Guesswork Brings Machine Learning to the Masses

Romin Irani, Contributing Writer
Aug. 08 2014, 03:08PM EDT

The benefits of machine learning are well-known. With the tremendous amounts of data being generated on a daily basis, the ability to analyze and interpret this data to make correct decisions is considered a strategic advantage by organizations. One of the key aspects to doing that is the ability to predict quickly and correctly the most likely scenario from the volumes of data that you have. This is where machine learning comes in, but developers will be the first to acknowledge that it is not an easy job to understand and implement it in your applications.

Guesswork, a machine learning platform startup founded by Mani Doraisamy and Boobesh Ramalingam, tackles that problem head-on and greatly simplifies the task of integrating machine learning in applications.

Speaking on the product launch, CEO Doraisamy said, "Google understands customers' intention with a few keywords in a search box. It proved that understanding intention can provide immense value to customers (by providing accurate search results), while creating the largest business in the world. Even with troves of data, CRM still doesn't understand what customers want. We want to help CRM companies to build that intelligence into their product and replicate what Google has done on the web in every company's CRM."

The initial focus of Guesswork is to predict customer intent. To do that, it takes a clever approach of using the Google Prediction API with its own rules engine layer on top to improve prediction accuracy. Developers have typically struggled with the Google Prediction API not just in terms of training the model effectively but also writing their own rules on top of it. Guesswork does all that heavy lifting for you, so all you need to do is provide the data and then query it based on your input data.

The Guesswork execution model uses social media data to build customer personas and combines that with users' business data to predict customer interests and action. The initial segment that Guesswork is targeting is CRM companies. This is a good segment to go after since the essence of such systems is to understand customer data and try to predict customers' next steps or what services they would like.

With CRM as the focus, Guesswork provides multiple project templates that users can employ straight away. This includes auto-responders to customer feedback, product recommendation and personalized newsletters based on customer interest. For CRM companies, this would be a great add-on integration that their customers would love to gain insights into customer behavior and tailor it to their needs.

You can take Guesswork for a spin right away with the free tier that allows five models and 1,000 predictions per month. Once you've signed up, the workflow to create your own prediction model and what you want from it is simple. The steps include Create a Project -> Design Your Project -> Upload your Data -> Publish your project.

To start your evaluation, you can choose from one or several project templates that have been created for you.


As an example, I chose the auto-response engine that will take customer feedback entered via a contact form or email and help me classify the response per category (food, service, etc.) or type (praise or complaint). Training a prediction model is tough and you need data. For that, Guesswork makes the task easy for the template projects by providing seed data that can help train the prediction model. A sample run from Guesswork's dashboard for my customer inquiries is shown below:


Guesswork exposes a REST API with JSON as the data format. The authentication mechanism is via a security token. When users create a project, it provides a nice JavaScript-based integration code that they can integrate into their applications. Client libraries in other languages are not available at the moment, but they are in the plans.


Guesswork is being piloted at three beta sites. One is a customer acquisition platform for universities in the U.K. to help predict which students are likely to convert, and other two are large CRM product companies that are integrating the platform into their offerings.

Guesswork has chosen machine learning, an area that has been traditionally difficult for developers to integrate. With its unique approach of hiding the details of ML behind the scenes, building itself on top of strong foundational services like Google Cloud Platform and Google Prediction API, along with its custom rules engine, it has greatly simplified the task for developers. With a few clicks, developers can integrate it in their code right away. While its current focus is on CRM, it would be great to see what kind of templates and target segments it goes after in the near term. Maybe a marketplace or mechanism for developers to submit their own models would be an interesting play for Guesswork.

To get started with Guesswork, visit the sign-up page.

Romin Irani Romin loves learning about new technologies and teaching it to others. His passion is to help developers succeed.