I once inherited a research setup where every product team had its own embedded researcher, each one reporting into a different roadmap, each one running their own version of the same studies. It made sense on paper. Faster access, tighter context, no shared queue slowing anyone down. Then two teams presented conflicting findings on the same user segment in the same week, and neither had any idea the other study existed until someone in the room asked why the decks disagreed.
That’s not really a story about picking the wrong structure. Embedded was probably right at the size the company started at. It’s a story about nobody deciding when it would stop being right.
Centralize or embed gets treated like a values question, the kind of thing you decide once and defend from then on. It isn’t. It’s closer to a lease with no end date written on it. What’s right for one researcher and three stakeholders is not what’s right for five researchers and twelve.
Strip the philosophy out and the actual decision is simpler: how many product lines need research at the same time right now, and does their demand overlap enough that one shared approach could serve all of them, or diverge enough that it would serve none of them well. Embedded wins when the questions genuinely differ team to team and speed matters more than consistency. Centralized wins when the questions rhyme across teams and consistency, one panel, one standard, one place where institutional knowledge actually accumulates, matters more than speed.
Most companies get the first call right. What they skip is deciding, at that same moment, what would tell them the call needs revisiting. Embedded research that fit at three researchers can quietly turn into five disconnected one-person shops solving the same problem five different ways. Centralized research that fit for one team can become the thing a second team routes around the moment their needs pull hard enough in a different direction.
The fix isn’t getting the first pick right. It’s naming, up front, what would signal it’s time to pick again. A headcount threshold. A specific kind of duplicated work. A complaint that shows up more than once from more than one team. Whatever it is, name it before you need it. By the time the structure is visibly broken, you’re already past the point where changing it is easy.
A couple of signals worth watching for, whichever way you’re currently set up:
Overlapping demand. If two or more teams are asking research questions that rhyme, similar audiences, similar decisions, staying embedded is quietly costing you the coordination a shared function would give you for free.
Findings nobody’s cross-checking. If two teams could reach opposite conclusions on the same question and neither would find out until a decision was already made, that gap is the actual risk, not the structure itself.
Neither signal argues for one structure over the other. They argue for treating the choice as something you keep managing, not something you pick once and stop thinking about.
This is the first in Who’s Actually Running This? Start with the series introduction here. Next: Build Versus Buy
Ground Truth is written by Shannon Coleman, PhD. Strategy and research leader with a background in psychology and quantitative methods. Her work spans consumer product contexts, with particular depth in enterprise AI platforms. Portfolio: shannonlcoleman.com


