Global automotive supplier TI Automotive operates at the heart of a rapidly transforming industry. The company designs and manufactures fluid storage, carrying, and delivery systems used by leading OEMs worldwide, from fuel tanks and brake lines to thermal management components supporting the shift to electric vehicles. As automotive manufacturing becomes increasingly digital, connected, and software-driven, the role of data and AI is expanding from operational support to strategic necessity.

In this first installment of a three-part series, Apurva Wadodkar, Senior Director and Head of Data and AI at TI Automotive, joins Merav Yuravlivker, Chief Learning Officer at Data Society, to discuss what it takes to build an AI practice from the ground up. Drawing on experience establishing AI capabilities multiple times, Wadodkar outlines a pragmatic, enterprise-first playbook centered on education, prioritization, and disciplined execution.

Start by removing the “AI Halo”

Wadodkar begins with a reality check. “Every time you begin an AI practice, remember there is a bit of a halo around AI. It almost feels magical, like you can throw anything at it and it’s going to solve the problem.”

That perception can create unrealistic expectations and derail early momentum. Her first move is always education. “Education is critical. Make sure you get your business partners educated.”

Rather than leading with technical concepts, she brings AI into the language of each function. “I would have roadshows with finance, HR, and supply chain people, and go with use cases that they would understand to explain in their own language.”

While generative AI dominates today’s headlines, Wadodkar stresses the broader AI landscape.

“Right now, the flavor of the year for AI is GenAI. But then, what about predictive? What about anomaly detection, time series? There’s so much that AI can do and that needs education.”

The Five-E Framework for building AI

Over multiple roles, Wadodkar developed what she calls the Five-E Framework:

1. Educate
2. Engage
3. Experiment
4. Expedite
5. Effectuate

At TI Automotive, she kicked off her journey with an educational video designed to spark curiosity across the enterprise. The result was immediate. “Shortly after the engagement, business partners started coming back with use cases. There were 80 use cases on my docket right after that session. That’s the kind of engagement you seek.”

From 80 Ideas to the right few

Wadodkar stresses that the momentum must quickly turn into focus. “You won’t be able to do 80. So, we selected the top things that we can put our energy into.”

This begins the experimentation phase, where feasibility becomes the central question. “Some AI use cases do not need as much experimentation, like GenAI. We are building a bot, which is pretty straightforward. There’s a high chance this is going to work.”

However, she maintains that predictive models are different. “It’s a hypothesis. It may or may not work. The experimentation helps us quantify if this is feasible or not.”

This disciplined filtering protects both time and credibility.

The workflow reality check most teams miss

Wadodkar highlights a mistake many AI teams make: evaluating models in isolation. “Never find the visibility of an AI use case in isolation. You have to pull back and see the whole workflow or the whole process,” she says.

She further offers a manufacturing example: “You are putting a fantastic AI model that you took six months to build, predicting based on temperature and pressure that a particular machine is going to create scrap.”

Technically, the model works. But the real question follows: “What is the operator going to do with it? Is he going to switch it off? Is the model going to start blinking, making sounds? What is it going to do?”

If the answer is unclear, the model has no operational value.“If the operator cannot turn this off, then what is my model doing for you? And I spent six months doing it.”

For industrial environments like TI Automotive’s global manufacturing footprint, operational integration is the difference between experimentation and impact.

3 filters before selecting an AI use case

Wadodkar applies three criteria before committing to any AI initiative.

1. Tie every use case to a business outcome

“It has to be connected to a business case. You look for the biggest bang for your buck. Feel free to say no to others.”

2. Evaluate the entire workflow

Zoom out. Understand how the solution fits into the real process before building it.

3. Buy over build

“Unless you are a software product company, which many of us are not, do a buy-over-build strategy. If somebody has built it and it’s available at a pretty penny, get it. You are saving the overall cost of maintenance, upgrades, all of those things.”

Avoiding the enterprise chatbot explosion

As the conversation shifts to generative AI, Yuravlivker raises the growing excitement around chatbots and agentic systems.

Wadodkar sees both opportunity and risk: “Everybody is in a chatbot frenzy these days.”

Her response is proactive: unify early. “Before my organization goes into that frenzy, I said, let’s come up with a unified chatbot before there are a hundred chatbots, and nobody knows which chatbot I should go to.”

At TI Automotive, the unified assistant is called TIA. It serves as a single conversational front door, hosted in tools like Microsoft Teams and integrated across internal platforms.

Behind the scenes, capabilities continue to expand:

• Creating ServiceNow tickets
• Approval workflows
• Submitting leave requests
• Surfacing HR and legal policies
• Managing approvals

“You can just grow this one face and expand what it can do in the background,” Wadodkar explains. Different teams can continue building their own capabilities while contributing to a single unified interface. “There might be several teams building this offering, and that’s fine. It doesn’t all have to be centralized.” The approach mirrors a microservices model, enabling decentralized innovation while maintaining a consistent employee experience.

She also emphasizes how dramatically the tooling landscape has evolved. “These frameworks and agents are right there. You can just add your code and go on with it.” The impact, she notes, goes beyond technical efficiency. “It’s a big productivity builder for your employees. It keeps them happy.” Instead of navigating multiple departments or systems, employees can simply ask the chatbot.

“I think every CDAO should consider a unified chatbot,” Wadodkar concludes.

In Part Two, Wadodkar explores how AI governance and decision intelligence help scale early AI success responsibly across the enterprise, including the operating model, accountability, and transparency required to build trust.

CDO Magazine appreciates Apurva Wadodkar for sharing her insights with our global community.