Despite these substantial and well understood benefits, the hype requires a caveat: AI is not the silver bullet for every problem and can be difficult to scale. Most modern AI applications solve discrete problems like visual recognition or natural language processing. We are a long way from the general artificial intelligence that has been popularized by Alex Garland’s Ex Machina and HBO’s Westworld. Instead we see a range of maturity:
To gain advantage from AI, organizations must understand the types of problems AI can solve, how they can start applying AI, and what steps need to be taken to implement AI solutions that empower the workforce.
AI Today
As we can see in the figure above, most AI applications in industry today is on the simpler end of the spectrum, with more complex flavours only starting to gain traction. General AI is still in early research stages.
Most adoption today is where companies have carved out niche applications or use cases that fit a certain mould. Typically, value from AI is derived from solving problems that fit the following criteria:
· High data volumes for sampling
· High data quality and common data structures
· There is a specific, measurable goal in mind
· Relatively low stakes, high potential reward
· Applies structured and repeatable methods
· Human supervision or decision-making authority is available to guide AI training
Solving these problems has proven commercial and consumer applications. The solutions are redefining business models, augmenting existing products and processes, and reshaping cost structures through operational efficiency. While the lines between applications are blurred, they fall roughly into the categories described below:
As these solutions mature, they push the envelope toward science fiction through deep learning and reinforcement learning. IDC predicts the global market for content analytics, discovery, and cognitive systems software will grow to $9.2 billion by 2019 – more than double that of 2014. In Canada, the government and corporations have pledged $180 MM to fund the Vector Institute and position Canada as a world-leading center for AI research. This institute will develop experts in deep learning and machine learning, and serve as an engine for an AI super cluster meant to drive economic outcomes in the Canadian marketplace. Leading organizations – Canadian and otherwise – are vying for a seat at the table. With this market growth, it’s easy to imagine how ubiquitous AI will be in the near future.
Applying AI to Your Business
Excited business stakeholders have kicked off small AI projects in silos with mixed results and value scaling challenges. It is important to prioritize. Proof of concepts are valuable to the extent that they advance the corporate strategy, prove potential, or have a clear business case. Organizations investing in AI pilots without a clear strategy can find that they are eclipsed by focused competitors. For this reason, it is important to honestly assess where specific capabilities can drive the highest return or enable key business priorities. Target value in centers of cost and processes with a high number of simple decisions because they are most likely to have challenges that fit the mould. Consider using both quick-win technologies (e.g. robotic process automation, chat-bots) and incubating longer-term deep learning and reinforcement learning applications.
Data is critical and AI opportunities in enterprises are often constrained by the data landscape. Architecture and processes to acquire sufficient volumes of high-quality, relevant information are fundamental to unlock AI potential. Investment in these areas should be guided by the organization's aspirations. Enabling real-time in legacy systems, while potentially fruitful, can be expensive. Consider real-time data flows when real-time adaptation and learning is tightly linked to competitive advantage.
Each flavour of AI has its own requirements: supervised and semi-supervised techniques are ineffective without critical volumes of labeled examples, reinforcement learning demands feedback to evaluate the performance of actions as they are made. Begin with getting data right for the scope of the problem. Incrementally improve data quality to harvest value along the journey rather than attempting solve all the data issues before embarking.
As cognitive technologies aid people in their day-to-day roles, talent strategy and training need to adapt. Understanding and addressing these shifts should be a key focus of an AI-oriented team, which will build core AI skillsets while also driving technology and process transformations. In parallel, companies can get started now by initiating partnerships with academia and research institutions, such as the Vector Institute in Toronto. Researchers need data and real-world applications to prove value, and enterprise environments are the perfect proving ground.
With these initial steps, businesses can lay the groundwork for future AI investment, creating value from applications today, while reducing technology, data, and talent barriers to future opportunities.
Thanks to Danica McLeod, Kevin Raymond, William McCluskey for their contributions to authoring this post.
*The views expressed in this post are the opinions of the authors and are not representative of Accenture’s opinion on this topic*