Small Business Advisor

The website front end

Inspiration

The pandemic hit small businesses hard with all their customers moving online. They had to figure out how to move their operations online all by themselves.

Today, with many countries with limited access to vaccines still operating on a limited, small businesses around the world still struggle to reach any customers while big-box retailers crush their already shrinking market.

With 4.48 billion global users using social media, it goes without saying that social media can be a major equalizer between small and big businesses. But, today, small businesses cannot afford social media managers, data scientists or B2B marketing firms.

The Small Business Advisor (SBA) was made to level this plane.

What it does

This acts as a data-driven personal business advisor. The process is as easy as 1, 2, 3! 1) Set the kinds of posts the business wants to have on their social media (memes, informational, trending posts, behind-the-scenes, etc) 2) The business sets priorities for a) Number of views, b) Follower count, c) Trying new kinds of posts. 3) The SBA will then suggest the type of post to post 4) After posting, the data (#views, follower gradient, # likes, etc) from the post is input to the SBA which will then use Reinforcement learning to readjust its advice for the next post.

This way the SBA continuously learns from social media posts and makes personalized suggestions for the Small Business it serves using Reinforcement Learning. The SBA abstracts out all the data science-y stuff so that the small business doesn't have to deal with that.

And importantly, the SBA explores and learns from mistakes so that the small business doesn't have to.

How we built it

We used a popular Reinforcement Learning algorithm for the SBA: TD Learning with epsilon-soft policy improvement. It basically learns on the go with every new data sample inputted with some discounting factor to adjust priorities give to old/new data. The algorithm learns a categorical distribution and outputs a maximal output.

We used Python3, Numpy, Pandas, SciPy. We DID NOT have to use any of the popular Neural Net libraries like Tensorflow or Keras. We built the algorithms from scratch.

We created a user-friendly front end to emphasize our mission to make digital marketing and social media advising data science accessible to everyone. We used HTML, CSS for this.