AI Boosts Advertising Effectiveness by Improving TV Audience Predictions

June 28, 2022


This advertising agency had a TV rating system in place that had become stagnant as TV viewership habits changed. Prediction accuracy was too dependent on varying skills of individual data scientists.


dotData’s platform the company a one-stop solution that could automatically analyze data and discover valuable patterns that had not been used in models previously.


With dotData, the company reduced audience prediction errors by 20% and was able to improve prediction times by 30% to 40%.

ADK Marketing Solutions, a comprehensive advertising agency, provides integrated marketing support to contribute to its clients’ businesses. Data utilization is one of these efforts, and audience data is a critical use case for television advertising. Spot-Navi® is a TV audience prediction system that can reduce the cost of buying advertising time by as much as 30%. The company adopted dotData to improve the system’s accuracy and reduce the impact of seasonal fluctuations. The adoption of dotData has provided significant qualitative and quantitative improvements.

“We introduced dotData as a way to revamp the TV audience prediction system we have built up over the past 20 years and move on to the next step,” – Yoichi Numata, Vice President, ADK Marketing Solutions Inc.”

– Yoichi Numatai | Vice President, ADK Marketing Solutions Inc.

The Challenge

ADK Marketing Solutions (“ADK MS”) has developed, operated, and improved its audience prediction system, “Spot-Navi®,” for 20 years. The system is used to predict viewer ratings per program for three months in order to optimize the purchase of advertising space from TV stations,” said Mr. Numata of the company.

Optimize Ad Buying with Predicted TV Ratings

The price of ad space is determined by average viewer ratings for the most recent weeks. Actual audience rates vary depending on various factors, resulting in ad spaces being overpriced or underpriced, depending on actual viewing patterns. If audiences can be accurately predicted, it is possible to analyze the cost-effectiveness of ad spaces to optimize the buying strategy.

Spot-Navi® calculates viewer ratings based on long-term audience data, like the past year. While this reduces the cost of buying ad spaces by about 30%, it lacks flexibility in responding to rapid changes in tv viewing habits. In addition, “For one-off, localized programs such as specials, we could not use historical averages, so we had to manually make corrections to improve accuracy. 20 years of continuous operation had maintained a high level of accuracy, but we felt that we had hit our ceiling,” said Mr.Numata.

Improving accuracy based on human experience requires a large number of man-hours and workload, and accuracy varies depending on the skills of each individual. We decided to consider AI-based audience prediction in search of a breakthrough to transform the audience prediction system into a system that can immediately respond to rapidly changing TV industry patterns.

Challenges in audience data accuracy due to manual operation

Mr. Numata explains, “In the previous model, which was based on averages from past performance, we have repeatedly improved the prediction logic to reflect the impact of seasonality and trends.

However, it was difficult to successfully combine these factors to calculate predicted values, and we also experienced variations and compromises due to the manual nature of the process,” he recalls.

Furthermore, trends and predictions using new data were required. For example, changes in the environment, such as the inclusion of recorded data in audience ratings and a shift away from television due to the expansion of the Internet.

Therefore, ADK MS decided to utilize AI for audience prediction, believing that AI could analyze data and learn constantly changing viewer trends, enabling highly accurate operations without relying on human labor. We calculated the predicted ratings for several tools in a proof-of-concept (POC) format and compared their accuracy, operation, and degree of labor savings, and as a result, decided to use dotData. 

When we compared dotData with the products of other companies, we were impressed by the concept and mechanism of the system: “We did not need to make full use of our expertise in AI or empirical knowledge of data trends, but just provided the data and the tool successfully discovered trends (features) in audience ratings hidden in the data and built a highly accurate model. This was the deciding factor in our decision,” said Mr.Numata.

Solution: improving prediction accuracy through rapid iteration

ADK MS selected Mr.Kenichi Fujimori as project manager (PM) to promote the project of AI audience prediction with dotData.

