Join us October 22nd to hear Coglate-Palmolive, IDC, and Sequoia Capital discuss moving to a digital-first environment
Learn more
Ankur Goyal

Practical AI: An interview with Ankur Goyal, CEO at Impira

In an interview with our founder and CEO Ankur Goyal, we explore Practical AI and what it really takes to put AI to action effectively.

Artificial Intelligence (AI) and Machine Learning (ML) are frequent buzz words that are thrown around by many companies. From self-driving cars to data analytics and smart home devices, AI has been at the forefront of innovation. Yet, how much of this is rooted in practical applicability—and isn’t simply an attractive marketing ploy? 

Recently, we interviewed Impira’s CEO and founder Ankur Goyal to get his take on how AI can be meaningfully and practically applied to help solve a business problem. Having worked with customers from eCommerce to consumer goods and financial services institutions, Ankur knows all too well the struggles and celebrations of putting AI to action in solving their most pressing problems.

Question: What does it mean to make AI truly useful for a business? 

Answer: For AI to be useful, it needs to demonstrate quantifiable ROI for a business, such as time saved or revenue generated. In order to do this, we need to satisfy two core conditions: one, craft a clearly scoped use case that we think can generate such ROI, and two, set up the AI and ML so that it can be applied effectively and successfully to solve that use case. And underlying condition number two is a really critical piece that I often see a lot of companies miss—the ability to incorporate an inexpensive and fast feedback loop that can take in new data so that future predictions can be continuously improved. Together, I like to call this practical AI because it materially moves the needle for a specific business problem or use case.

Question: In today’s world, do you see a lot of companies applying practical AI—that is, in the way that you have defined it?

Answer: I don’t see many people thinking holistically about the combination of both of those conditions, particularly the second one on implementing AI effectively with the incorporation of a feedback loop. However, that is understandable because feedback loops can be both costly and time-consuming. 

For example, implementing a feedback loop for self-driving cars to learn has a very high cost—human death—making it morally difficult to iteratively improve on wrong predictions. Another example is reporting & insights dashboards, in which there is often too long of a time lapse between getting feedback on whether a particular insight has helped and then incorporating that feedback into the dashboard for further refinement. I don’t believe that any business would wait for the next insight, no matter how good it is, with such a long time lag.

As a result, today’s use cases for practical AI are fairly nascent, and I think there is still a long way to go before we can realize the full benefits of practical AI. In fact, if you hear any company stating otherwise, proceed with caution before you engage further!

Question: What about the use cases that are better-positioned or better-scoped for the application of practical AI? What have you seen today that has actually driven measurable business value? 

Answer: We’ve seen a few clear use cases that are better suited for practical AI, including data extraction from documents, image labeling, and so forth. To showcase the data extraction use case, let’s examine invoice document processing for a moment. For many, this process typically requires an actual human to identify that a particular line item corresponds to the purchase of, for example, new office chairs, and then enter that dollar amount into a spreadsheet. What we’ve found and been able to experiment with is that we can properly set up machine learning-based AI to make a prediction and take in feedback from the user to determine if that guess was indeed correct. This use case satisfies the two conditions of practical AI: a clearly scoped use case with quantifiable ROI, and the right set up for AI and ML to be applied successfully with the inclusion of a feedback loop.

What this means is that AI can help users get to cleaner, usable, and business-ready data by understanding the details within documents, with minimal human intervention. In other words, AI can help make sense of the data so that users can then ask questions about their data that can drive towards business insights and quantifiable ROI.

Question: What are your thoughts on the future of practical AI, and what problems do you see practical AI solving? 

Answer: The business problem of managing data within documents and images is and will continue to be a struggle for many companies. Especially with the COVID-19 pandemic and with many workforces going completely remote, documents and images are now the key tools by which businesses interact with their customers. Processing this data is going to continue to be manual and hard, whether it’s extracting data from invoices or identifying objects in images, and will prevent people from drawing business insights that facilitate fast decision-making. 

Practical AI can solve this problem by enabling businesses to understand the data within documents and images, so we can ask meaningful questions about that data. Now, more than ever, if we can be smart about how and where to apply AI, then we can transform how we work, sell, and manage our businesses, and improve the productivity for an increasingly remote workforce of the future.

What’s your view on practical AI? What pressing business problems are you looking to AI to help solve? We’d love to hear from you! Connect with us at or on Facebook, Instagram, LinkedIn, and Twitter.

Subscribe to Impira's blogStay up to date with all things Impira, automation, document processing, and industry best practices.

Highlight and extract data with Impira AutoML.