The 5 Steps for Proceeding with an AI Project: How to Successfully and Effectively Manage AI Projects (Part 3)

The 5 Steps For Proceeding With An AI Project: How To Successfully And Effectively Manage AI Projects (Part 3)

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.

In the last blog, we studied essential considerations when implementing issue setting, such as company-wide cost-effectiveness and evaluation indicators. Today, I will introduce next steps for proceeding with an AI project. Let’s start with step 2. Proof of concept (POC).

Step 2: Proof of Concept (POC)

Once the targets have been set and the AI project has started, the next step is the proof of concept. This term doesn’t just apply to the world of AI, but also for the implementation of ordinary systems. 

In this stage, project verification and demonstrations are conducted prior to the trial implementation. The Proof of Concept (POC) acts as a test to see if the project is likely to be a success or failure. For AI, it is a preliminary exercise where AI is programmed and tested.

To begin the process, a temporary AI is run to verify whether the target performance can be achieved. However, there are many aspects that cannot be properly understood until sample data is collected and fed into a full-scale AI model.

However, we cannot progress towards building a full-scale program for production operations. Instead, we can proceed with the project while validating its functionality by setting up a POC.

For this phase, the time schedule is on a case-by-case basis, but it generally takes several months. It is also common to repeat the trial-and-error process several times during this step.

Analyzing Suspected Tumor Sites

One POC case study I would like to share is an AI project for analyzing suspected tumor sites using MRI machine data. The main objective of this project was to provide diagnostic support to inexperienced physicians that matched the quality of experienced physicians. 

The first step in this POC process is to collect data to be read by the AI model. Using the medical images required during actual MRI exams, this data will be analyzed and formatted for the model. The AI is then trained using methods such as deep learning to measure the performance of the tumor screening. 

While observing the results, we tried various approaches to improve performance. This included adding more data for training, changing the data processing method, and changing the settings for the AI training. This work is carried out over a set period of time to achieve the POC targets set earlier.  

The Reasons of POC Deaths

Unfortunately, in the process of actually proceeding with an implementation project, there are many cases where the project is terminated at the POC stage. For failure cases like this, where the project never goes into production, we refer to them as POC deaths. 

There are many reasons that lead to POC deaths. In some cases, the vendor company’s technical capabilities aren’t able to keep up with the client’s problem setting. In other cases, there have been problems with the quality or quantity of the data collected.

At the beginning of this section, I mentioned that the POC stage is the basis for determining if the AI implementation project is viable or not. For this reason, it is unavoidable to abort the project at this stage if satisfactory results are not obtained. 

Step 3: Pilot Testing

If the POC phase is the building of a provisional AI and verifying its performance, the pilot phase is the integration of an actual system and conducting tests. For example, in the case of the previous medical AI for MRI diagnostic imaging, the pilot testing is conducted by actually integrating the AI into the hospital’s system. It is then run so that doctors can use it on-site to test its functionality.  

Integration in Manufacturing Field 

Looking at a different industry, in the case of inspection-based AI in the manufacturing field, the pilot would be integrated into the workflow on-site. The results of the system’s inspection performance would then be discussed with the field personnel. 

Since at this point, the integration is only for a trial system, it will be kept to a small scale and multi-facility introduction should be avoided. With a small number of test sites, it is also possible to receive important feedback from the field. 

Although this is just a trial run, the system is still being implemented into the company’s actual operations. During this time, it is key to work closely together with the IT department to ensure a smooth test run. 

In the next blog, I will introduce the following steps for proceeding with an AI Project from step 4. full-scale implementation and step 5. project operation. For your updated knowledge and insight about AI technology, subscribe to our newsletter or visit HACARUS website

Subscribe to our newsletter

Click here to sign up