Whether you are an early adopter or someone who merely follows technology advances through media, it is abundantly clear that we are in the midst of a significant wave. Google Search trends for 'artificial intelligence' (AI) and 'machine learning' (ML) provide validation for this view. Today, every aspect of our lives as consumers is influenced in some way by advancements in ML and AI.
As businesses seek to leverage these advancements to drive commercial growth and efficiencies, they recognize a journey lies ahead. Possibly the best example of this journey is the evolution of AI in vehicles. Features like blind spot detection, lane departure warning, and front collision warning were necessary prerequisites for self-driving capabilities. However, as recent news reports demonstrate, there is much to be done before self-driven vehicles become the norm.
With this backdrop, we examine eight elements that will help businesses assess their ML/AI maturity, as well as manage their adoption journey.
1. Stakeholder Engagement
Needless to say, ML/AI investments should begin with securing the buy-in from the impacted business process stakeholders. Who better understands the decision-making process and the opportunities to 'automate', 'assist' or 'augment' through ML/AI than the process stakeholders? It is not uncommon for us to see organizations dive into ML/AI, while the impacted stakeholders sit disengaged on the sidelines. They are either not actively participating in defining opportunities or worse still, are debating the longevity of their jobs within the organization. The latter should be a non-issue if the strategic objective driving the ML/AI investments is profitable business growth.
2. Capability Cycles
Academic origins for ML/AI date back at least four decades prior to 1997, when IBM's Deep Blue machine defeated then the reigning chess champion, Gary Kasparov. While it has taken about half that time for it to penetrate our everyday lives, it helps illustrate that advancements come in waves. Two of the most common reasons for the oft-repeated "ML/AI hype" are: [a] an inability to see the limitations of the current ML/AI wave, and [b] a short-sighted vision for the longer-term transformation that future ML/AI waves could drive. In contrast, businesses that are successful in disrupting the norm, see discrete capability cycles within their ML/AI journey. An appreciation for what can and should be targeted in each cycle is critical to continued success.
3. Data and Model Governance
Data is central to the success of ML/AI adoption. With that comes an understanding of the role of data strategy, data architecture, data management, data security, data privacy and permissible use. These facets have been brought front and center in the recent weeks with the potential non-permissible use of social media data. As new data sources emerge over time and, existing data sources and ML/AI solutions multiply in volume, a strong stewardship and governance practice becomes an essential ingredient to balancing the risks of ML/AI with its rewards. Further, all analytic and ML/AI models developed and deployed to support business processes also require appropriate governance controls to ensure compliance with operational or regulatory requirements.
4. Data Architecture
As an extension of the Data Governance point, the need for a mature Enterprise Data Architecture is a natural foundational step. Emerging data sources, various data types/formats, sensor and IoT data streams together with a push towards near real-time decision-making capabilities, are all calling for a rethinking of enterprise data architecture. We recommend following an agile strategy in regard to designing your data architecture. However, sooner than later, a sufficiently robust 'baseline' enterprise architecture needs to emerge from the agile work streams, that will catapult the organization forward in their ML/AI journey.
5. Operational Focus
Selecting the right set of tools and technologies is essential to your journey attaining its operational objectives. There is no dearth of ML/AI vendors and solutions for every business process. Which solution is right for your operational business needs and can at the same time evolve with your business? We encourage our clients to pursue the path of open-source centered proof-of-value before investing in commercial off-the-shelf (COTS) solutions. Open-source packages also provide businesses the opportunity to set a benchmark for any COTS investments in the future. Open-source does not have to become your operational solution, although it is certainly up to the task for many of the more commonplace ML/AI applications. While the total cost of ownership should be a consideration, we recommend balancing it with the solution's ability to adapt and evolve with your organization's needs.
6. Feedback For Learning
If your business is actively pursuing ML/AI opportunities, it has quite likely moved past 'Descriptive' and 'Diagnostic' analytic capabilities. As companies venture into the 'Predictive', 'Prescriptive' and 'Cognitive' analytic capabilities, we encourage distinguishing your needs from your wants. Learning solutions require a robust mechanism to record the feedback from decisions taken and effects, outcomes or responses from those decisions. You may also consider setting up A/B testing strategies, so that your models can draw from a broad set of experiences to learn. Too often we see organizations trying to reach for the 'Cognitive' capabilities without the necessary elements to capture the feedback and support the learning mechanism.
7. Complementary Skillsets
Whether you decide to hire the appropriate skill sets into your organization or outsource to a services delivery consultancy, having the right mix of skill sets on your team is essential. ML/AI requires a healthy mix of mathematical/statistical genius, programming prowess, parallel computation know-how, data wrangling skills and business acumen. Finding all of these skills at the right levels of competency in a single or even a select few individuals is akin to seeking a unicorn. Businesses have to be willing to invest, bringing together the talent and/or training up proven resources to be successful with ML/AI. In addition, a clear charter, organizational structure and career development pathways are essential ingredients for an ML/AI center of excellence.
8. Continuous Improvement
Last but not least, a culture that recognizes the importance of continuous improvement holds the key to disrupting the norm. Continuous improvement must span all seven of the preceding facets and when appropriate reach outside of these as well. Expanding on the latter point by building on the self-driving vehicles example provided earlier; there are a number of ethics questions that have not been touched on in regard to self-driving vehicles. For example, should self-driving vehicles take decisions that benefit the vehicle occupants or equally benefit occupants, pedestrians and other vehicles in its path? Further, who is responsible when a self-driving vehicle is involved in a fatal accident: the vehicle owner, the manufacturer, the ML/AI software provider or all three? Continuous improvement should not draw the line at the data and embedded ML/AI algorithms attaining maturity. There are and will be a host of related questions that must be asked and answered in the context of a truly mature ML/AI solution.
Follow us through this series of posts as we dive deeper into each of these facets to illustrate our methodology for assessing your ML/AI maturity. We will provide a yardstick against each of the eight facets that will assist with measuring your ML/AI maturity and help direct your ML/AI strategy and investments.
By Brian Monteiro - Chief Data Scientist & Director, Analytics Practice at WorldLink