**This is a continuation of last week’s blog post which can be accessed HERE.**

As mentioned in last week’s blog, Algorithms serve as the backbone of Stitch Fix. From the creation of the profile, to styling the user, to shipping the clothes, and gathering feedback from the user to make their algorithms more accurate and stronger.

1. Creating a Profile:
   1. The first step when logging into Stitch Fix is to create a profile that encompasses your gender, what you are looking for with Stitch Fix, your personal style, body insecurities, by what time you would like the Fix (a box full of your 5 items of clothing), and any other comments that you feel will help your stylist best style you. Once you create a profile, an algorithm determines which stylist has the potential to style you the best and matches you up with a stylist.
   2. The algorithm is a type of comparison algorithm (algorithm that sorts elements) to compare your answers with a stylist’s information. It is loosely based off an algorithm called the Needleman-Wunsch Algorithm  that was originally created to compare biological sequences and match and align proteins and nucleotides! But now, it is used in a number of different industries, such as in retail with Stitch Fix to compare huge amounts of data and make optimal matches between them! It does this by utilizing a scoring system. In this specific system, they look at a pieces of data and score them based on if it is a match (the two things are the same), mismatch (they are different), Indel (insertion or a deletion-aligns to gap in other string-but this part is not used as widely in Stitch Fix). Matches earn one point while mismatches and InDels lose a point. The points are added to the existing score (oftentimes 0). The points are then put into a table in a matrix format and can be calculated from the top, left, or diagonals in the chart. The highest score is the best score with the most matches. Stitch Fix may also be using a different point system based on their criteria. I highly recommend reading the article to further understand this.
2. Styling the User
   1. Ok. So now you are paired up with the stylist…What next? The next step is for the stylist to get you styled! There are three of algorithms to help the stylist pick the outfit for you from hundreds of thousands of clothes in their inventory.
   2. The first algorithm helps sketch a base idea of an outfit by picking three example items of clothing that serve as “parents”. This means that they are pieces of clothing that can serve as a template for a whole new outfit! The second algorithm takes your user input to manipulate the “parent” items. If you said that you were uncomfortable with wearing sleeveless, the stylist would know to add sleeves if it was not there in the “parent” items. The third algorithm takes your data and generates something completely random to your style. Though this may seem counterintuitive, Stitch Fix believes that it is important to take risks with fashion sometimes and try a new style once in a while!
   3. However, it is important to note, that the stylist still has the final say. Though these algorithms are effective, they are not perfect. Though it can create a mathematically sound piece of clothing, it is unable to evaluate the cultural context, etc. Stitch Fix CEO calls it the blend of human and machines.
3. Finding and Shipping your Fix
   1. Now, your outfit is styled online, it is time to actually find the items of clothing, pack it, and ship it to your doorstep! Of course, there are multiple algorithms used in this step as well for maximum efficiency :).
   2. There are multiple different Stitch Fix Warehouses all across the country. First, a matching algorithm like the Needleman-Wunsch algorithm is used to compare items in the warehouses to see which one has the inventory for the one that will be styled for you. Next out of those warehouses, will be an algorithm that calculates a cost function for each of their warehouses in relation to your location to find which one is the optimal location. This can be looked at as a type of binary optimization problem that will show the route from the warehouse to your house by maximizing the efficiency of it.
4. Feedback for Future Algorithmic Work
   1. Stitch Fix algorithms, though pretty good, are not perfect. That is why they depend on your feedback of the Fix. The more comments about your Fix that you give, the more the algorithms can become accustomed to your likes and dislikes. The algorithms learn from your comments, and in the future, it will provide the stylist more suggestions for outfits that will better match your taste and liking :).

So that was that for the Stitch Fix Algorithmic journey from the beginning to the end! Of course, my explanation of this was in no way correlated to the level of complexity that I imagine Stitch Fix has (as many of their algorithms are most probably secret, customized ones used for business), but hopefully it gave you guys a bit of exposure into how Algorithms govern Stitch Fix’s business :).

Until next time,

Inchara