Mr. Fujimori’s goal was to replace conventional human and empirical knowledge with the ability to input program-related attributes and past audience rating data into dotData and calculate and quantify viewing trends and patterns as “feature values.”

To optimize the use of dotData, the project team focused on preparing the appropriate data. Mr. Fujimori said, “dotData is a powerful tool that can discover trends and patterns from data as feature quantities. There were many iterations in preparing the appropriate data that would work for the audience ratings, such as type and duration,” he recalls.

We had to ask ourselves, “What data can we use to make better predictions? We used dotData to make predictions for past programs that already had actual measured audience ratings and repeatedly verified what kind of data could be used to improve accuracy. With ordinary AI tools, this type of iterative process is extremely time-consuming and labor-intensive since it requires manual analysis of data trends and feature values. With dotData, we could implement it quickly because it automatically analyzes the number of features once the data is input,” said Mr. Fujimori.

Ultimately, the accuracy and quality of AI-based predictions depend on how much meaningful information is contained in the input data. Therefore, by steadily developing the data, dotData achieved higher accuracy in predicting audience ratings.

Results: reducing viewing rate errors by 20%.

Short-term trends can now be considered in prediction:

In developing the prediction model, new and old models were quantitatively compared. Mr. Fujimori says, “dotData allowed us to take into account monthly trends as characteristics of the patterns used for prediction, and combined with other factors such as day of the week and time of day, we were able to identify short-term trends in the process.”

The error between predicted audiences and actual audiences was reduced by 20%. Furthermore, the correlation coefficient between the predicted audience and actual outcome has increased from 90.7% in the previous model to around 95%,” says Mr. Fujimori. He also says that the time required to predict three months’ worth of audience has been reduced by 30% to 40% compared to the original model.

Retail Owner Finding Missing Items

Improved Ad Efficiency and Better Client Retention

ADK MS released Spot-Navi®, which employs AI audience prediction by dotData, as Spot-Navi® AI Edition in March 2021. Recently, we’ve been getting more and more comments from salespeople saying, ‘The accuracy is improving. In terms of actual results, it has led to the acquisition and retention of new clients and improved advertising efficiency, and understanding of the use of AI is spreading within the company as well, and the effectiveness of the introduction is being evaluated,” said Mr. Fujimori.

Going Forward

The Spot-Navi® AI version is constantly updated to maintain and improve its accuracy. Audience trends change based on influences on viewers and the environment,” said Mr.Fujimori. The model does not end once it is created. We check the model’s accuracy daily, and we feed fresh audience tracking data into dotData and rebuild the model every month,” he explains.

ADK MS has a variety of other data and plans to expand the use of dotData for predicting and discovering insights from data in addition to Spot-Navi® in the future. In particular, since dotData automates the entire process from feature design to the creation of predictive models, we expect data utilization to expand through a wide range of ADK MS operations. Based on the results of Spot-Navi®, we are in the process of promoting its use within the company,” says Fujimori.
Mr. Fujimori offers the following advice to users considering the introduction of AI: “Introducing AI requires a lot of iteration, but dotData can greatly lower the barriers. The first step is to learn about AI through a quick trial with dotData, and then expand the use of AI.” he says. AI with dotData will likely lead to the creation of new business within ADK MS in the future, following Fujimori’s activities to promote its utilization.

Case Study Highlights: Preventing Retail Fraud with AI and Predictive Analytics

About ADK Marketing Solutions Inc.

Solutions company provides integrated proposals and executions to solve marketing issues, planning and buying of digital and mass media, data-driven marketing, etc. The company provides comprehensive solutions in the marketing domain by addressing a multitude of client challenges.

Share On


dotData Inc.

dotData Automated Feature Engineering powers our full-cycle data science automation platform to help enterprise organizations accelerate ML and AI projects and deliver more business value by automating the hardest part of the data science and AI process – feature engineering and operationalization. Learn more at, and join us on Twitter and LinkedIn.