UX Research Team Structure: Centralized, Embedded, and Hub-and-Spoke Models Compared (2026)

The short answer

There are four viable research operating models, and the right one is decided by two variables: how fast product teams need answers, and how much standardisation your decisions require. Centralized wins on rigor and career development; embedded wins on speed and influence; hub-and-spoke is what most organisations converge on as they grow; ops-enabled democratization is what AI-native tooling has made viable below the headcount that used to be required.

The benchmark data is unusually clear on two points. First, adoption is almost evenly split: Nielsen Norman Group's survey of 557 UX and design professionals across 356 companies in 59 countries found 29% centralized, 32% decentralized, and 31% hybrid/matrix (NN/g) — nobody has won this argument. Second, researchers specifically are rarely on a research team at all: only 6% of researchers sit on a dedicated User Research team, and nearly a third report into Product (User Interviews, State of Research Strategy).

That second number is the one that should drive your design. If research is already distributed, the question is not "centralize or embed" — it is what the centre owns when almost everyone doing research reports somewhere else.

---

The four models

1. Centralized

All researchers sit on one team under one manager and take intake from product groups.

Strong when: methodological rigor matters (regulated products, safety, high-stakes decisions); the team is small enough that context-switching is cheaper than duplication; you need to grow junior researchers. Fails when: intake becomes a queue. The most common failure is the research team becoming a service desk with a six-week backlog, and product teams routing around it with a survey tool and a weekend.

2. Embedded / decentralized

Each researcher sits inside a product team, reporting either into that team or dotted-line to a research lead.

Strong when: speed and influence matter more than consistency; researchers need deep domain context; the org runs continuous discovery. Fails when: it fragments. Among decentralized teams, 43% align to products and 27% to internal departments or business lines (NN/g) — and without a shared standard, each alignment invents its own screener, its own consent language, and its own repository. Career growth also stalls: an embedded researcher with no research manager has no one who can evaluate their craft.

3. Hub-and-spoke (hybrid/matrix)

Researchers are embedded in product teams but report — solid or dotted line — to a central research lead who owns standards, tooling, and development. NN/g's data shows hybrid structures correlate with larger organisations (statistically significant), which matches the lived pattern: teams start centralized, embed under pressure, and rebuild a centre once quality problems appear.

Strong when: you have more than roughly four researchers and more than one product line. Fails when: the dotted line has no teeth. If the hub owns standards but not performance reviews, tooling budget, or hiring, it owns nothing.

4. Ops-enabled democratization

A small central core (often one to three people) owns standards, tooling, participant operations, and quality review, while PMs, designers, and support staff run most studies themselves on a platform that enforces the method.

This is the newest model and the fastest-growing. 71% of organisations now have people conducting research who are not researchers, and the share of organisations where research is essential to strategy at all levels nearly tripled in a year, from 8% to 22% (Maze, 2026). See the research democratization playbook for the enablement side.

Strong when: demand massively exceeds researcher supply — which is the normal condition. Fails when: the platform does not enforce quality. Democratization without a method-enforcing tool is just more bad research, faster.

---

Model comparison

---

Where research should report

Reporting line determines what research is allowed to influence. NN/g's data shows 27% of design organisations report into product management and 23% directly into the C-suite; research specifically most often sits under Product.

• Under Product — fastest path to roadmap influence; the risk is that research only gets asked questions Product already thought of, and evaluative work crowds out generative work.
• Under Design — protects craft quality and gives researchers a manager who can assess their work; the risk is being scoped to interface questions rather than business ones.
• Under a C-level Insights or Strategy function — the only structure where research routinely shapes decisions above the roadmap; rare below a few hundred employees.
• Under Marketing or Growth — fast funding, but the questions skew to messaging and acquisition.

A practical test: if research reports into the function whose plans it most often contradicts, expect findings to get softened. That is an argument for a dotted line to someone independent, not for a reorg.

---

Staffing math: how many researchers do you actually need?

NN/g's ratio benchmark is roughly 1 researcher : 5 designers : 50 developers — improved from 1:5:100 — with the important caveat that ratios describe a state, not a maturity score. NN/g is explicit that "ratios are not maturity scores." The complementary NN/g finding is that headcount tracks organisation size mechanically: on average one additional designer per extra 200 employees.

Use the ratio as a sanity check, not a target. The better question is throughput: how many decisions per quarter require evidence, and what is the largest number of studies a researcher can run well? Human researchers manage roughly four to six in-depth interviews a day before fatigue degrades quality, and a traditional study runs about six weeks end to end.

For the cost side of this calculation — salary bands by level and region, fully loaded cost, and research budget as a share of ARR — see UX researcher salary and team cost benchmarks. For the build-the-team decision, see hiring a UX researcher and agency vs. in-house.

---

How AI changes the org design

Every model above was designed around a constraint that no longer holds: a researcher can only moderate one session at a time, and analysis takes longer than fieldwork.

When moderation and synthesis are automated, three things change structurally:

1. The hub can serve more spokes. The central function's scarce resource stops being interview hours and becomes study design and quality review — which scales far better. A one-to-three-person core can credibly support a dozen product teams.
2. Democratization stops degrading quality. The historical objection to non-researchers running studies is that they ask leading questions and over-generalise from four conversations. A platform that enforces the method — structured question types, consistent probing, automatic thematic analysis, an interview quality score — removes most of that risk. Koji scores each conversation 1–5 and only counts sessions at 3 or above.
3. The centre's real deliverable becomes templates and standards, not studies. The highest-leverage artefact a central team ships is a validated study template that any PM can launch on Monday.

