Our Approach

Holistic approach with Data Science

We take a holistic approach to any data science project. Data Science is not about just creating models. We follow a rigorous process starting with understanding the business goal.

Understanding Business Goal

Business Goal is defined in terms of impact you want to see in the business. Think of the final outcome you want to achieve like improving sales or profitability or customer satisfaction. Building a machine learning model is not the end goal, it is the means to the goal. It often requires more than one machine learning model, some experiments and optimization to achieve the final business goal.

We work with you to make sure we understand your business goal and then find a solution based on data science and modeling to achieve that goal.

Existing & New Data

We start with getting a deep understanding of available data. Many a times this data is sufficient to proceed with creating a model, but sometimes there isn’t enough data available to build models and answer all questions related to the business problem.

New data may need to be collected by performing carefully planned experiments. This process can be challenging and time consuming, but a well-planned experiment generates highly valuable data.

We look carefully within existing data to extract as much insights as possible before considering gathering new data. Sometimes insights can be inferred from existing data even if it was not collected by running experiments. We work diligently to extract meaning from available data to make the best out of it.

Building Models

Here comes the most exciting part of Data Science. Depending on the problem we could build a single or multiple models. Some solutions require one simple model, like model to predict price of a used car. While other solutions could use multiple models that work together to arrive at a decision. For example, a model that scores every customer interaction across various online and offline channel and another model that calculates customer lifetime value based on all those interactions and past purchase history.

Evaluate & Tune Performance

Before developing a model, its evaluation criteria a validation and a test dataset are set aside to objectively evaluate the model on a preselected criterion. Once the model is developed, validation dataset is used for tuning the hyper-parameters to optimize performance of the model. Hyper-parameters are parameters that do not affect the architecture of the model, but tweaks some parameters for selected architecture.

Finally, the model’s performance is evaluated on test dataset. This is the true test of whether the model performs as expected on future data.