Building an AI Foundation, Not Just an AI Stack

Procurement teams are under real pressure to show that they are doing something meaningful with AI. All too often, this leads to a collection of tools that were grabbed quickly and dropped into existing workflows. These ‘solutions’ were left to prove their value without much thought given to whether they were solving the right problem in the first place. 

Hemangi Tawade has seen this pattern up close, and she has spent considerable time thinking about what a more effective approach could look like.

Hemangi spent nearly two decades in sourcing and procurement, working as an SAP consultant, in enterprise transformation roles, and as an IT director of procurement before starting her own advisory practice. Along the way, she completed formal programs at Harvard Business School and elsewhere focused on the leadership and technical dimensions of AI. 

Developing an AI Foundation

In a recent episode of The Sourcing Hero podcast, she brought that foundation to bear on some of the opportunities procurement teams might still be missing.

Understanding the Tool Before Applying It

Hemangi’s interest in AI started not with the outputs but with the mechanics. She wanted to understand why a prompt produced a particular result, what was happening inside the model, and what conditions led to better or worse outcomes for the business. That curiosity drove her formal education, and that education now shapes how she advises procurement teams.

“If I am now using a tool or a product, I need to understand how it works,” she said. The practical value of having a better understanding of the tool you’re using, she says, changes how you evaluate provider claims, how you diagnose when something is not performing as expected, and even how you make decisions about where AI should and should not be applied.

Most procurement professionals have not had access to that kind of training. They get a tool, maybe a workshop, and then they are largely left to figure it out. Hemangi’s perspective is that closing that knowledge gap and empowering teams with a deeper understanding of the technology changes what is possible, not just for individual practitioners but for the programs they build or the outcomes they produce for the business.

Governance as a Guardrail, Not a Gatekeeper

Hemangi pushes back hardest on conventional thinking around governance. In many organizations, mentioning governance evokes thoughts of restrictions. Teams become hesitant to experiment at all, which defeats the purpose of having new technology in the first place.

She explains that governance, when done well, is what gives teams the confidence to try things. It defines the boundaries clearly enough so that people know when they are operating safely, and they can move forward without fear of inadvertently creating a much larger problem.

"Governance steps in and gives you the confidence to fail in a protected environment and not fail majorly without knowing the impact of it.

— Hemangi Tawade

Her recommendation to CTOs and technology leaders is to show up to conversations around risk and governance not as enforcers delivering policy, but as guides helping procurement understand what is possible within a safe operating range.

Start With Monday Morning

On the question of where to begin with AI, Hemangi suggests that procurement should ask themselves a practical question: what does Monday morning look like for your team?

For example, if a category manager is spending three to five days preparing for a negotiation because relevant data is scattered across multiple systems, buried in contract versions, or stored in ways that make it hard to access quickly, that is a real problem with a practical solution. AI as a preparation and support tool, pulling together context before a negotiation rather than conducting the negotiation itself, is both achievable and valuable in a way that is easy to measure.

The mistake she sees repeatedly is teams bringing the same four or five perennial pain points to every transformation cycle without asking whether the approach will be different this time. 

“The same pain points are again brought up while doing your ERP transformations, and if it has not brought the value you expected, bringing the same pain point again to a new transformation will not help,” she said.

Choosing the Right Kind of AI

It’s important for teams to think clearly about which type of AI is appropriate for a given problem. The tendency to treat AI as a single category leads to mismatches between the tool and the task, either over-investing in large language models for problems that a simpler agentic solution would handle better, or under-investing in capability for problems that genuinely require it.

For complex, global supply chain risk management in a volatile environment, large language models that can interpret signals across many data sources have a genuine role. For high-volume, repetitive tasks like three-way invoice matching across millions of line items, a more targeted agentic solution is often more appropriate and far less expensive.

Her recommendation is to go hybrid, evaluating each use case on its own terms: how complex is the problem, how much accuracy is required, what is the cost of error, and how quickly does the solution need to deliver value. 

“Solve the simplest problems first,” she said, “and then you will come closer to the bigger ones.”

That approach builds the organizational muscle and the internal credibility that makes larger AI investments easier to justify and more likely to succeed in the long run.

For more, listen to Hemangi's full episode here: