Hi! My name is Lauren Chircus, Fellow at Insight Data Science, and I’m a data scientist. Tindie came to me with a plethora of questions about their platform, and I’m here to share the insights I’ve gleaned from their database with the Tindie community.
Most importantly, we wanted to know what makes a product successful on Tindie.
Here’s the TL;DR version…
1. Length of time on Tindie → Don’t get discouraged if no one has bought your product yet!
Products are discovered at all lengths of time since posting, and the longer a product has been posted, the more likely that it has been ordered.
2. Description Length → Provide a detailed description of the product and at least a few clear images!
Products with description lengths longer than 375 words are up to 15% more likely to have been ordered, and having at least 2 images gives it an additional boost.
3. Title Length → Use a short but informative title!
The probability that a product has sold tends to decrease with the length of the product title.
Now the longer answer & results…
How do you define success?
First we need to find a definition of “success” that we can (1) quantify and (2) leads to useful insights. The total quantity ordered meets the first requirement, but unfortunately does not meet the second requirement. About half the products on Tindie do not get ordered at all, and the other half follow what’s called a “power law distribution” – the number of products decreases rapidly as the total quantity ordered increases. You can see this in this histogram:
As a result, the variation in products that have smaller quantities ordered is huge, and there are only a few products that have large quantities ordered, so we can’t learn very much with this metric.
On the other hand, we can ask what the difference is between a product that has been ordered and a product that has not been ordered since the data splits pretty evenly between the two classes.
- # of products that have not been ordered = 1863
- # of products that have been ordered = 1343
Definition of success (for our purposes): a product that has been ordered at least once
**Note: we’re limiting our analysis to products that have been listed on the site and orders that have occurred since 1/1/14 because of major differences on the site prior to 2014.
Picking Features
Now that we have a definition of success, our new goal is to understand what factors differentiate a product that has been ordered from a product that has not been ordered.
To achieve this, we will use a random forest classifier (you can learn more about how it works here). The random forest will allow us to find both the how important a particular factor is as well as the relationship between that factor and the likelihood of a product selling.
A random forest classifier trained on the products listed on Tindie since 1/1/14 determines whether a product has or has not sold correctly 72.5% of the time. It categorizes products that have not sold (“true negatives”) correctly 83% of the time, and it categorizes products that have sold (“true positives”) correctly 58% of the time. This discrepancy in the ability to predict true positives and true negatives happens when there’s a lot of variability in the negative (have not sold) class. In our case, many products that have not been ordered may be products that have not been ordered yet, resulting in factors that should be associated with “have been ordered” appearing frequently in the “have not been ordered” class.
The mean decrease in Gini index, a measure of how important a factor is in separating two classes, shows us the most important factors for distinguishing products that have been ordered from products that have not:
The time a product has been on Tindie is a very important factor in determining whether a product has been ordered, while having a Youtube link on the product page is not.
Let’s take a closer look at how some of these factors impact the probability of being ordered.
1. Time on Tindie
Products on Tindie are “discovered” and ordered at all lengths of time following their initial posting. If your product has not been ordered yet, don’t worry – that doesn’t necessarily mean it won’t ever get ordered! By looking at the plot below, we can see that the probability of product getting ordered goes above 50% (assuming all other factors are equal) by 1 year after posting:
II. Length of Description
Having a long description of the product increases the probability that it has been ordered, while having a short description decreases it. Based on this data, I suggest bumping up your description to at least 250 words (though 375 words gives you an added boost). Only a small number of products meet that pretty low bar! Check out ElectroPi IoT PCB (315MHz – Helical Ant) for an example of a nice, long description.
While having a long description improves the probability of your product selling, having additional documentation on Youtube links doesn’t hurt or help. Buyers primarily want a lot of information in the description.
III. Number of Pictures
While the number of pictures on a product page is one of the least important predictors of whether a product will be ordered, including 2 or more pictures does give your product a little boost. If you have one of the nearly 1200 products that only have one picture, upload at least one more!
Look at the beautiful pictures on the CMOY Headphone Amplifier for inspiration.
IV. Title Length
Largely, longer titles take a bigger hit in terms of probability they will sell. Short, catchy titles tend to sell better (think AirPi). Think about taking extra information (i.e. model numbers or specs) out of your product title.
Hope this helps you improve your product pages! As Tindie keeps growing, we’ll be able to learn even more about what makes products successful.
Keep Reading ›
Images from Making Society
