Thursday, May 23, 2024
HomeTechnologyThe era of high unit prices for AI is over | To...

The era of high unit prices for AI is over | To an era where AI can be provided for 1.5 million yen

As the use of AI increased, the number of vendor companies developing AI surged in the late 2010s. However, just because the number of vendor companies has increased does not mean that the use of AI will advance in the world.

That’s because the “unit price” of the development project was high. When outsourcing AI development to a vendor company, various costs such as consulting, development, and verification of effects are added, and depending on the number of engineers, there are cases where the cost exceeds 10 million yen.

KICONIA WORKS Co. , Ltd. (hereafter, KICONIA WORKS), which provides application and system development services that incorporate technologies such as machine learning according to customer issues, will launch a service to build a demand prediction AI for 1.5 million yen in 2021. Released on May 18th .

This time, we interviewed Mr. Takuro Shogami, the representative director of KICONIA WORKS , about the company’s internal system, why it can be built at a low price of 1.5 million yen.

Contents

  • AI development market with a market value of several million to tens of millions
  • Company structure that can provide demand prediction AI construction for 1.5 million yen
    • Cost savings from specialists
    • Discoveries from in-house performance
  • In the future, we will incorporate the latest algorithms and expand the range of support
  • lastly

AI development market with a market value of several million to tens of millions

ーーHow do you feel the unit price of AI projects is changing? Please let me know because it doesn’t matter how you feel.

Mr. Kakigami

From around 2015 to 2018, the unit price for a machine learning project (PoC) was 10 to 50 million yen , and if you asked a major vendor, it would be close to 100 million yen.
In fact, when I hear stories from customers who requested KICONIA WORKS, I hear that they invested more than 50 million yen and failed. From around the end of 2018, I think that the number of development vendors who can handle it for about 5 million to 8 million yen has increased.KICONIA WORKS is also a company that was launched in 2018, but it is often carried out for about 3 million to 6 million yen. Among them, the unit price is still in the tens of millions of yen, so I think there are many development vendors who are suffering from not being able to receive orders.

Behind this is the emergence of various tools and open sources that make machine learning projects easier, the movement to learn AI development in -house , and I think there is a background that we started AI development .

Company structure that can provide demand prediction AI construction for 1.5 million yen

ーーWhy can you provide AI for around 1.5 million yen?

Mr. Kakigami

As I mentioned earlier, KICONIA WORKS has been supporting PoC of machine learning model building in about 2 to 4 months for about 3 to 6 million yen.
After completing the project, we conducted a questionnaire survey with the customer, and we constantly made improvements, wondering if the project could have progressed faster and better within the company, and if there was a better algorithm in the world.And, as a result of handling more than 30 AI-related projects a year, and more than 150 projects including the career experience of individual members, we have the know-how to implement the latest algorithms and the optimal feature vectors. etc. have accumulated, making it possible to build the best machine learning models in a short period of time.

In particular, there is a need for demand forecasting, which we are dealing with this time, and we have implemented many projects.

The price of 1.5 million yen has been realized through the efforts of our customers who have grown while doing projects together, and the efforts of KICONIA WORKS.

Cost savings from specialists

ーーIn KICONIA WORKS, how have the human costs and calculation costs for individual projects changed? And why?

Mr. Kakigami

At the time of our founding, about three project managers, data scientists , and engineers were involved in the AI ​​development of demand forecasting .
Although not fully committed, these members will work for 2-3 months, so the human cost within KICONIA WORKS accounted for the majority of the project cost. In addition, computational costs were incurred due to the calculation of huge amounts of data on the cloud.However, we have been able to reduce the calculation cost by investing the profits earned in business management into in-house server expenses and calculating to some extent in-house without using the cloud .

Furthermore, by repeating improvements and automation, it has become possible for two people, a project manager and a machine learning engineer, to respond in about two months. I made it.

Discoveries from in-house performance

ーーHow do projects become more efficient through the accumulation of in-house achievements?

Mr. Kakigami

First of all, we have accumulated know-how through our achievements, and our key points have improved, so we are now able to efficiently research and implement algorithms and feature vectors .
Furthermore, we have incorporated various data and algorithms from Kaggle and the world in addition to our in-house achievements, enabling us to always build learning models of the highest standard.

ーーWhat trends have you seen as a result of the company’s accumulated track record? Also, the number of cases of providing tools as SaaS is increasing, but what is the importance of developing for each client?

Mr. Kakigami

Even if we have accumulated a wealth of know-how and knowledge through the accumulation of achievements, there are still many subtly different parts for each customer, so we felt that it would not be good to completely automate and make it a service.
Therefore, we decided to realize PoC at a low price in order to maintain our goal of creating the best AI customized for each individual customer, instead of making it a SaaS type and saying that any company will have a certain degree of accuracy. I was conscious.

ーーWhy do you stick to low prices to the extent that profits are suppressed?

Mr. Kakigami

To be honest, if we can reduce internal costs by improving internal efficiency and continue to operate at a unit price of 5 million yen or more as before, profits will increase and the company will be stable.
However, we originally started this company with the idea that ” the ones who are really in trouble are small and medium-sized enterprises for which the value of introducing AI is very high .”We believe that lowering the price is the only way to provide services to such companies, and we have been implementing the industry’s lowest price standard for three years.

Thanks to that, we have been very pleased with both large companies and small and medium-sized companies, but in order to further accelerate the creation of value, we are committed to a low-priced project this time, regardless of our company’s profitability. Released.

In the future, we will incorporate the latest algorithms and expand the range of support

ーーPlease tell us about your future prospects.

Mr. Kakigami

Even if you build a demand forecasting AI with PoC at a low price, you need an environment (system) to operate it . This system will also be made into a template and will be available at a low cost, so we plan to release it following this release.
Transformers, which revolutionized NLP, have recently begun to be applied to time-series and tabular algorithms. Demand forecasting AI is constantly incorporating the latest algorithms, so it will evolve day by day, and we would like to expand the scope of support beyond demand forecasting.Also, recently there have been an increasing number of optimization projects that do not use machine learning, so we will continue to improve efficiency in this regard as well. In addition to commissioned development projects, KICONIA WORKS will release new services incorporating AI in fields such as entertainment and sports.

lastly

One of the bottlenecks in introducing AI is the high cost of development and introduction. As a result, many companies are missing out on opportunities to create value with AI.

RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Recent Posts

Most Popular

Recent Comments