Apply HN: Statistics as a Service
This is in the idea stage.
Business Model: Freemium
Pricing tiers: 1. Free 2. Pro (additional, more advanced statistical smörgåsbord* to choose from, suggested stats to run, etc.) 3. API (programmatic access to FREE and PRO tier stats) 4. Custom (could be custom stats implementations w/ API access, more of a consulting project, or even, a statistician on call, etc.)
* Though, initially the options would start relatively small, and be added to as time progresses and it makes sense to do so.
Use Cases: 1. Knowledge worker at a SMB or local government entity with little access to data analytics or statisticians, but still need to make use of statistics. 2. Dev shops needing stats, but don't have the time, in-house resources, etc. to bother with it.
It is not an attempt ... * at big data analytics * at building an AI/ML service * to be a replacement to R, Python, Julia, et al.
Perhaps you can find some ideas there.
For the Free Tier the idea is provide some basic descriptive statistics with maybe a histogram overlaid with the normal distribution curve on top. Some example values that might be calculated are the mean, median and mode along with several standard deviations. Down the road a bit, I could envision a feature to suggest further exploration, which would likely guide the user towards some of the Pro features. (Although, I would like to expand what is available in the free tier, where it makes sense.)
For the Pro Tier, they would of course, have access to the same results as the Free Tier, but I imagine a wizard like interface (which can be turned off) to help guide the user through choosing what they want done. Initially, the focus would be on helping spot basic trends and more obvious data anomalies. Further down the road, I could see the options expanding.
One of the challenges to be handled is the "width" of the data set. For example, if User A has a CSV with 100 columns, the basic calculations may not be a big deal, but do we present a 100 histograms? What if there are 1,000 columns. My current thinking is for the user to prioritize the columns they want analyzed if they exceed a certain threshold (say, 25 columns). The results would be done on the columns up to this threshold.
I am open to ideas and suggestions on this, and even on things such as which "basic" descriptive statistics are most important.
Edit: Clarification (last sentence)