The Full-Stack Centralized Insights Org
Keep data, research and engineering in one central team — objectivity is the asset you're protecting.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 88%
Netflix's deliberately unfashionable org design for data. Most companies at Netflix's scale either embed data people into business lines (ads, games) or split them functionally (data engineers here, data scientists there, consumer researchers somewhere else). Netflix resisted both and kept one centralized team that is functionally diverse — data engineering, data science, analytics engineering and consumer/user research combined — serving nearly every area of the business. The payoff is objectivity: a team that doesn't report to the people relying on it can tell the truth instead of the story someone wants to hear.
Origin
Netflix's own data and insights organisation, which Elizabeth Stone led as VP of Data and Insights before becoming CTO. The merger of the consumer insights (research) team with data science and engineering into a single full-stack org happened roughly two years before this February 2024 conversation.
Core principles
- 01Objectivity is the point: the job is not to tell the story someone wants to hear with the data, nor to solve the problem someone thinks is most important.
- 02A centralized functional home means you can ask 'are we the world's best data engineers/data scientists?' and keep getting better at the craft.
- 03Combining attitudinal (qualitative research) and behavioural (quantitative data) expertise in one org is a superpower on the right problems.
- 04The cost of centralisation is that it demands extraordinary partnership — people work on problems for teams they don't report to.
- 05Centralisation must be balanced with being a genuinely good partner: deliver the agreed priorities, stay flexible, and earn the agency.
How to run it
- 1
Resist the two default org moves
Do not embed data people into business lines, and do not split them functionally into separate data-engineering, data-science, analytics and research orgs. Keep one centralized, functionally diverse team that works on nearly every area of the business from within.
Watch out This is the harder path. Stone understands why companies move away from it — it requires extraordinary partnership across reporting lines.
- 2
Merge research into the same org as data
Put consumer/user research (attitudinal, qualitative) in the same organisation as data science, data engineering and analytics (behavioural, quantitative). This makes it a full-stack data and research capability rather than two teams producing conflicting narratives.
Pro tip This also solves the credibility problem user research teams commonly face — under the same org as data, the qualitative work stops being dismissed as anecdote.
Watch out Don't overdo it. Not every problem needs both. Stone explicitly says they try not to insist on collaborating everywhere, because that's the wrong expectation.
- 3
Protect and use the objectivity
Because the team doesn't report into the teams relying on it, it can hold its own perspective, be a truth teller, and be curious about problems nobody asked it to look at. Treat that agency as the core asset and expect the team to take it seriously.
Pro tip This is what uplevels the whole organisation — a group whose incentive is truth rather than narrative.
Watch out Objectivity has to be earned continuously. Balance it by delivering on the priorities you agreed and being flexible about where you spend time, or the partner teams will lobby to embed you.
- 4
Bank the functional and career benefits
Use the centralized functional home to push craft excellence, create mobility across problem spaces, and cross-pollinate ideas between domains — all of which are structurally unavailable in an embedded model.
In the wild
With research and data in one org, Netflix can tackle a problem like 'what's the right way to think about recommendations and how best to surface them' by combining attitudinal research — qualitative and quantitative — with behavioural research from the data science, data engineering and analytics side.
→ Stone says the team calls the combination of these skill sets a superpower internally, and that it produced answers she isn't sure they'd have reached under a different org structure.
The consumer insights arm spans content screenings (making titles the best version of themselves before they hit the service), traditional UX research on title discovery and accessibility, internal research on tools for studio productions, and regional teams with local expertise in entertainment consumer needs.
→ A single research capability with global remit that sits inside the same org as the behavioural data, giving both credibility and reach.
Common mistakes
Embedding data teams to make partners happy
Embedding makes the partnership problem disappear — and takes objectivity with it. The team's job quietly becomes telling the story the business line wants to hear, and the truth-telling function is lost.
Splitting research away from data
When user research sits outside the data org, it faces perpetual backlash ('this is all anecdotal') and you get duelling narratives — data says X, research says Y, and nobody can adjudicate.
Mandating collaboration everywhere
Stone explicitly guards against this: not every problem space benefits from the full combined stack. Insisting on it sets the wrong expectation and wastes the superpower on problems that don't need it.
Is it for you?
Best for
Data/analytics/research leaders at scaling companies under pressure to embed or functionally split their team
Not ideal for
Small companies where a centralized team would just be a bottleneck, or businesses where data needs are shallow and purely operational per business line
From the transcript
“I sort of understand why a lot of companies move away from this because it really does require basically extraordinary partnership that we would have…”
“it also allows us to be really objective that is probably the most important thing that our job is not to tell the story that…”
“we could tackle a problem like what's the right way to think about recommendations and how best to surface them in a way that combines…”
“we try not to overdo it and say we need to”
From the episode
How Netflix builds a culture of excellence
Elizabeth Stone (CTO)