Machine Learning / AI / Statistics as a Service – an overview of use-cases and cloud services that address them

The problem

There are many reasons one would want to use machine learning or AI algorithms. Sadly there are some reasons people don’t get around to it. Some include:

  • Lack of access to (quality) data-sets.
  • Confusion around goals to be achieved.
  • Intimidated by the myriad choices available in terms of algorithms / programming tools.
  • Lack of programming experience and hence steep learning curve for some.

Some of these are real issues which require work to fix, others thankfully are solvable. In this post I’ll try and list (not exhaustive by any means) the motivation someone might have to use machine learning in their day jobs and the tools / services available out there to help them achieve their goals.

Fraud Detection

Run a business thats fraught with risk/fraud? From banks giving out loans to credit card issuers to even transaction marketplaces like Paypal / eBay / AirBNB face such.

Services like Sift Science already have a lot of adoption with clients like Instacart, AirBNB etc. They reduce chargebacks, marketplace fraud etc. Services like Riskified specialize in the eCommerce vertical providing similar value.

Run a call-center? Pindrop Security will help lower phone fraud and provide a phone reputation service.

Data Classification / Tagging

Say you have a million images in your digital library. How do you make these easily accessible or available? On the consumer side Facebook and Google both have internal tools to do image recognition, and Google/Stanford reported even more advances.

But for those without the resources of a large multi-billion dollar corporation give Clarifai or MetaMind a shot. See this example I ran through MetaMind (it was a tough one and it got coffee mug right with high confidence!).

Coffee mug MetaMind

Conversion Optimization (make more $)

What should you send in your weekly email newsletter to convert the user most effectively? How about even what time should you send the email so it gets read! You might consider using a service like Ersatz Labs that lets you upload a spreadsheet with data, wrangle with it, train models and then potentially be able to make predictions. Wrangle data and train models with ease!


Recommendation Engines

This is a common requirement for most medium-large sized eCommerce platforms. “People who bought this, also bought….” or “You might also like….”. This can be done simply via running correlation algorithms / regressions or might be somewhat more advanced using click patterns or search or purchase history etc. For example algorithms have become very smart in detecting if you’ve recently moved cities or homes or expecting a kid or recently married etc. based on subtle non-transaction related signals. A quora link explains the role of machine learning in recommendation engines.

Here’s a quora link with some example services that enable this. Rich Relevance has been in this space for a while.

My problem is very custom / I don’t like using external services

If you find yourself falling in this bucket, fear not! There are so many tools / libraries out there that with some effort and willingness to learn you can put together a solution for yourself. Some libraries and tools you might find helpful:

  • Pandas – to manipulate your data / run experiments
  • scikit-learn – machine learning algorithm implemented in Python.
  • TernsorFlow – a new library for machine learning from Google.
  • Azure Machine learning – Marketplace for ML services that you can purchase and implement quickly.
  • AWS Machine learning
  • – Open-source platform with enterprise support / customizations you can pay for.


Identify the class of problem you’re trying to solve and map it to existing services / capabilities available out of the box to get a proof of concept / prototype up and running quickly. Machine Learning / AI as a Service is here and will only become more accessible and easy to integrate into enterprise stacks over time!

Image courtesy: By Chire (Own work) [CC BY-SA 3.0 (], via Wikimedia Commons