I have led analytics teams inside Fortune 100 companies and inside government, and I have advised more than I have led. The teams that struggle tend to struggle in patterned ways. There are four failure modes I see often enough to name them.
The reactive shop
The reactive shop is built around ticket throughput. Stakeholders file requests. Analysts pull numbers. The team measures itself by volume and turnaround time. There is no roadmap, only a queue.
This team feels productive from the inside and looks invisible from the outside. Leadership cannot describe what the team is working on at any level higher than the individual ticket. When budget conversations come around, the team has no story to tell about value created. The work was real, but it was diffuse, and diffuse work does not survive a downturn.
The fix is not harder. It is rationing. Move ad hoc requests into a clearly bounded service tier with a published SLA, and reserve the majority of capacity for product work that has a name and an owner.
The platform team that forgot the customer
The opposite failure is the team that has decided it is a platform team and stopped doing customer work. The roadmap is full of internal frameworks. The team is rebuilding ingestion for the third time. There are no live dashboards in front of users.
The internal language is sophisticated. The external value is hard to find. When the head of product is asked what analytics shipped this quarter, the answer is internal scaffolding the product team cannot see. This is a team that is grading its own homework on a curve only it can read.
The fix is to put a customer-facing deliverable on every sprint, even if it is small. Platform investment is real, but it has to be paid for in user-visible outcomes, or it is being financed on trust the team has not yet earned.
The model factory with no production line
The model factory ships notebooks. The data scientists are talented. The papers cited in standups are recent. The number of models in production is zero, or close to it.
This is not a science failure. It is a deployment failure. The team has not built or has not been given the infrastructure to move a model from notebook to production safely. Every project ends at the prototype stage because there is no path beyond it.
The fix requires a partnership the team usually does not have. Either an MLOps function gets built next to the data science team, or the data science team learns to ship and the org restructures around that capability. Either way, the question "how do we get a model into production by next quarter" needs an answer at the org level, not at the IC level.
The dashboard maintenance shop
The dashboard maintenance shop has 400 dashboards. About 30 are viewed weekly. The other 370 are viewed by the people who built them and no one else. The team's calendar is consumed by request intake, dashboard updates, and broken pipeline triage.
This team has been promoted into operational support without anyone noticing. The original analytical mandate has been displaced by maintenance overhead. There is no time for new work because the existing surface area is unmanageable.
The fix is portfolio management. Audit every dashboard. Sunset the ones with no recent views. Move recurring requests into self-service. Negotiate a maintenance budget separately from the new-work budget, and protect both. If you do not protect the new-work budget explicitly, it will be eaten by maintenance, and the team will spend its career inside the queue.
The diagnostic question
When I assess an analytics team, I ask leadership two questions. What did the team ship in the last quarter that a customer or operator can name. What is in the next quarter that has the same property.
If the answer is a list of internal initiatives, the team is in one of these four modes. The path back is rarely a reorg. It is usually a portfolio decision about what work counts and what does not, made by leadership and held to over multiple quarters.