How To Keep A Great Project From Getting Stuck In The Pilot Phase: Scaling Strategies From Siddharth Pani

In 2026, experts began to increasingly discuss a new challenge facing the artificial intelligence market. While just a few years ago the main concerns were model quality and computing power, today the focus has shifted to a much more practical challenge – deployment. According to the industry study Enterprise AI Deployment Gap, published in May 2026, about 70-80% of corporate AI pilots launch successfully, but only 20-30% reach full-scale implementation and widespread use in business. Experts cite integration challenges, a lack of clear KPIs, data issues, and companies’ insufficient readiness for organizational change as the main reasons.

In other words, modern companies are increasingly facing not a shortage of technology, but a lack of understanding of how to turn technology into a functional business tool.

Siddharth Pani, Senior Technical Project Manager at Anicca Data Science Solutions, is well acquainted with this challenge. Over the years, he has led projects for Microsoft, McDonald’s, NVIDIA, Intel, and other major organizations, and developed solutions in the fields of artificial intelligence, computer vision, cloud infrastructure, and cybersecurity. He managed international engineering teams and his own product FAVO development, an enterprise learning management platform. We spoke with him about why many AI projects remain pilot projects, what mistakes companies most often make, and what actually determines the success of technology initiatives.

Siddharth, industry research shows that most AI pilot projects never make it to full-scale implementation. Why do you think this happens?

Because many organizations start with the wrong question, instead of asking what business problem they want to solve, they start thinking about which specific AI tools they want to use. In my experience, the most successful projects always started with an operational challenge. For example, when working on solutions for the restaurant industry, we first identified the metrics that truly impact service quality and operational efficiency, and only then selected the technologies. In one project, we created a system of 26 distinct metrics to evaluate the customer experience and operational processes. Until a company understands which metrics it intends to improve, any AI initiative risks remaining an experiment.

The restaurant industry example you mentioned is clearly the McDonald's Anicca Vision project. You worked on the Anicca Vision project for that company, which included the front desk, dining area, and drive-thru. How did the project begin, and what were the primary objectives?

The project began with an analysis of actual business processes. We needed to understand which employee and customer actions affect service speed and the quality of the experience, and what data could be collected for automated analysis. As I mentioned, we developed a system that allowed us to track process efficiency in real time, including cashiers’ performance, drive-thru service speed, and customer interactions. Only then did we move on to building a computer vision model and integrating cameras and edge servers for real-time video analysis.

You scaled this project to include 21 restaurants – what challenges did you encounter?

Scaling up significantly changes the project; it’s always a complex and time-consuming process, since we had to integrate the systems of all the restaurants into a single system. Across 21 restaurants, the system now has to work with different cameras, networks, and local servers, while all data must be collected centrally. We installed local servers directly in the restaurants, configured the system to quickly process data from multiple cameras simultaneously, and automated the connection of new locations. This made implementation much easier, and the system was able to operate reliably in different restaurants without constant engineer intervention. As a result, the solution became stable and reproducible outside of laboratory conditions.

Developing an AI model is only one part of the journey. Making that model deployable, maintainable, and scalable across hundreds of real-world locations often requires collaboration between multiple technology partners. Based on your experience working on projects that were later presented at Embedded World in Germany and recognized by Microsoft, what role do partnerships play in the successful commercialization of AI solutions?

Such partnerships are critical to the successful implementation of AI. The technology itself is only part of the solution. In the case of Intel, we were able to use cost-effective NUC processors with integrated graphics, which helped reduce hardware costs and accelerate the system’s deployment in restaurants. But just as important is that collaboration with partners helps scale the product and ensure its stable operation in a real-world environment. External partnerships, such as Microsoft’s recognition as a key partner in bringing the product to market, provide confidence that the technology not only works in the lab but can also be reliably used in hundreds of locations without constant intervention from engineers.

After McDonald’s, you continued working on infrastructure projects for Microsoft. What key lessons did you carry over from the restaurant project to your cloud projects?

One of the biggest lessons was that technology alone is never enough. At McDonald’s, I saw that even the most advanced AI solution creates value only when it fits naturally into everyday operations and solves a real problem. I carried the same mindset into my work with Microsoft. Regardless of the industry, success depends on understanding people’s needs, building reliable processes, and making technology easy to use. In the end, the goal is not simply to create innovative tools, but to make them useful in real life.

In addition to corporate projects in Microsoft, you also led the product vision and led the development of his own product, FAVO — an enterprise learning management platform. FAVO included session recording, online quizzes, assignments, and user-based access control. How has working on your project changed your perspective on AI implementation and project management?

Building your own product teaches you very quickly that technology alone does not create value. In enterprise projects, the business problem is usually already defined. When you're creating a product from scratch, you have to identify the problem yourself, validate assumptions, prioritize features, and constantly evaluate whether users actually benefit from what you're building.

With FAVO, I used large language models to analyze learning materials and generate educational content, but the biggest challenge was not the AI itself. It was understanding what users truly needed and how to deliver that value consistently. That experience reinforced something I've seen throughout my career: successful AI projects are driven by outcomes, not by technology. The model is important, but solving a real problem is what ultimately determines success.

Looking back at your projects and extensive experience, when you compare the AI landscape today with the one you entered at the beginning of your career, what stands out most?

The biggest difference is that AI has become much more practical and accessible. A few years ago, many discussions focused on what the technology might be able to do in the future. Today, AI is already part of everyday life, from smartphones and recommendation systems to business operations and industrial applications. Another important shift is the growing ability to run AI directly on local devices rather than relying entirely on the cloud. This makes the technology faster, more affordable, and easier to use. To me, the most exciting part is seeing AI move from experimental projects to tools that deliver real value to people and organizations every day.

Published on: Thursday, July 23, 2026, 06:59 PM IST