I spent two days at the American Medical Device Summit at the end of October, and three phrases followed me from room to room: patient safety, product data, shared accountability.
Patient safety is the stake every function agrees on. The other two are where the agreement ends.
I am not a medical device specialist. I have spent my career in PLM, most of it with regulated manufacturers, and I went to listen. What I kept noticing was a mirror.
The engineers and operations people I have worked with for years describe quality as the tail wagging the dog. The quality and regulatory people who filled the Summit described themselves as the bad guys, the ones who have to say no. Same wall, two sides of it.
The program kept arguing that compliance should not belong to quality and regulatory alone, and I agree. But I came home convinced of something more specific.
Shared accountability is a data problem before it can become a culture problem. The behavior change is real work, and I will come back to it. It cannot start while two functions are looking at two versions of the product.
Shared accountability was the word of the week
Some decisions that can create a compliance problem are made upstream: in design, in sourcing, and in the promise marketing makes about what the device will do. The people in the room wanted engineering and product leadership to own the outcome with them.
The push and pull is three-way, which is what makes it hard.
- Marketing wants to promise the most the device can do.
- Engineering has feasibility and cost limits.
- Regulatory needs the device to fit the box the FDA has set, or at least to land on a position it can justify.
Shared accountability has to cover all three or it is a slogan.
In practice, quality and regulatory are still the gatekeepers. The audience was mostly regulatory and quality people rather than product development, and a good share of my booth time went to explaining what PLM is and that it can carry their work at all. They don’t know what they don’t know. The engineering side rarely knows what quality is missing either.
You cannot share accountability across different versions of the truth
How are you going to have shared accountability if you don’t have shared perspective? If engineering, quality and regulatory are looking at different datasets, they will have different KPIs, and each of them will be right by their own numbers. It’s not possible.
If shared accountability is the destination, shared visibility and shared data availability are the first steps on the path.
So here is a practical test. Can an engineer, a quality manager and a regulatory lead open the same product record and see:
- the same change history
- the same where-used
- the same as-designed against as-built
- the quality record attached to the part, rather than filed next to it
If the answer is no, accountability is a meeting, and it stays a meeting.
The fracture I kept hearing about is between the quality management system and the PLM. The two rarely know about each other. A quality record points at a part number, and the part has no idea the record exists. Some of the quality leaders I spoke with did not know a PLM could hold quality at all, or connect to the system that does.
Take one example. A new regulation forces a part change. The part change forces an SOP change, and the SOP change forces a documentation change.
When the part, the procedures that reference it and the documents that describe it are linked, a change to one flags the others. That linking can live in one system or across several. What matters is that the relationships are kept, not reconstructed.
When they are not, someone has to remember, and the week they are on leave is often the week the thread breaks. Those breaks are the practical case for shared data, and they are the same breaks that show up when there is no one product truth across systems.
AI was everywhere, and the data foundation kept interrupting it
Every session had AI in it. In every conversation about AI, the data foundation surfaced as the prerequisite. It is the unglamorous part, the one most AI roadmaps skip past, and it is where the results get decided.
Send AI into a lake of unstructured product data and you get wrong answers faster, or you get lucky once in a while. Structure the data deliberately, with the relationships and the meaning set up, and you are set up to succeed sooner.
AI run over a fragmented record scales the fragmentation.
Data migration has always forced a company to look at the state of its data. The AI conversation adds two questions: what the data means, and how it relates. Most organizations have not done that contextualization work yet, and saying so plainly seemed to land harder with executives than any demo would have.
We made the longer version of the argument earlier this year in AI readiness is a people and data problem, so I will leave it there.
Defensible means traceable
Compliance is not going anywhere. But what I heard in conversation after conversation, especially around devices with embedded software and AI, was that the question regulators ask is less “did you follow the rule” and more “show me why you did what you did”.
The FDA’s recent guidance reads the same way. Its January 2025 draft guidance on AI-enabled device software, still a draft, describes a comprehensive approach to managing risk throughout the device total product life cycle. Its final guidance on predetermined change control plans lets a manufacturer implement the modifications described in the plan without a new marketing submission for each one, once the FDA has reviewed the planned changes, the methods and the impact assessment.
Both put the weight on evidence you can produce on demand.
Defensibility is traceability: who changed what, when, under which justification, and what it touched. Traceability is a data question before it is a policy question. If the change history lives in one system, the SOP in another and the risk file in a third, the defense is a reconstruction project every time someone asks.
The part nobody budgets for
Change management came up in about three of every four conversations I had, and I will admit I steered a few of them there. We know about people.
Shared accountability is a behavior. A system makes it possible rather than making it happen. A product record that all three functions can open is necessary. Three functions that actually open it is the work.
That work is organizational change management, the line item that is usually trimmed first. We have seen enough rollouts to know that go-live is not adoption.
What this means for the engineering leader
If you lead engineering or product development, you already own the record. The design history, the change orders, the where-used, the as-designed: they live in your systems and your people keep them.
The ask from the Summit is to open that record up and treat quality and regulatory as co-owners of it, rather than as the function that receives the file at the end.
Here is a test you can run this week.
- Pick one product change from the last quarter.
- Ask an engineer, a quality manager and a regulatory lead to find its rationale, the records it touched and the approvals it carried, each from their own desk.
- Time how long it takes and count how many people they had to ask.
That number is your shared accountability gap.
It also changes how you judge your PLM. Judge it on whether it carries the whole lifecycle rather than the CAD vault: the quality record attached to the part, the regulatory position attached to the change, the SOP that references both.
Shared accountability across engineering, quality and regulatory works when every function is working from the same product record. That record is the same foundation your AI program is going to need anyway. Build it once, for both.
If you want an outside view of where the gaps sit between what engineering, quality and regulatory can each see today, talk to our team.
