Unlocking the Power of AI: A Practical 5-Step Guide to Implementation
Transitioning from a desire to utilize AI to a practical and sustainable approach for solving business problems and managing expectations does not require extraordinary abilities.
Erick Brethenoux, a Distinguished VP Analyst at Gartner, Inc., specializing in machine learning, artificial intelligence, and applied cognitive computing, along with Frances Karamouzis, provide valuable insights in their analysis article published by Gartner on “How To Implement AI in 5 Practical Steps.”
According to Gartner, by 2020, 50% of organizations lacked the necessary skills in artificial intelligence (AI) and data literacy to derive business value. Furthermore, by 2023, 80% of digital business industry visions will rely on AI adopted from AI industry use cases.
Introduction
Developing an artificial intelligence (AI) strategy without first evaluating the organization’s readiness to adopt AI techniques is akin to devising a battle plan without knowing if troops have undergone training and preparation (unknown skills), lacking intelligence on enemy movements and capabilities (unknown data), having no knowledge of available weapon systems (unknown technology), and lacking understanding of the objectives (unknown goals).
The practical implementation of AI techniques within organizations of any size can be achieved through the following five steps:
This five-step approach offers a tactical perspective on implementing AI techniques, focusing on achieving quick time-to-value. It does not encompass a long-term strategic outlook, which can be developed once the organization has assessed its current strengths and weaknesses, both culturally and technologically, in terms of utilizing these techniques effectively.
Analysis
Step 1. Build a Portfolio of Impactful, Measurable, and Quickly Solvable Use Cases
The primary objective of the first step is to focus on measurable outcomes. Similar to successful AI and machine learning (ML) initiatives, it is crucial to start with a well-defined answer, one that is as comprehensive as possible. Each project leveraging AI techniques should begin with a clear understanding of its final business impact. Stakeholders from various lines of business (LOB) should be able to articulate the concrete business benefits they expect to derive from AI. While these expectations may change or evolve over time, establishing them firmly at the outset is important.
In a recent Gartner survey, respondents indicated plans for a significant increase in deployed AI projects. On average, organizations plan to go from five new projects in the upcoming year to nine and then 13 projects in the subsequent years. Gartner’s CIO survey also revealed that approximately 14% of CIOs have already implemented AI solutions, while 23% plan to do so within the next 12 months.
Three principles are crucial for this foundational step:
Selecting Use Cases
There are various techniques available to establish a small portfolio of use cases. Gartner provides guiding principles and practical examples for organizations to prioritize and select the most promising areas where AI techniques can be leveraged. The toolkit titled “How to Select and Prioritize AI Use Cases Using Real Domain and Industry Examples” and the Gartner Trend Insight Report “AI Use Cases, Tales From the Trenches” offer valuable resources in this regard.
Simple tools like the impact versus feasibility matrix (Figure 2) can aid in efficiently and clearly prioritizing use cases.
In selecting use cases, another important consideration should be time. Despite the vast number of factors to consider, a simple rule-of-thumb has proven effective in many proof-of-concepts (POCs): Think big, start small, act fast. The idea is to identify a critical problem for the organization, one that has not been successfully addressed by other techniques (think big). However, it is essential to scope the problem in a way that can be accomplished within a nine-week timeframe (start small). Finally, it is crucial to begin iterating and executing the project quickly to uncover any issues promptly and make necessary adjustments (act fast).
Describe Use Cases Clearly
Even for small and rapid POCs, and particularly when employing AI techniques, it is essential to articulate the value propositions as clearly as possible. A value proposition canvas, as depicted in Figure 3, can serve as a valuable tool for CIOs to assess business outcomes and identify potential barriers transparently.
Do Not Compromise on Established Metrics
To establish an appropriate set of metrics, the Gartner Business Value Model can be utilized. This model, explained in “The Gartner Business Value Model: A Framework for Measuring Business Performance,” provides a structured framework and definition of nonaccounting metrics. It can be applied universally to assist organizations in identifying the impact of their business activities on financial performance.
By creating concrete and measurable metrics, organizations can effectively apply three types of value to the AI techniques they intend to implement:
Refer to “Data and Analytics Strategies Need More-Concrete Metrics of Success” for further details.
However, it’s important to note that due to the unconventional nature of AI techniques, early adopters should maintain an open mind as metrics may change or evolve during the exploration and advancement of these techniques.
For additional information, consult the following resources:
Step 2. Assemble a Set of Talents Pertinent to the Use Cases to Be Solved
Depending on the prioritized use cases, three personas can provide an ideal balance to initiate AI efforts:
领英推荐
These personas form a complementary triumvirate of AI, IT, and domain expertise. However, within each area of expertise, a range of skilled individuals will be pivotal in solving the proposed use cases (refer to Figure 4 for a visual representation).
AI Thrives Through Data Literacy
Contrary to popular belief, AI skills are not necessarily scarce, expensive, or mysterious. Individuals with a curious mindset, such as database administrators or mathematically inclined data engineers, can become excellent data scientists without needing a Ph.D. or even Python skills. While this notion may challenge the prevailing trends in the machine learning (ML) world, the reality is that valuable predictive analytics models can be built without writing a single line of code. However, a solid understanding of ML principles, statistics, and a passionate curiosity for data exploration and manipulation are essential.
Collaboration skills are equally important. Successful AI initiatives are more likely to be achieved through tight collaboration among individuals with expertise in different areas, rather than relying solely on deep knowledge from a single person with unrealistic expectations. Motivated, open-minded, competent, and focused champions, often emerging from within the organization, prove to be a better recipe for successful proof of concepts (POCs).
