What We Learned Asking Customers About AI for BOM and PLM? 

When customers speak about AI in PLM, they don’t talk about AI at all. It sounds like a paradox, you think. This year was a year when every single PLM vendor introduced “some AI”. While it was very expected, I think the jury is still out. I see some amazing companies and tools. At the same time, I see how (similar to the time of cloud/SaaS), some PLM vendors are dressing up their old tech with “AI features”. I will talk about it in my other Beyond PLM blog, but today, I want to focus on what we learned by asking customers about AI for PLM (and for OpenBOM)

2025 was a year of exploration for us at OpenBOM. We developed OpenBOM MCP, created the beta version of OpenBOM co-pilot chat, and spent a lot of time researching AI, its capabilities, underlying technologies, and what it could realistically mean for PLM. But more importantly, we focused on understanding what customers actually want from AI in their day-to-day work. What we learned came less from demos and experiments, and more from listening carefully to how people describe their problems and how those problems slowly turn into expectations.

One thing I have noticed in almost every conversation about AI and PLM is that customers rarely begin by talking about AI itself. They start by describing what is not working in their daily work, which then leads them to ask, “we need an AI that solves this problem”.  

Only after these problems are described does AI enter the conversation, and even then, it is not introduced as a feature request or a technology discussion, but as a quiet question about whether something can finally reduce this friction.

In 2025, we ran a continuous survey with our users and customers about what they expect OpenBOM AI to do, and today, we want to share some results

To make this more explicit, we asked a simple question and collected responses from survey participants about what they would most like to see from OpenBOM AI. The answers clearly reflected the reality we hear in conversations every week.

From Problems to Expectations

What stood out immediately was that none of the expectations were abstract. They were grounded in everyday tasks that consume time, introduce risk, and require constant manual attention. In that sense, the responses do not describe a future vision of PLM, but rather a list of unresolved issues that companies have learned to live with for years.

The chart does not describe curiosity about AI. It describes fatigue.

Why Automated BOM Creation Comes First?

Automated BOM creation ranked highest by a wide margin, and this result is difficult to misinterpret. BOM creation remains one of the most manual and fragile activities in product development, even in organizations that have invested in modern CAD and PLM systems.

In practice, BOMs are still assembled through a sequence of exports, restructurings, spreadsheet edits, and late corrections, with engineers and operations teams acting as intermediaries between systems that do not fully align. Errors appear not because people lack discipline, but because the process itself depends too heavily on manual coordination.

When companies ask for automated BOM creation, they are not asking AI to make design decisions for them. They are asking it to remove repetitive work that adds little value and creates unnecessary risk, while making product structures more consistent and predictable across teams.

ERP Is Still the Moment of Truth

The second most requested capability was help with sending data to ERP, which continues to be one of the most stressful moments in the product lifecycle. Even when integrations exist, the transition from engineering to manufacturing often triggers hesitation, additional checks, and parallel spreadsheets that exist solely to confirm that the data can be trusted.

What is interesting is that respondents did not frame this expectation as a request for new integrations or different systems. They framed it as a readiness problem. Data exists, but it is not always ready to be used operationally, and AI is expected to help reduce the uncertainty at the moment of release.

This expectation highlights that integration alone does not solve the problem. Confidence does.

From Recording Changes to Understanding Consequences

The proud place #3 in the list of responses pointed toward predictive cost and lead time analysis, change impact analysis, and change impact forecasting. Together, these expectations suggest a shift away from simply recording what changed and toward understanding what will happen if a change is made.

Traditional PLM systems are effective at capturing revisions and maintaining history, but they offer limited help when teams need to anticipate downstream effects before acting. Companies are increasingly asking for help answering questions about consequences, not just compliance, and this requires understanding relationships between parts, suppliers, costs, and schedules rather than treating data as isolated records.

Chat Is Helpful but Not the Main Expectation

An AI chat assistant for quick answers ranked well, but it was clearly not the primary expectation. This aligns with what I see in practice, where conversational interfaces are useful only when the underlying data is already structured, reliable, and meaningful.

Without that foundation, chat becomes another way to surface uncertainty rather than reduce it. Customers do not want AI to speak more fluently. They want to understand their product data more accurately.

The Quiet Importance of Data Quality

Smart data cleanup and error detection appeared lower in the ranking, but it underpins every other expectation. Most organizations are aware that AI cannot deliver reliable results if the data itself is inconsistent, duplicated, or incomplete, even if this concern is not always stated explicitly.

Behind many of these expectations is a simple desire to regain trust in product data, because without that trust, automation increases speed but not confidence.

What These Expectations Really Say

Taken together, these responses make one thing clear to me. Companies are not asking AI to reinvent PLM or to replace engineering judgment. They are asking AI to make PLM finally behave in a way that feels reliable, predictable, and aligned with how work actually happens.

AI is expected to absorb friction that has accumulated over years of manual workarounds, partial integrations, and fragile handoffs, while allowing engineers and operations teams to focus on decisions rather than coordination.

When customers talk about AI in PLM, they are really talking about long-standing problems they have learned to live with, and AI becomes relevant only insofar as it offers a path toward simpler, calmer, and more trustworthy workflows.

That perspective, more than any feature list or technology discussion, defines how companies approach AI in PLM today.

2026 will be another great year when we will continue to explore AI and will expand the AI capabilities of OpenBOM. The current AI Support Agent and chatbot beta is a foundational element of what we do. More will come in 2026.

In the meantime, REGISTER FOR FREE and check how OpenBOM can help you already today.

Best, Oleg

​