Where to Put Design, Data, and Ops: Embedded or Central
At some point a specialist function outgrows the arrangement where one person does all of it, and you have to decide whether those people sit inside the teams they serve or together as a group. The debate is usually conducted as a matter of principle, with the function’s leader arguing for centralisation and the product leaders arguing for embedding, and both citing the last company they worked at. Neither is wrong in general. This page is about the actual trade-off and how to decide it per function rather than as company doctrine.
What’s inside
The real trade-off: craft versus proximity
Three models
How to choose, per function
The hybrid that usually wins
What breaks in each model
The career-path problem nobody mentions
Where Blomma fits
The real trade-off: craft versus proximity
Strip away the advocacy and there is one trade-off underneath.
Centralising buys craft.Specialists sitting together set a shared standard, review each other’s work, build reusable components and conventions, and develop juniors. A designer surrounded by designers gets better faster than one surrounded by engineers. This is not a preference; it is how craft skills develop.
Embedding buys proximity.A specialist inside the team hears the customer conversation, understands the constraint before the brief is written, and can make small decisions without a handoff. They also build trust, which is what turns a specialist from a service provider into a participant.
You cannot have both at full strength, and every model is a position on that line. Which means the useful question is which one your current work is more sensitive to. Work where quality of craft is the differentiator suffers more from embedding. Work where speed of iteration and context are the differentiator suffers more from centralising.
One more thing the debate obscures: the answer differs by function inside the same company, and it changes with stage. Adopting a single company-wide doctrine is usually the worst available choice.
Three models
Fully central — the function is one team, work arrives as requests.Specialists report to the function lead, are allocated to projects, and move between them. Strongest craft, clearest standards, best development path. Costs the most in context and handoff, and reliably produces a service-desk dynamic where the function is treated as a supplier rather than a partner.
Fully embedded — specialists report into the teams they serve.Strongest context and speed. Costs craft consistency, and it isolates individuals: a lone designer or data person on a team has nobody to learn from, no review, and often no manager who understands their work well enough to develop or evaluate them.
Dotted-line hybrid — specialists sit and work with their teams, and report to (or are professionally managed by) the function lead.The most common arrangement at scale and the most frequently botched, because it works only if the two lines have genuinely different remits, which is rarely specified.
Most companies pass through all three. The failure is not choosing the wrong one; it is choosing one and never revisiting it as the work changes.
How to choose, per function
Four questions, answered per function rather than for the company.
How coupled is the work to a single team’s context?Product design is tightly coupled — the designer needs the customer conversation. Brand design is not. Analytics supporting one product is coupled; data platform work is not. Tightly coupled work argues for embedding.
How much does consistency across the company matter?If users experience the output as one thing — a single interface, a single set of metrics definitions, one voice — consistency is a real requirement and central wins. If each team’s output is independently consumed, it matters less.
How many specialists do you have?Below about four in a function, embedding isolates them and you lose the development entirely. Central is usually right until you have enough people to embed without stranding anyone.
Who currently owns the standard?If the answer is nobody, centralise until someone does. A function with no established standard cannot be embedded without the standard dissolving.
Run those four on design, data, research, ops, and any other specialist function separately. You will typically get different answers, and that is correct rather than inconsistent.
The hybrid that usually wins
For most companies past about eighty people, some version of the hybrid is right. What distinguishes the ones that work is specificity about which line owns what. Worth writing down explicitly.
The team owns:what the person works on, priority, day-to-day direction, and whether the outcome shipped. The specialist is a full member of the team, in its rituals, accountable to its goals.
The function owns:the craft standard, hiring bar, review and critique, career development, levelling, and — importantly — the performance assessment of craft quality.
Named jointly:compensation and promotion decisions, with the function lead holding the craft judgment and the team lead holding the impact judgment.
The two failure modes are both about vagueness. If the function line has no real authority, the hybrid is embedding with extra meetings, and craft degrades on schedule. If the team line has no real authority, the hybrid is central with extra seating, and you get the service-desk dynamic anyway. Write the split down and tell both leads.
What breaks in each model
Central breaks at the interface.Requests arrive under-specified, the function pushes back, both sides conclude the other does not understand the work. Symptom: intake processes, prioritisation committees, and the phrase “we need to be brought in earlier.” That phrase is diagnostic — it means the model has put the specialist outside the moment where the decision is actually made.
Embedded breaks at the standard.Six teams produce six conventions, the product feels assembled from parts, and metric definitions diverge until two teams report different numbers for the same thing. Symptom: nobody can say what good looks like, and quality varies by which team you look at.
Embedded also breaks the individuals.The lone specialist has no peer review, no development, and a manager who cannot evaluate their craft. This is the cost that shows up last and hurts most, usually as the quiet departure of your strongest specialist.
Hybrid breaks on ambiguity.When both lines believe they own priority, the specialist is caught between two managers and resolves it by pleasing whoever is most present. Symptom: a specialist who seems stressed and is producing less than their capability suggests.
The career-path problem nobody mentions
One consequence of this decision that rarely enters the discussion and matters more than most of the above.
Whoever owns the reporting line owns the person’s development. A designer embedded under an engineering manager has a manager who cannot assess their craft, cannot coach it, and will not know which skill they should build next. That person’s growth stops, quietly, and it is usually attributed to them rather than to the structure.
The same applies in reverse: a specialist in a fully central team may have excellent craft development and no exposure to the business context that would make them senior.
So whichever model you pick, name explicitly where development comes from for these people. If the reporting manager cannot assess the craft, someone else must — a function lead, a senior peer, a mentor, or coaching that does not depend on the manager’s expertise. Leaving it implicit is how companies lose specialists at exactly the seniority where they become expensive to replace.
Where Blomma fits
The practical difficulty is that this decision is argued by people with a stake in the answer. Your function lead wants centralisation partly for legitimate craft reasons and partly because it is a larger, clearer role. Your product leads want embedding partly for legitimate speed reasons and partly because it gives them control of resource. Both are advocating sincerely and neither is neutral, and you are the only person in the discussion without a position — except that you also have a preference from wherever you saw this done before.
Blomma is an always-on AI career coach with no stake in your structure. Use it to run the four questions per function rather than adopting a doctrine. Use it to write the hybrid split specifically enough that it holds — which is the step that determines whether the hybrid works or becomes the worst of both. And use it to prepare the conversation with whichever leader does not get the model they argued for, which is the part of this decision that actually determines whether it lands.
The second application is the career-path problem above. Where a specialist’s manager cannot develop their craft, coaching that does not depend on the manager’s expertise is the difference between keeping that person and losing them in eighteen months..
Embedded or central is not a question of philosophy. It is a trade between craft and proximity, answered per function, revisited as the work changes — and it carries a development consequence that outlasts the structure itself.