Besides the business benefits, early POCs also provide valuable insights into the organization’s readiness to adopt AI techniques. Initial skills can be improved through upskilling programs, such as local academic programs offering certificates, or through online courses on platforms like Coursera, Udemy, or DataCamp. However, to maintain competitiveness and ensure scalable progress, enterprises will eventually need to pursue a more comprehensive educational program.
Gartner’s research, exemplified by “Artificial Intelligence Demands That CIOs Foster a Data-Literate Society,” highlights the importance of fluency in “data” as a critical capability in the digital society. While executives and professionals may be well-versed in the people, process, and technology aspects of business change, the ability to “speak data” fluently is increasingly becoming a crucial skill (refer to Figure 5 for further details).
Step 3. Gather the Appropriate Data Relevant to the Selected Use Cases
One common misconception in early AI exploration is the belief that massive amounts of data are required to build successful AI models. In reality, many use cases can be accomplished using a reasonable amount of data, as long as the dataset is of high quality, including factors such as normalization, completeness, and diversity. While a lack of volume can be compensated for by reducing the project scope, poor data quality inevitably leads to failure in proof of concept (POC) efforts. Data-driven techniques, like machine learning (ML), heavily rely on data to generate insights, making data quality issues particularly acute throughout the ML life cycle.
Refer to Figure 6 for a detailed overview of the data quality challenges involved. It’s important to recognize that the processes in place to ensure data quality will require iterative improvement.
However, the definition of a “reasonable amount of data” can vary depending on the selected AI techniques. ML techniques typically necessitate more data compared to logic-based or optimization techniques. The same holds true for natural language processing (NLP) systems, whether they utilize subsymbolic techniques (ML-based) or traditional symbolic approaches (linguistics). The skills identified in Step 2 will play a crucial role in determining the appropriate amount and type of data needed for the chosen techniques.
Data Pipelines
In addition to focusing on data quality and completeness during the data gathering process, CIOs should also consider the sustainability of the data. This involves understanding the nature of the data sources, such as whether they are anecdotal or systematic, partial or complete, discrete or continuous. Furthermore, it is essential to determine the frequency at which the data can be obtained, whether it’s in subsecond intervals, daily, weekly, and so on. Considering these factors is crucial for assessing the scalability of the proof of concept (POC) and can be facilitated through the implementation of data pipelines. Detailed information on this practice can be found in “Enabling IoT Data Pipelines for Machine Learning Inference.”
Refer to the following resources for additional insights:
Step 4. Select the AI Techniques Linked to the Use Cases, the Skills, and the Data
Many AI techniques have been in existence for decades and are mature in their development. These techniques are suited for specific problems, types of data, and skill sets. For instance, probabilistic reasoning techniques like machine learning (ML) are effective in uncovering hidden patterns within large datasets, such as fraud patterns, churn issues, or risk variables. However, ML techniques require a sufficient quantity of high-quality data, along with analysts who possess knowledge of analytical mechanisms and algorithms.
On the other hand, optimization techniques are well-suited for tasks like finding an optimal balance in inventories, optimizing routes within a supply chain, or generating workable plans under multiple constraints and time limitations, such as managing landing, gate, and crew assignments at an international airport during a snowstorm. These techniques rely on talents in operations research and the ability to gather data appropriately, particularly when considering the operationalization of models in production.
There are various AI techniques available (refer to Figure 7), some of which may already be integrated into enterprise solutions (e.g., SalesForce, SAP, Oracle), decision-modeling solutions (e.g., FICO, Enova Decisions, ACTICO), or standalone platforms (e.g., RapidMiner, KNIME, FlexRule, Decisions, Gurobi, Frontline Systems, Attivio, Narrative Science, Maana, Grakn, Swarm Technology, Matroid, Deepomatic, PROPHESY). Open-source libraries and platforms also offer the possibility of leveraging these techniques.
Step 5. Structure the Expertise and the Accumulated AI Know-How
In the context of AI adoption, it is crucial to consider not only the AI techniques themselves but also the surrounding technology ecosystem. The success of a proof of concept (POC) relies on several other technology factors that should be addressed at this stage. These include the IT infrastructure, which encompasses the resources available for algorithm training and integration with target systems, as well as existing process automation software where AI models will be integrated. Additionally, the availability of adequate user interfaces to interact with these models is another important consideration.
Refer to the following resources for more information:
AI techniques, despite their potential, have sometimes failed to deliver a satisfactory return on investment (ROI), leading CEOs to become skeptical about the money invested. By executing POCs on various business problems across the enterprise, organizations can identify gaps in skills, data, technology, culture, readiness, and general education related to AI. It is also important to consider the different levels at which AI techniques can be leveraged and how they can complement human tasks, as the level of adoption may vary from one department to another.
AI competence can be found in various areas of the business, from line-of-business (LOB) units to the IT department or other functional departments. Typically, organizations tend to consolidate their AI skills into competency centers after going through multiple experiments with AI techniques. A common organizational model that emerges is the establishment of a separate “AI lab” that operates independently of both LOBs and the IT department. This lab usually reports to a neutral corporate function, enabling close ties with the business while staying aligned with technical capabilities, investments, and overall strategy. AI experts within the lab may also collaborate closely with SMEs and IT professionals in LOBs, working on ongoing projects to foster cross-pollination of AI capabilities. This proximity facilitates the generation of serendipitous ideas from both technical and business stakeholders.
For a visual representation, please refer to Figure 8.
If you are interested in this article, you are welcome to join the Tikflow RPA?Discord community, where you can engage with a large community of technology enthusiasts and discuss cutting-edge information.