Concretely with Koji: the central team owns the research briefs and the six structured question types — open_ended, scale, single_choice, multiple_choice, ranking, yes_no — so every team's data is comparable and aggregatable rather than trapped in one team's notes. Spokes launch AI-moderated text or voice interviews without booking a moderator, and a customisable AI consultant carries the organisation's context and tone into every study. Reports are generated in real time, so the hub reviews finished analysis instead of transcribing. Legacy stacks make the opposite trade: SurveyMonkey or Qualtrics lets anyone launch a survey but nobody enforce a method, while a traditional agency enforces method but only for the studies you can afford.

Pricing supports the model too: at €29/month (Insights) and €79/month (Interviews), a spoke team can be given a seat for less than the cost of a single recruited participant, and credits are consumed only by conversations that clear the quality gate.

---

A worked example: a 40-person product org

Four product teams, two researchers, demand for roughly 20 studies a quarter and capacity for six.

Diagnosis. Centralized by default. Six-week backlog. Two of the four teams have quietly started running their own unmoderated surveys, and nobody can find last quarter's findings.

Design chosen: hub-and-spoke with ops-enabled spokes.

• Hub (2 researchers, one designated lead): owns the study template library, screener and consent standards, the repository, quality review of every published finding, and all high-stakes generative work.
• Spokes (1 trained PM or designer per product team): run evaluative and continuous-discovery studies from hub-approved templates on Koji, with the hub reviewing before findings are published to the repository.
• Reporting: researchers report solid-line to the research lead, dotted-line to their product group. The lead owns hiring, craft review, and the tooling budget — the dotted line has teeth.
• Quality gate: no finding enters the repository without hub sign-off; every study uses structured question types so results aggregate across teams.

Result to expect: study throughput rises well past the six-per-quarter ceiling because the two researchers stop moderating routine sessions, while method consistency improves rather than degrades — the spokes are running the hub's templates, not inventing their own.

---

Changing models: a 90-day plan

Days 1–30 — Diagnose. Inventory every study run in the last two quarters, including the ones run without the research team. Count decisions that needed evidence and did not get it. Map current reporting lines and where findings live. Interview four to six stakeholders about what they do when they need an answer this week — see stakeholder buy-in.

Days 31–60 — Design and pilot. Pick the model, write down what the centre owns (standards, tooling, participants, quality review, hiring) and what the spokes own. Publish three study templates. Pilot with one willing product team rather than announcing a reorg.

Days 61–90 — Institutionalise. Fix the reporting lines, set the quality gate, agree how research capacity is requested and prioritised, and instrument the outcome: studies per quarter, time from question to answer, and share of decisions with evidence attached. Review the model every two quarters — structures that fit at 40 people break at 120.

---

Common mistakes

1. Reorganising instead of fixing intake. Most "we need to embed" conversations are really a prioritisation problem wearing a costume.
2. A hub with no authority. If the centre does not own standards, tooling budget, and craft review, it is a mailing list.
3. Embedding a single researcher with no research manager. Their craft stops developing and they leave within a year.
4. Democratizing without a method-enforcing platform. Volume goes up, evidence quality goes down, trust in research goes with it.
5. Using ratios as targets. They describe the current state of a sample, not what your decisions require.
6. Never revisiting. The model that fits your org today is a snapshot; the NN/g size correlation says you will need a different one after the next two hiring waves.

---

Frequently asked questions

What is the most common UX research team structure? There is no majority model. NN/g's survey of 557 professionals across 356 companies found 29% centralized, 32% decentralized, and 31% hybrid/matrix. For researchers specifically, only 6% sit on a dedicated research team — most are embedded elsewhere, with nearly a third reporting into Product.

What is a hub-and-spoke research model? Researchers are embedded in product teams for day-to-day work but report to a central research lead who owns standards, tooling, participant operations, hiring, and craft review. It is the model organisations typically converge on past four or five researchers and more than one product line.

How many researchers should we have? NN/g's benchmark ratio is roughly 1 researcher to 5 designers to 50 developers, improved from 1:5:100 — but NN/g cautions that ratios are not maturity scores. Size the team from decision throughput instead: how many decisions per quarter need evidence, and how many studies a researcher can run well.

Where should UX research report? Most often Product, which buys roadmap influence at the cost of being scoped to questions Product already asked. Design protects craft quality; a C-level insights function gives the most strategic reach but is rare below a few hundred employees. Whatever the line, keep an independent path for findings that contradict the plan.

Does democratization mean we do not need researchers? No. 71% of organisations already have non-researchers running research; the effect is to change what researchers do — from moderating every session to owning standards, templates, quality review, and the hardest generative work.

How does AI change research team structure? It lifts the constraint the models were built around. With AI-moderated interviews and automatic thematic analysis, a one-to-three-person core can support a dozen product teams, and the centre's main deliverable becomes validated study templates and a quality gate rather than individually run studies.

---

Related resources

• Structured Questions Guide — the six question types that make findings comparable across teams
• Research Democratization Playbook — enabling non-researchers without losing rigor
• Hiring a UX Researcher — when to hire, levels, and the interview loop
• UX Researcher Salary and Team Cost Benchmarks — the cost side of the staffing model
• UX Research Agency vs. In-House — the four sourcing models and their TCO
• ResearchOps Guide — the operational layer every model depends on
• Stakeholder Buy-In for User Research — securing the mandate before you restructure