How to use AI/machine learning for targeted marketing

The recent evolution and spread of AI has been remarkable. Just the other day, I saw the news that DataRobot, a global leader in the AI business, was making its AI platform available for free to support analysis related to COVID-19. While the adoption and use of AI is progressing in areas such as anomaly detection in manufacturing and demand forecasting in retail, it seems that many companies are still in the trial and error stage. In this article, I would like to introduce the challenges we faced and the solutions we found, based on our company's efforts in target marketing using AI/machine learning.

AI/machine learning requires a large amount of data preparation (collection and processing)

Although the data is a little old (from 2016), a survey of data scientists asked them which task takes the most time when analyzing data using AI/machine learning, and the results were as follows:

What do data scientists spend the most time on?

The survey results above suggest that many data scientists spend the most time on data preparation tasks such as collecting and processing data, rather than on applying algorithms and tuning them. This suggests that, at least at the time of the survey, preparing the data to be analyzed is the most important (essential) task in utilizing AI/machine learning, but also the biggest challenge for data scientists. This is because approximately 60% of data scientists responded that "data preparation takes too much time, and the work itself is monotonous and boring."

Creating learning/prediction pipelines without programming

There is an accelerating trend towards using AI/machine learning to analyze huge amounts of data related to customer purchasing behavior and preferences in order to conduct effective marketing activities, and our company is also using DataRobot for targeted marketing to users of our flagship software package product, HULFT. HULFT provides two types of technical support services: daytime weekdays and 24 hours a day, and DataRobot is responsible for predicting which users who are likely to switch to 24-hour support from among those who receive daytime weekday support.
By using our data integration software "DataSpider" + "DataRobot Adapter" to automate the collection and processing of the necessary data, we have created a system that enables the entire process from model creation to release to be completed without programming and in a short time (about one month).As mentioned above, this not only reduces the burden of "monotonous and boring" data preparation, but also automates post-processing operations such as emailing prediction result files to business personnel, successfully improving work efficiency.

Image of targeted marketing using DataRobot

Using DataSpider is the shortcut to data analysis with DataRobot

To utilize AI/machine learning, it is necessary to prepare the appropriate data, but this is not easy as data processing is often required, such as preparing different data formats for each system, correcting missing data, or combining multiple data to generate the necessary data. Also, in order to achieve business results, there are times when the results of predictions made by AI/machine learning must be passed on to (linked with) external systems for use. This kind of linking function with external systems such as business systems is exactly where DataSpider excels.
We currently offer a "linkage service" that utilizes our products, including HULFT and DataSpider, to support the integration of data and systems. If you are interested in this service or would like to know more about our successful experience implementing and utilizing AI/machine learning, please feel free to contact us.

The person who wrote the article

所 属：リンケージビジネスユニット ビジネス開発部 データエンジニアリングコンサルタントチーム

Yasuyuki Yoro

Related Content