Gast is transforming the way people discover and interact with their personal design style thru our custom algorithms. Our customers desire a dream home personalized to their design preferences, and we offer solutions tailored to an individual’s style without the burden of endlessly searching or trying to keep up with current trends. Our products are curated by our in-house team that continually searches for leading creative and innovative options. We are connecting customers with the style they love (and wouldn’t have found on their own).
So how do we do this? We start by having our customers complete a questionnaire, which provides us useful data with the least effort. Customers can choose to create an account, which can be updated at any time, and can select filters to remove styles from consideration they want to avoid. Customers then choose between different images, selecting the images they most prefer, which provides more data to our algorithm on their individual style preferences. We also continually receive feedback from our customers.
Our algorithm evaluates the relative likelihood that a particular customer will like a particular style and product. This is a difficult to do, and the algorithms engage this problem from many different angles.
- We tag each item multiple times with match scores from different algorithms, and rank the scores.
- With a database of different customer’s preferences, we are also able to use standard collaborative filtering algorithms (i.e. other people who like what you like also like…).
- We use mixed-effects modeling, which lets the algorithms learn our customer’s preferences over time.
- Our algorithms layer in a lot of photographic data to determine style, not only with our side by side comparison selection tool, but also Pinterest boards and style image selection tools. Often it is difficult for people to describe their style in words, but they know it when they see it. Our algorithms can look at photos of clothing that customers like, and look for visually similar items.
The different algorithm scores are ranked, and the results are sent to one of our human stylists to review. While our machines are great for calculations based on data, some tasks require human knowledge of norms. So we have our stylists review and sometimes finalize selections, and add notes regarding what items may work well together.