Every company’s artificial intelligence (AI) journey is unique. So is its application of AI.
Maybe your company has invested in big data systems and wants to apply AI but is unsure where to start. Maybe you want to identify the areas in which AI can add maximum business value. Perhaps it’s finalizing the underlying technology stack.
There is no general-purpose AI, and that’s a frustrating reality for organizations. Getting up to speed on AI is a journey rather than a sequential maturity curve for good reason: Every application and use case requires different tools and algorithms. The chatbot for a financial service provider’s IT help desk, for example, can’t be applied to the company’s home-loan call center. An algorithm trained in pharma to read documents for adverse event recognition can’t be reused in a banking context to recognize anomalies in mortgage applications, even though the technology stack and technique may be the same.
With AI, each use case requires a unique training process as the system learns the relevant patterns. Moreover, AI systems take much longer to learn some tasks than others. For example, neural network-powered computer vision requires extensive training and data sets to recognize and analyze patterns in images.
AI is also a very different experience from other digital exploits. While your organization can scale its learning of cloud and analytics, AI requires a fresh look at existing approaches to help take advantage of new techniques, different data sets and accelerating advances in core technologies.
Getting to Outcomes
To ensure your company’s AI efforts are smooth and achieve their desired business outcomes, we recommend the “Five E” process:
With AI on every organization’s 2018 to-do list, companies are attending conferences and hosting workshops. Despite the buzz, however, confusion remains over even the basics of AI, which requires its own literacy. For example, how does an advanced form of AI such as machine learning (ML) differ from other forms of AI?
In addition to such distinctions, it’s important to understand that AI forms a continuum: There is no start or end, and it’s the combination of tools and techniques, applied to the right business problems and processes, that will deliver personalized experiences with efficiency and scale.
Success with AI depends on an organizational openness to discovering new and at times unexpected business needs. When a healthcare client conducted a pilot in natural language processing (NLP) for more efficient review of social workers’ and physicians’ notes, it discovered that factors such as economics and access to transportation significantly affect the health of 12,000 patients in a pilot market — yet standard documentation practices include no fields for social factors.
While many organizations look forward to experimenting with AI as they have with digital tools and technologies, AI differs in several areas. It requires return on investment (ROI)-driven outcomes, specific setup and support, and increased tolerance for failure.
It’s easy for organizations to get lost here. Determining whether a pilot has produced definitive results is tricky, as is the question of whether to extend a pilot for further iteration or the acquisition of additional data sets. For example, a client that provides credit-card services for small and medium businesses (SMBs) discovered that while it typically segments customers by industry and revenue, a more telling metric is whether the founder is still involved: SMBs in which the original owners remain hands-on often have little time to evaluate new financial products. The client’s next step is to determine whether the additional campaign’s ROI will offset the costs. The lesson? Be willing to stop the pilot if the cost-benefit doesn’t work out.
Given AI’s growing profile, it’s common to find multiple business units within an organization — operations, technology, and lines of business — each pursuing its own initiatives. Use cases abound, and prioritization is a challenge. Which pilots share a common AI core that all functions can leverage? Which ones can the larger company learn from? The end goal is to establish AI as a capability that the larger organization can embrace more effectively. CIOs and business leaders who want to encourage AI experimentation while imposing order and discipline on the process can ask several questions to determine which AI projects to greenlight. Always start with business value. Does a proposed project deliver limited, incremental value, or is it reinventing a process through clever use of data and technology? Early successes that feature positive business benefits such as top- or bottom-line boosts or productivity improvements help organizations embrace AI faster than those that are technologically possible but have limited value.
In the exploration phase, organizations are typically deciding whether to continue to focus their AI efforts within a single functional area, or to apply it more broadly across the company. Many are also watchful of unfolding government regulations regarding compliance and liability, as legislative and judicial branches tackle AI-related challenges. The important part of this step is for companies to consider how they can better organize themselves around AI. What processes can they create that are useful forapplying their AI learnings to other parts of the organization?
There is no utility AI, so every company must chart its own path to success. While each route is different, the sequence remains consistent.