Frequently Asked QuestionsFAQ

Here are some of the most frequently asked questions about Package Design AI.
The information includes predicted values, images to use, the fee structure of the service, and free plans.

If you have any other questions or concerns, please feel free to contact us using our dedicated inquiry form.

Question.What are the five levels of predicted value?
Answer.The weighted average predicted value of 5 "like", 4 "somewhat like", 3 "neither like nor dislike", 2 "not like much", and 1 "not like much".
Question.What kind of images should I upload?
Answer.Please make sure you have a "front view" image of the package design.
If you do not have a photo, you can use an Illustrator image, but we recommend that the shading be as close to the image as possible.
Also, if the brightness or saturation of the image varies from design plan to design plan, this may be reflected in the forecast results.
See the About Images section for more details.
Question.I don't see a category I would like to try.
Answer.Currently, 51 categories in 8 fields are available and more will be added in the future.
If you do not see the category you want, please select "similar categories" or "other than the above" to try it, although the accuracy is expected to be lower.
Question.Which data set and how many images are used?
Answer.Data was collected through a Package Design Ranking (our web research) conducted each spring and fall. For each image, 1,000 respondents are divided equally into 125 by gender and age group (20s/30s/40s/50s). The dataset currently includes 9,200 products and 9.2 million respondents.
Question.What kind of programs and processes does AI run?
Answer.The Package Design Preference Score AI uses a CNN to extract image features and calculates predicted values based on a specific algorithm based on the features.
Question.In which situations do you recommend using the system?
Answer.In evaluation, it is often used to refine the design that emerged from the initial presentation and to validate the final design.
Design generation can be used to expand on initial ideas or to make minor changes to the final design to increase its degree of completion.
Question.Can I expect the same predictions with different survey methods?
Answer.Different survey methods will produce different results.
For example, the results will differ depending on the number of respondents, or even if the survey method is different, such as web research versus on-site testing. The predictions in this service are based on the "sequential survey method based on web research conducted in the spring and fall of each year".
Question.How accurate are the predictions?
Answer.予Prediction accuracy is the relationship between the predicted values and the actual values of the design evaluation for validation, which is prepared separately from the package design images used in the training model.
Prediction accuracy is based on the mean absolute error and correlation coefficient, which are 0.133 and 0.748, respectively.
The error is a measure of the accuracy of the score, and the correlation coefficient is a measure of the accuracy of the design ranking. For more information, see Accuracy.
Question.Why Preference?
Answer.Purchase intention (the degree to which one wants to buy the product) and usage intention (the degree to which one wants to try the product) are strongly influenced by the product's features and concept. On the other hand, design preference (like/dislike) can be treated as a variable independent of concept evaluation.
Question.Which products or designs are suitable for evaluation?
Answer.AI scores are based on a vast amount of past research, so they are more accurate in traditional design categories where the tone and manner of the design is already set.
It is also not suitable for predicting designs that include Olympic symbols in an Olympic year, for example, because it does not take into account temperature, weather, or events of the time.
Question.Why is image preparation necessary for generation?
Answer.Design generation is a service that generates combinations of design proposals under consideration and repeatedly evaluates them to complete a design that is preferred by consumers. To achieve this, multiple design proposals are needed as the basis for generation.
Question.What is the maximum number of users?
Answer.A company or branch (headquarters or research center) can have a maximum of five users.
Question.More information about the price
Answer.For evaluation, we offer "ad hoc plans" for one-time use (15,000 yen per piece) and monthly plans for unlimited use (~700,000 yen).
For generation, a project costs 300,000 yen and can be used as many times per month as needed. (A separate evaluation plan contract is required.) See the fee plans for details.
Question.I am worried about applying for a paid plan from the beginning.
Answer.Don't worry, we have a plan that allows one company or branch office (head office or research center) to use up to 10 free copies. Although membership registration is required, please try the free plan first.
(*) Each company/branch (head office/laboratory) is limited to one person for up to two months.
Question.Resubscribe after canceling, will my past data be preserved?
Answer.Please be assured that past data will be retained even if you cancel your subscription. However, image data is deleted periodically, so it may be lost depending on the timing.