Bias in training data for computer vision AI can lead to models that perform poorly in the real world and can even leave companies open to charges of discrimination. This is particularly important in a public facing industry like retail. AI has the potential to transform many aspects of in-store shopping, making the customer experience better whilst improving profit margins for retailers.
However, if this technology is to be fully accepted and adopted it needs to function in a wide range of contexts. Accessing reliable, accurately annotated and varied image and video data is crucial. The data annotation process can form a vital part of AI companies' overall effort to avoid bias in their models.
We will look at the AI use cases that are beginning to make their way into stores across the world. Then we will show how partnering with annotation services, like Keymakr, can help AI innovators to manage bias issues through smart training dataset creation.
Emerging AI applications in retail
The following use cases for AI in the retail sector promise to streamline the in-store shopping experience:
- Smart checkouts: Automated checkouts can speed up shopping and reduce staffing overheads. However, they can present issues in terms of loss prevention and theft. Smart checkouts can monitor the products being scanned and alert security staff in the event of theft.
- Real-time inventory management: AI systems have the capacity to monitor shelves and inventory space in real-time. Alerts can be given when a particular item is running low on stock, or when a particular brand has a low shelf share. Automated inventory management promotes efficiency and product visibility.
- Sentiment analysis: Retailers can gain a deeper understanding of customers’ relationship to products with the help of AI powered sentiment analysis. AI systems monitor customers facial expressions, providing vital data as to how they respond to certain products and promotions.
- In-store traffic analysis: AI powered cameras can track customers’ movements around retail spaces. This can in turn allow promotions and specific products to be placed in optimal positions.
- Face recognition: This use case enables personalised shopping experiences. Automated customer identification means that product and promotion recommendations can be tailored to their shopping history. It also means that repeat customers can easily be rewarded with loyalty bonuses.
A key factor in overcoming bias for retail AI companies is conscientious data collection. Data collection for computer vision projects involves searching open source image and video repositories for suitable data, or using web scraping tools to find images on the internet. Effective data collection can be a time consuming task for busy AI companies. Outsourcing to the collection experts at Keymakr makes it far easier for developers to gain the benefits of high quality training data.
Overcoming bias in retail can sometimes mean having to create data from scratch. Training data may not be available in sufficient quantity or quality, or it might not feature a diversity of light levels and store layouts. Keymakr’s in-house data creation facilities give retail AI innovators the options and flexibility to specify the image and video data that their models need.
The advantage of data annotation outsourcing
A lack of ethnic and cultural diversity in training data can lead to models that do not function well in varied regions and contexts. Facial recognition AI may struggle with different skin tones, and store layouts may vary from country to country. Keymakr’s data collection and creation expertise allow AI companies to craft datasets that reflect the complexity of a diverse world.
Contact a team member to book your personalized demo today.