Analyzing 10,000 ProductHunt products to determine which products are most successful

min read

Want to know what’s hot and what’s not on ProductHunt?

Resident Data Evangelist, Michelangiolo Mazzeschi & Head of Marketing, Adam Marsh have analyzed the user reviews and product descriptions of over 10,000 ProductHunt listings to find out the answer.

ProductHunt can make or break a launch for a business or product. Which is why founders (that’s you people!) tend to invest a significant amount of time & money to build up a launch strategy, image assets, video, copy and external marketing campaigns.

Sometimes though, despite your best efforts, you won’t achieve great success as some business models just aren’t popular on ProductHunt by design.

Given I have a great interest in understanding why and how things work, I wanted to pinpoint what these factors are.

For me the best way to do this is to perform topic modeling on the product descriptions of 10k+ products on PH – all to find insights and determine which type of products were most popular.

If you would you like to see the ProductHunt Product Clusters in the dashboard, whilst reading along then you can view the Clustering Dashboard here

What makes a successful ProductHunt listing?

  • The ProductHunt audience LOVE Web based apps, automation products, email productivity, community focused products and screen recording / screenshotting software.
  • Don’t expect Pomodoro, QR Code, Habit Tracking, VPN, Paywall or Notification products to get high engagement.

The Good

  • Products that achieved the highest number of reviews were based around emails (25 in total) and rèsumes (41 apps).
  • You will face the lowest amount of competition and may prefer to build these type of products below
    • Timezone (9 apps)
    • VPN (10 apps)
  • 1st – As an overall leading type of product, building a webapp is the way to go. This type of product achieved:
    • 1st highest avg upvotes (11.11)
    • 1st highest avg reviews (30.63)
  • 2nd – Automation based products that were focused towards marketing, tech, AI & productivity score quite high.
    • 2nd highest avg upvotes (10.52)
    • 11th highest avg reviews (10)
  • 3rd – Email Productivity products tend to solve an in-demand problem.
    • 3rd highest avg upvotes (10.42)
    • 93rd highest avg reviews (5.74)
    • This was interesting as although the upvotes were high, the average reviews were in the bottom 50% of all products.
  • 4th – Community focused apps relating to community management on social media or within forums scored highly.
    • 5th avg upvotes (9.13)
    • 51st avg reviews (7.02)
  • 5th – Screenshot or screenrecording apps were also in high demand. These types of apps were very popular with both upvotes and reviews.
    • 6th avg upvotes (8.85)
    • 19th avg reviews (9.18)

The Bad

  • Generally you may want to avoid building in these saturated categories as they had the highest most competition amongst
    • Automation (148 apps)
    • Chatbots(118 apps)
    • Design (118 apps)
    • Video (105 apps)
  • Overall, the products that were reviewed the lowest amount were based on web domains, QR Codes & habits / behaviour focused products.
    • My hypothesis here is that both of these apps tend to be used away from your average ProductHunter’s desk, or have long timespans to give an accurate review, which requires a greater motivation than reviewing a Browser extension.

The Ugly

  • Products that achieved the lowest average results and upvotes were as follows.
    • 1st – Paywall & VPN related products
    • 240th lowest / bottom 9% for average reviews (2.1)
    • 208th lowest / bottom 17% for average upvotes (3.5)
  • 2nd – Update and notification based products
    • 239th lowest / bottom 11% for average reviews (2.29)
    • 236th lowest / bottom 7% for average upvotes (2.67)
  • 3rd – Google Suite focused extensions / products
    • 238th lowest / bottom 11% for average reviews (2.41)
    • 234th lowest / bottom 10% for average upvotes (2.91)
  • 4th – QR Code / Barcode orientated products
    • 237th lowest / bottom 12% for average reviews (2.44)
    • 240th lowest / bottom 4% for average upvotes (2.25)
  • 5th – Productivity Apps & Pomodoro Timers
    • My hypothesis here is that these apps are simple and rehashed, with little innovation between them. There’s a low engagement here as new products entering this market aren’t innovative or impressive enough.
    • 236th lowest / bottom 12% for average reviews (2.47)
    • 218th lowest / bottom 15% for average upvotes (3.33)
  • 6th – Habit Tracking & Behaviour based products
    • 239th lowest / bottom 6% for average reviews (2.44)
    • 230th lowest / bottom 15% for average upvotes (2.88)

Dataset

Compiled from over 2000+ tickets taken from our project management system Linear. These were then exported into an CSV file and subsequently uploaded as a dataset in the Relevance AI platform, then vectorised & clustered.

Technical Write Up

Using the all-MiniLM-L6-v2 encoder, 10,000 product descriptions on 4 main product categories, Marketing, Productivity, Tech, and AI, have been converted into 786-dimensional vectors (which are coordinates based on a 3d plane).

A K-Means clustering algorithm was applied to check which the similarity of reviews in groups of 60, 120 and 240 clusters.

These results were then fed into a Relevance AI dashboard.

Due to other features in addition to the product description, such as the number of upvotes and the number of written reviews, we have used aggregations to show the total number of upvotes and reviews per cluster, as well as the keywords extracted from every cluster using a technique called zeroshots.

Solution

We can have two variations of the same dataset, one portraying the description of the products, and the other one showing the logs (static png or gif) of the different products.

Dataset

Standard web scraping tools such as Beautifulsoup did not work on ProductHunt which required a different technique.

By using pyautogui (a python library to control the keyboard and the mouse movements), to automate the page scrolling, we were able to scrape this differently.

Thanks for reading!

If you would you like to see the ProductHunt Product Clusters in the dashboard, feel free to follow along here –
View the Clustering Dashboard here

If you would like to DIY this data experiment, feel free to use the following links:
View Juypiter notebook here
Download the ProductHunt Dataset here

Analyzing 10,000 ProductHunt products to determine which products are most successful
Benedek Zajkas
March 7, 2022
Want to mine external data sources for business and product strategy?

Find insights in unstructured data with Relevance AI's Clustering platform.