Hello everyone, My name is Kenshin Fujiwara and I am the CEO and founder of HACARUS Inc.
Through this series of blogs, I will discuss a wide range of topics of AI, from the history of AI to practical tips for successful application of AI projects. I hope my blog posts will help you gain a better understanding of AI and solve your business issues.
Using the knowledge we have gained so far, today, we will discuss the process and key points for successfully implementing an AI project. The process, beginning from the review stage, through the operation management, is divided into five major steps as shown as below.
Step 1: Problem Identification
Step 2: Proof of Concept (POC)
Step 3: Trial Operation (Pilot)
Step 4: Full Implementation
Step 5: Operation Management
At this point, it is also worth noting that AI development rarely goes as planned, and is achieved through repeated trial and error. For this reason, it is not uncommon to run a POC multiple times or to revise a POC to resolve issues that come up during a trial. Even after the AI system is fully implemented and the plan enters the operation management phase, the AI itself is often updated as time goes on. This additional training is utilized if the nature of the target data changes.
I’d like to introduce each of these steps in greater detail through the next following blogs. First, let’s take a look at the problem identification step, especially through three case studies.
Step 1: Problem Identification
From the beginning, identifying the problem or setting tasks is an important step for AI implementation. If the project begins with ambiguity in these areas, the probability of failure is significantly higher. As mentioned earlier, you should set a general goal for the AI project in order to solve a current problem. By first identifying the problem, you can better define the specifications and requirements during the AI development.
As an example, set a major goal at the beginning such as “to eliminate the problem of labor shortages.” Other examples may include reducing on-site workloads or accelerating research and development speeds. Next, I will use three case studies to illustrate this point further.
Case 1: Solving Labor Shortages
At some point, skilled workers will reach retirement age and leave the company. During this process, it will take time to train new workers to fill the void. Looking at Japan specifically, due to the declining birth rate, there is no guarantee that the company will be able to hire talented individuals in the future. In this case, we want to prepare for this shortage of labor by digitizing their knowledge into data using AI.
Case 2: Reducing On-site Workloads
In the medical field, diagnostic imaging is becoming increasingly important. As a result, the workload of doctors specializing in reading CT, MRI, and other imaging technologies is increasing. The practice of diagnosing and determining the presence or absence of disease often leads to frequent overtime work for these specialists. In order to reduce the burden on physicians by introducing AI-based diagnostic imaging technology and incorporating it into the initial decision-making process.
Case 3: Accelerating the Rate of Research and Development
In the field of drug discovery research, pharmaceutical companies need to test a massive number of drug combinations. Currently, it takes over ten minutes to test a single drug candidate. By introducing AI into this process, we can drastically speed up the testing process and save valuable time.
PDCA (Plan-Do-Check-Act) Cycle
After identifying the initial problems and defining the project requirements and specifications, we can proceed to the proof of concept stage. I will describe this stage in more detail later, but it is also important to remember that we will continue repeating small trials from this point and we call this the PDCA (Plan-Do-Check-Act) cycle.
Even though the PDCA cycle is ideal, we don’t have unlimited time and money to keep running new trials. Therefore, it is a good idea to also consider the criteria for when to stop the trials. At the same time, it is also important to work together with the people who will actually introduce and work with the AI and consult with them. Failure to incorporate the opinions of frontline workers might lead to a divided workforce.
Of course, not all top-down AI implementation projects lead to failure. The most important factor for success is that the department in charge of the project has well-defined tasks and goals. However, in many cases, the probability of success is higher when a field-driven problem-based approach is taken.
In the next blog, I will I continue to discuss other important factors in the step of problem identification. For your updated knowledge and insight about AI technology, subscribe to our newsletter or visit HACARUS website https://hacarus.com.