How to Use AI to Deliver Customer Value (Real World Examples)

The Core Argument

Guided by host Sanjay Saini, engineering leaders Amit and Deepak Bhadouriya detailed how Barco transformed abstract AI hype into practical, revenue-saving infrastructure. By building a dedicated "AI Accelerator" framework, their teams shifted from isolated machine learning experiments to deploying scalable, autonomous intelligence directly into edge devices. They argued that organizations successfully deploying AI are those treating it as a core platform engineering challenge—solving data cleanliness, deployment pipelines, and cultural resistance upfront, rather than just forcing LLM wrappers onto legacy software.

“The technological challenges are the least of the challenges in AI. The real challenge is realization: if we have invested in this, how do we get revenue out of it?” Amit (45:13)

Detailed Answers

1. Could you give a brief intro about yourselves and your backgrounds?

Amit brings over 25 years of IT experience, serving as a Senior Technical Program Manager at Barco, where he drives the adoption of AI practices across business units. Deepak has 15 years of backend and data engineering experience, currently serving as an Engineering Manager for Applied Machine Learning and Data Platforms at Barco. Together, they lead the central software platform team that powers enterprise, healthcare, and entertainment product lines.

2. What are the five pillars of your AI accelerator ecosystem?

Amit explained that their AI Accelerator acts as an internal engine with five strict pillars: 1) Ready-to-use AI integration assets (like CI/CD model deployment pipelines), 2) Traditional AI vision and NLP services, 3) Agentic AI frameworks for autonomous action, 4) Generative AI capabilities focusing heavily on RAG (Retrieval-Augmented Generation) architectures, and 5) Leveraging the Microsoft 365 Copilot ecosystem for internal business efficiency.

3. How does your 'Model as a Service' approach empower data scientists?

Deepak noted a common industry bottleneck: a data scientist builds a great model but lacks the system engineering knowledge to containerize, host, and deploy it. Barco solved this by providing "Model as a Service" artifacts. A data scientist simply uploads the model weights, and the accelerator automates the entire continuous deployment workflow to end customers, completely removing the dependency on DevOps.

4. How are you using AI for predictive maintenance on hardware devices?

For high-value hardware like cinema projectors, zero downtime is critical. Rather than relying on simple failure alerts, Deepak’s team utilizes connected IoT telemetry data. They run predictive models that analyze historical performance deterioration (e.g., fan RPM drops) to accurately predict exactly when a component will fail, allowing field technicians to preemptively replace it during scheduled maintenance windows.

5. What is the difference between pure automation and an autonomous AI system?

Deepak clarified the difference through execution limits. Pure automation relies on strict, hardcoded rule sets—if step three fails, the process breaks. Autonomous systems are goal-oriented. Driven by reinforcement learning, an autonomous agent can handle ambiguity; if it hits a roadblock, it independently recalculates a different execution path to achieve the final objective without human intervention.

6. How does your Agentic RAG platform support enterprise knowledge retrieval?

Enterprises have fragmented knowledge buried across SharePoint, Salesforce, and legacy SQL databases. Deepak explained that their custom RAG platform aggregates this unstructured data. When an admin or field tech asks a complex support question, the system identifies the product classification intent, searches the unified vector database, and generates an exact resolution guide, drastically cutting down SLA resolution times.

7. How do you handle AI deployment when cloud connectivity is restricted?

Many enterprise hardware clients completely prohibit cloud connectivity for security reasons. To accommodate this, Deepak detailed their hybrid deployment approach. By modularizing their AI components, they can deploy customized, smaller parameter models (like Llama implementations) directly onto the customer's on-premise hardware, executing localized edge AI without ever transmitting data to an external cloud.

8. Are the cultural challenges of AI transformation similar to Agile transformation?

Amit strongly agreed that AI transformations directly parallel the digital and Agile transformations of the past two decades. The central blocker is never the technology—it is the organizational culture. If a company lacks an "Agile DNA" that openly tolerates early experimentation and failure, they will struggle to extract any genuine business value or ROI from their AI investments.

Stop running isolated AI experiments and start building scalable business value. Learn the architecture required to deliver real organizational transformation in our Artificial Intelligence training program.