The goal of artificial intelligence (AI) for an enterprise is to streamline business processes and improve customer experience to drive positive bottom-line results. It’s about moving enterprises into the on-demand economy by finding that magical balance of increasing customer satisfaction while reducing costs. But artificial intelligence is no silver bullet that works on its own. It’s an integral piece in a broader end-to-end solution that combines the best of technology with the learned behaviors of skilled humans. And the challenge of enterprise AI is finding a solution or approach that provides measurable results.
There’s so much hype around artificial intelligence that it’s easy for enterprises to get lost amid the buzz. Many enterprises acknowledge they need to do something with AI, but they don’t know exactly what. This usually results in businesses deep diving into large-scale AI and automation programs without a thorough strategy for success. This approach is costly and relies on the artificial intelligence to learn, grow and perform the required business processes instantaneously. But, unfortunately, we’re not yet at the stage where AI is advanced enough to perform in such a way.
What’s needed is a totally different approach that doesn’t place all the burden on the technology to work miracles and automate complex business processes upon launch.
Enter the pragmatic approach to AI.
Pragmatic AI leverages a systematic implementation that gradually increases automation and machine learning. In the context of enterprise implementation, there are three phases to Pragmatic AI. At the end of these three phases, the goal is to have an artificial intelligence program that learns and develops on its own. However, before we get to that stage, we must focus on highly repeatable business processes that can easily be automated.
It’s important to note that pragmatic AI fits within a larger digital transformation strategy. AI is by no means a silver bullet that solves all organizational problems on its own. Rather, it augments processes and complements technology that are already in place across the organization. Contrary to the beliefs of many industry pundits, AI is still not advanced enough to work on its own. It needs human management, guiding and careful development.
“Long Tail” and “Fat Tail” Processes
Central to Pragmatic AI is the delineation between “fat tail” and “long tail” processes. Processes that are generic, repetitive and common are classified as fat tail processes. Examples include basic customer service inquiries (technical assistance, account updates, etc.) or bill processing. In contrast, long tail processes are less frequent and require a high-degree of customization to resolve.
The foundation of pragmatic AI is focusing first on solving the fat tail processes before progressing to the complex long tail issues.
The 3 Phases of Pragmatic AI
Phase 1: Automate Repetitive Processes
The first phase of Pragmatic AI is analyzing all the fat tail processes and identifying those that can be easily automated by deploying an AI program. The benefit of starting with these fat tail processes is that there is already a vast amount of data available to leverage for faster engineering. This data means there is less need for the program to “learn” on its own. Instead, it’s programmed, or “taught,” with specific business processes in the development phase so that it’s already functional upon deployment.
Phase 1 includes a combination of Natural Language Processing (NLP) with guided user journeys.
Consider that fat tail accounts for nearly 80% of customer inquiries, meaning the vast majority of inquiries are repetitive and highly common. With this in mind, you can get a sense of the cost reductions and the potential for bottom-line efficiencies when this fat tail of inquiries is automated. All the while, live agents and operators can still be used for the remaining 20% of customer inquiries, the infrequent long tail processes and a safety net for the AI.
With this combination of AI and live agents, enterprises can enjoy the benefits of having always-on customer support without the additional cost of resources, thus delivering a responsive, on-demand and highly effective customer experience that drives satisfaction.
Phase 1 of Pragmatic AI is often sufficient for both small- to medium-sized businesses and large organizations. The significant cost reductions and increases in customer satisfaction can provide significant ROI. Phases 2 and 3 expand the automation and include greater sophistication and more complex machine learning.
Phase 2: Add Advanced Cognitive Services
With AI already implemented to manage the repetitive, high-volume processes, Phase 2 moves further down the tail, addressing more specific processes. More advanced cognitive services are implemented to augment and widen AI’s capacity by analyzing the interactions and identifying patterns to be replicated. These additional services include language translation, image processing and video processing.
Key to success in Phase 2 is the use of analytics to improve the automated processes established earlier. Improvements are made to failure points — i.e. when a bot has to transfer to a live agent for resolution — and fixed to increase the bot’s success rate. Gradually, the AI’s capacity to solve complex problems grows, increasing the utility for the end-user. In addition, NLP is improved to provide more freedom for people to interact with the bot using natural language.
Phase 3: Building the Framework for Machine Learning
The final stage of pragmatic AI is enabling an element of machine learning and growth. Leveraging a combination of cognitive services and pre-programmed business processes, the AI-powered bots learn from their interactions and adjust flows under the supervision of human operators. The machine learning is used to improve existing processes or to train the machine regarding new processes that are currently being managed by humans.
Again, NLP is increased, allowing the bot to comprehend and process a wider degree of words and terms. At this point, the AI should be at a level of sophistication that the need to switch to live agents is less often. Throughout the three phases, the majority of failure points have been addressed and used to train the AI.
The three-phase approach to pragmatic AI allows automation to be developed and deployed using a strategic process that’s unique to an organization. Each phase directly addresses specific challenges within an organization.
See also: Robots and AI—It’s Just the Beginning
Why Pragmatic AI?
Currently, there’s a gap between what enterprises expect of AI and what the AI can actually do. Instead of letting this gap drive costs higher through ill-fated AI programs, enterprises need to adopt a practical approach. The journey to enterprise AI is about pursuing what’s practical with clear ROI and readily available now and then building upon this foundation with large-scale, sophisticated automation and machine learning programs.
Pragmatic AI’s strength is that it is embedded and grows from current business processes and that it can immediately deliver a positive ROI by increasing customer satisfaction while reducing costs. With these kinds of measurable results, AI programs move from a place of aspiration to an actionable strategy — and finally to a readily deployable solution.