AI Strategic Sourcing: Just A More Expensive Spreadsheet?

Part I: The Problem

Key Takeaways:

Most AI approaches in procurement just recreate the same old problem in a new form. Generic AI hands procurement teams a more sophisticated problem disguised as a solution. It swaps the fragility of a legacy spreadsheet for the complexity of a system that demands constant prompting, expert interpretation, and ongoing maintenance, without reliably delivering the sourcing outcomes that justified the investment in the first place.
The real cost of the wrong AI approach isn’t what you spend maintaining it — it’s the direct material cost savings that never make it to the P&L. A purpose-built solution like aiSource tells you why it’s high; which specific cost drivers in machining time, material grade, or overhead structure are inflating the quote, and gives buyers the fact-based ammunition to close that gap in the negotiation itself.
Direct materials sourcing teams are being asked to integrate AI into their organizational model; one that will impact how efficiently they can move a part from RFQ to negotiated close, how many cost reduction opportunities their buyers can actually handle in parallel, and how consistently they can capture savings across a team with varying experience levels. Most are making that decision without a clear framework for evaluating whether an AI approach will actually move those numbers. This two-part series examines why.

Part 1 looks at the three paths available today: building in-house, adapting general-purpose AI, and adopting purpose-built solutions, and the trade-offs each one carries. Part 2 goes inside the purpose-built approach to explore what it means to design AI around the direct materials sourcing cycle itself, and why that distinction changes everything downstream.

The Full Article-Part 1

80% of digital transformation initiatives fall short of expectations — and layering AI on top doesn’t automatically change that.

Digital transformation in procurement has never been simple. Introduce artificial intelligence (AI) in procurement, and the complexity only increases.

Today, procurement leaders are being asked to make a critical decision. Do we build our own AI capabilities, adapt a general-purpose platform, or adopt a solution designed specifically for sourcing? Each path promises progress. But not all of them deliver transformation.

Direct materials sourcing teams are being asked to integrate AI into their organizational model — one that will impact how efficiently they can move a part from RFQ to negotiated close, how many cost reduction opportunities their buyers can actually handle in parallel, and how consistently they can capture savings across a team with varying experience levels. Most are making that decision without a clear framework for evaluating whether an AI approach will actually move those numbers. This two-part series examines why.

The Familiar Pattern

From cost models to AI models, it’s the same story on repeat.

Most procurement teams have lived through a version of it. Often, they’ve relied on a cost model built to solve a specific problem. Expanded over time to handle more complexity. Maintained by a small group of people who truly understand it. And increasingly impossible for anyone else to trust, scale, or update.

In other words, it works until it doesn’t.

What begins as innovation quietly becomes infrastructure: fragile, resource-intensive, and hard to evolve. The team that built it moves on. The market shifts. The model stays where it is.

Procurement leaders are left to manage a tool that is no longer an asset, but instead, a liability. And many organizations are unknowingly repeating this exact pattern with AI. Most AI approaches in procurement just recreate the same old problem in a new form.

Three Paths, Three Trade-Offs

So, what can procurement leaders do? They need to either build, adapt, or adopt.

Here are three scenarios that define those outcomes:

Path 1: Building AI in-house: maximum control, full responsibility

Flexible
Slow to scale
High maintenance
Expert-dependent
Custom-tailored to their exact needs at that moment, but not beyond it/not equipped for long-term goals

Building internally offers flexibility. You own the system, control the data, and can theoretically customize everything. But it also recreates every challenge procurement teams have already lived through with manual tools, including:

Slow to deploy and scale across the organization
Continuous model and data maintenance to stay effective
Heavy reliance on specialized technical expertise
Difficulty ensuring consistency across different users and teams

The result is often a highly sophisticated system that behaves a lot like a legacy cost model. But one that is difficult to maintain and harder to trust.

The team that built it is the only team that truly understands it. And the moment those people leave, or the market shifts faster than the model can be updated, the organization is back where it started.

What begins as maximum control tends to become maximum responsibility — with diminishing returns over time.

Path 2: Adapting generic AI (LLMs): accessible, but not always actionable

Fast to start
Shallow in depth
Expert-dependent
Unreliable at scale

The rise of tools like ChatGPT and other forms of generative AI, driven by natural language processing, has made it feel immediately within reach. Many procurement teams are asking a fair question: Why not just use a general AI tool with our existing data? It’s accessible, fast, and feels like a reasonable starting point.

