R.

Where R fits on a long engagement.

R exists because statisticians wrote it for themselves. That shows in how directly statistical ideas map onto code, and in the depth of its package ecosystem for methods that have no equivalent elsewhere. For time series, survival analysis or econometrics, an analyst in R is frequently faster than the same analyst in a general-purpose language.

It shows equally in what R is not. Production deployment, long-running services and large engineering codebases are not what the language was built for. The common pattern is analysis in R and delivery in something else, and being explicit about that boundary early avoids an argument later.

What an assigned team does with R.

Analysis in R is assigned in a rhythm rather than a sprint. Questions arrive from the business, the work is investigative, and the output is a decision rather than a deployment. That suits a standing assignment far better than a project, because most of the value is in knowing the data already.

The recurring failure is that the analysis is sound and the data underneath it is not. Where that turns out to be the case, the honest answer is to fix the pipelines first, under managed data services, rather than to model around the problem.

What we use R for.

• Statistical work with a defensible method Analysis where the method has to stand up to scrutiny, not just produce a number that looks plausible.
• Experiment and test design Sizing, power and analysis for tests whose result will actually change a decision.
• Regulated or reported analysis Work that has to be reproducible from raw data to published figure, months after it was run.

How R capacity is assigned.

R capacity is assigned under data science. Where the constraint turns out to be the pipelines feeding the analysis rather than the analysis itself, that is a different assignment and we will say so at scoping.