I am in charge of Cinnamon AI public relations.

After the current coronavirus pandemic, there have been major changes in the way people work in Japan. I believe that various companies are currently moving forward with measures to transition to remote work and introduce internal systems at a rapid pace. Under these circumstances, the reality is that many companies have trouble implementing AI in-house because there are many differences from traditional system development.

In this article, we will introduce three points that can cause problems during the consideration stage when introducing AI.

1 Purpose of introduction is not clear

There are cases where you end up considering the introduction of a product based on news about another company's introduction. Once we start considering the possibility of introducing AI,``Actually, there wasn't that much need in the field.was easy to implement”As a result, the implementation of"In the end, the work has to be done by humans, which means you have to work twice with AI." "I tried introducing it, but it's not being used in the field!"That's what happens.

Therefore, it is necessary to think about which department and what kind of work has a heavy workload. In some cases, efficiency can be improved even by creating a system that does not use AI, so it is necessary to be careful not to consider "AI-based".

2 Requirements include things that AI cannot do

If we have too high expectations for AI, we may end up trying to accomplish things that even AI cannot do. In many cases,It is difficult for AI to perform tasks that exceed the quality achieved by humans.is the reality. (There are exceptions that sometimes make the news.)

In many cases, AI is not as good as humans, but it is possible to improve overall productivity by incorporating AI and reviewing the entire business process as long as the workforce remains. It is necessary to understand the characteristics of AI (e.g., it is good at repetitive tasks, can work according to rules, etc.), review the entire business flow on the assumption that it is not perfect, and consider redesigning it.

3 Setting an unrealistic schedule

When trying to introduce a system on a large scale, there are cases where it becomes difficult to judge whether the development was successful after PoC. AI development is different from traditional system development.It is possible that the accuracy of the developed product may not be as expected. Also, it is not uncommon for the development period to take 2-3 months.. If the subject of verification is too large, it is likely that large-scale AI development may not go well in the first place, and sometimes it may be affected by surrounding systems, making it difficult to see the results of the AI itself.

When developing AI, it is important to start with small and easy-to-understand tests so that the effects of the AI part can be clearly measured.

summary

 ■Clarify the purpose of the introduction and consider whether it will improve work efficiency.

 ■Understand the characteristics of AI and consider changing the overall operation

 ■Proceed with a development scope and schedule that allows evaluation of the AI part

This time, we have introduced the basics of considerations necessary for a successful AI project.

When making a proposal to a customer using Cinnamon AI, a consultant visits the customer, selects the target business together, formulates a development roadmap, and confirms that the points introduced today are clear. While I'm at it, I'm starting PJ.

When you are in charge of introducing AI, please keep the above points in mind.

(Written by: Morita)