Scaling AI: From Pilot to Product

Saurav Chakravorty • November 07, 2019
Share This Article


With AI becoming the forerunner for the empowered organization of tomorrow, the inflection point for its culture inclusion still seems far when it comes to scaling AI. Despite AI becoming the boardroom conversation now, the organization still falls short in utilizing the full potential of AI.

Senior executives have recognized the transformative potential of AI, yet they are unable to ‘scale’ this potential to transform every part of their organization. Our analysis shows that there are five major reasons blocking the path of AI.

Firstly, winners in this area are still grappling with AI strategy enterprise wide owing to reducing novelty quotient for AI. The need of the hour calls for focus shift towards ROI from AI investments. Secondly, this problem is compounded by the technology maze that organization find themselves in.

The third area of concern is talent. Strong AI engineers are highly coveted. Ensuring that those engineers focus most of their time on the most pressing problems of the organization is critical for success. Adding to this, functional siloes prevent this kind of cross leverage of ideas and resources.

Finally, the challenge is around data. The problem around centralizing data and ensuring its good quality is still something we are in the process of solving completely.

Scaling AI : How to do it effectively

Elements enabling AI at Scale

To create long term value, Chief Analytics Officers have to immediately move from being a cost centre to a profit centre. The move needs to start from inculcating a product engineering mindset from the current research-oriented mindset.

CAOs need to start thinking of their projects through 4 different lenses of outcome, namely ownership, cost, team structure and revenues. These lenses can be further applied to solve the CAOs’ most common problems as following.

Research oriented problems see quite a common occurrence and typically have lower ROI expectations. Consequently, the outcome ownership should be with business teams who own the problem statement and so should be the cost ownership. The code development team could consist of data scientists and data engineers.

Other kind includes applied data science problems. With more certainty around outcomes in this area, the central CAO team could create a cross functional team of Product Managers and Engineers to run the projects in it. There needs to be a formal charge back mechanism for all components developed by this team.

In the final category, we see problems hovering around areas that are verticalized and well solved. These are formal data products that need a wholistic product management team to develop and manage. These projects also need strong financial management to price them properly.

Let’s look at some use cases around the above-mentioned problems. One of our leading Pharma clients wanted to automate a particular process in protein crystallization. The image classification task herein, comes with limited anticipation around reaching human level accuracy. Owing to this, the project is categorized as R&D with business unit benefiting from the automation. The business unit assumes the risk in case the model accuracy isn’t at par with what they were doing prior to this.

One of our clients in Financial Services had multiple text mining project in production. Each of these text mining projects were developed by independent team using their own methods and algorithms. Working with our client we realized that there was an opportunity to improve all the models by creating a custom Word2Vec model trained on financial domain. All the models in production would benefit from such a model. We floated a project to train this model and this was funded by the central AI team with the cost being rationalized based on chargeback to all the team that were doing test mining and would benefit due to this Word2Vec.

Lastly, within Brilio, we realized that aspect-based sentiment analysis from a service perspective would aid a lot in areas such as customer feedback analysis, product reviews etc. This is a perfect candidate for a data product and to create this product, we created a cross functional team consisting of product managers, design thinkers, data scientist and software developers who built this product end to end. It is available to internal customers via APIs and is charged per use.

Brillio recently had Saurav Chakravorty, Principal Data Scientist, deliver a keynote session on “Scaling AI: How to do it effectively” at CYPHER Bangalore, 2019 organized by Analytics India Magazine. The session highlighted the product engineering mindset adoption towards enterprise wide embedding of AI.

If your organization is on an AI powered journey, reach out to us at to know more on how our AI capabilities and solutions can aid you right in the adoption, acceleration and scaling of it effectively.

Let’s create something amazing together!

Contact us Next
Latest Blog
LinkedIn Instagram Facebook Twitter