But generic AI hands procurement teams a more sophisticated problem disguised as a solution. It swaps the fragility of a legacy spreadsheet for the complexity of a system that demands constant prompting, expert interpretation, and ongoing maintenance — without reliably delivering the sourcing outcomes that justified the investment in the first place. In practice, teams consistently run into the same walls:

Lack of depth in complex sourcing scenarios: General-purpose AI often requires significant prompting and several iterations to reach the specificity needed for a meaningful negotiation. A significant share of sourcing automation opportunities simply can’t be fully developed with a model that wasn’t built for this context.
Data challenges — too little or too much: Static, siloed data points with poor data quality , or a lack of reasoning behind that data, may not contain enough structured information to guide the AI effectively. But feeding too much unstructured data creates misinterpretation, incorrect prioritization, and outputs that feel authoritative but aren’t.
Inconsistent reliability: Users must constantly evaluate whether outputs are valid, relevant, or subtly wrong. The risk isn’t always visible upfront. It tends to surface during execution, when it matters most, and there’s less room to course-correct.
Expert-dependent usage: Getting real value requires knowing which data to use and when. It also necessitates crafting prompts that elicit the right depth and recognizing when an output is flawed. That’s not a skill most buyers have, nor should it have to be.

The result is more powerful tools. However, there is still a fundamental reliance on human interpretation, validation, and rework.

Only the most experienced, engaged users consistently extract value, which limits what can actually scale across the organization. Most importantly: the real cost of the wrong AI approach isn’t what you spend maintaining it. It’s the direct material cost savings that never make it to the P&L.

Missed opportunities that aren’t fully developed. Savings that aren’t fully realized. Inconsistent outcomes across teams.

General AI shifts the burden from spreadsheets to prompts. The expertise requirement doesn’t disappear — it just changes form.

Path 3: Purpose-Built AI for Strategic Sourcing: Designed For The Outcome

Actionable insights
Scales to every buyer
Vendor-accountable

A different approach is emerging. It’s one that starts not with the technology, but with the sourcing decision itself. Rather than asking teams to build systems, configure workflows, or interpret outputs, purpose-built AI is designed to deliver consistent, actionable insights grounded in real procurement data and real sourcing use cases.

The key distinction is who carries the burden of expertise, especially when dealing with complex machine learning models. With generic AI, that burden stays with the user. With purpose-built AI, the system carries it. It means the analytical rigor, domain context, and reliability of the output don’t depend on who’s running the process.

A purpose-built solution like aiSource doesn’t just tell you a supplier quote is high — it tells you why. Which specific cost drivers in machining time, material grade, or overhead structure are inflating that quote. And it gives buyers the fact-based ammunition to close that gap in the negotiation itself. That’s the difference between AI-generated insight and realized value on the P&L.

Unlike building in-house, purpose-built solutions come with something equally valuable: accountability. Since you are not maintaining a system and absorbing every failure internally, you can hold a vendor to a standard. Challenge them when outputs fall short, expect improvements over time, and operate against an established benchmark rather than one you have to define and defend yourself.

The Pattern Worth Breaking

Every generation of procurement technology has promised to remove complexity. Spreadsheets replaced paper. ERPs replaced spreadsheets. Advanced analytics replaced gut instinct, at least for the teams that could afford the expertise to run them.

Every wave delivered real value. And each one quietly recreated the same underlying problem: the insight lived in the tool. However, the judgment still lived in the person. Scale the tool, and you scaled access. You didn’t scale outcomes.

AI is following the same arc. Unless procurement leaders deliberately choose to break it.

The question worth sitting with isn’t which AI approach is most sophisticated. It’s which approach actually changes where the expertise lives? Because that’s the only change that scales. And it’s the only change that shows up as realized savings — not as potential, not as outputs, but as dollars that actually make it to the P&L.

In Part 2, we go inside that question, looking at what it actually means to build AI around the sourcing decision itself, and how purpose-built AI works differently at the level of the buyer experience. We’ll also explore why the gap between “AI-generated insight” and “realized value” almost always comes down to one thing: whether anyone trusted the output enough to act on it. Be sure to read Part 2, coming soon.