Data Scientist

You know the question. What you want to know is whether the answer is true, and how sure you can be of it. The Data Scientist answers that. It takes clean data, works out what kind of data it is, runs the statistics that fit, and gives you a number you can rely on.

Many numbers look convincing and still mean little. A jump in sales might be one lucky week, not a trend. Like a good doctor reading test results, the Data Scientist looks at the numbers, rules out the noise, and only then gives its view. Like a navigator checking the map before the turn, it confirms the ground is solid before it tells you which way to go.

It is careful about what it claims. Every answer comes from a calculation it ran on your data. It names the test it used and the size of the sample, and it never treats a link between two things as proof that one causes the other. If the data cannot answer your question, it says so instead of guessing.

Think of a referee at a match. The players make the plays, but the referee applies clear rules, keeps the time and decides whether a goal counts. The Data Scientist takes that role for your numbers. It applies the rules, marks what counts, and says plainly when a call is too close to make.

• Work out what kind of data you have, such as sales and customer records or a general table, and run the analysis that matches it.
• Track key numbers over time: growth from one period to the next, moving averages, conversion, churn and repeat-purchase rates.
• Group customers into segments, such as loyal, at risk or lost, using how recently, how often and how much they buy.
• Find links between columns, spot unusual values, and run standard statistical tests, naming the test and the sample size each time. For example, it can show whether two measures move together, and it tells you that this alone does not prove one causes the other.
• Answer plain-English questions with one exact number and the evidence behind it.
• Compare two datasets and write up what changed between them.
• Save its findings to one file, so the charts and the report draw on exactly the same numbers.
• Read your data where it sits: in the files panel, in Google Sheets and in Google BigQuery, once you connect those apps.

It works in the files, artifact and browser panels. It reads your data in place and writes its results only to the project's own output folder. It never changes the data you gave it.

The Data Scientist has three skills. Ask in plain words and it picks the right one.

query answers one plain-English question about your data with a precise, evidenced number. It checks the column names first, so the answer uses the right fields. Use it for "What is our average order value?"

compare puts two data projects side by side. It shows how their structure, quality and metrics have shifted, and writes a diff report. Use it for "What changed between Q1 and Q2?"

pipeline-checks is a set of optional checks that run before and after an analysis. They cover a data check before the run, a log of past runs, and a webhook notice when the run finishes. Each one runs only when you approve it. Use it when you want a safety net around a large analysis.

Here is a common job, from a churn question to a tested list.

1. You ask: "Which customers are at risk of churning?" The Data Scientist receives the clean customer and order table from the Data Engineer.
2. It checks the table first. It confirms the columns it needs, such as a customer ID, an order date and an order value. If one is missing, it tells you and stops.
3. It works out the kind of data it has, here sales and customers, and picks the matching analysis.
4. It measures how recently, how often and how much each customer buys, and sorts them into groups such as Champions, Loyal, At Risk and Lost.
5. Before it names anyone "at risk", it asks you which cut-off to use: 60, 90 or 120 days without a purchase.
6. You pick one. It runs a statistical test on the groups and states the test, the sample size and the confidence level.
7. It saves the findings to a file and passes them to the Data Visualizer for a chart.
8. You get the list of at-risk customers, the numbers behind each group, and the test it used. It also says what the data could not show, for example, customers who left for a competitor and so never appear in your records. Nothing is changed in your original data.

At the end, you have one answer you can check: the number, the test behind it, the sample size, the confidence level and the findings file. The findings file means the charts and the report draw on the same figures, so nothing drifts apart. Anything hard to undo waits for your yes.

Call it when a number has to hold up to a hard question. A board meeting, a pricing decision, a claim in a sales deck. Ask it when two periods look different and you want to know whether the difference is real. Ask it when a chart shows two lines moving together and someone is about to say one causes the other. And ask it when you are unsure which customers, products or regions deserve attention first, because ranking by evidence is its daily work.

You do not need to know the names of the tests. Describe the question in your own words and the data you have. It picks the method, explains the choice in one plain sentence, and tells you how confident the result is. When the sample is too small or the data too patchy to answer, it says that first, before any number, so you never build a plan on a figure it could not stand behind.