Skip to content
CompanyBenchCompanyBench

Is your data pipeline ready for a data scientist?

Ten checks to run on your data pipeline before you hire a data scientist, so they spend their first month building models instead of fixing data.

You have decided to hire a data scientist to build forecasting, churn prediction or personalisation models. If your data pipeline isn't ready, they will spend the first month wrangling data instead of modelling, and you will be paying a data scientist's rate for infrastructure work that should have been done earlier. Run through this checklist before they start.

The readiness checklist

1. Your data sources are centralised or mapped

Will your data scientist know where to find customer data, product events and financial data? If it is spread across 10 databases, Salesforce and a set of spreadsheets, the first month goes on exploration, not analysis. At the very least, write down where each kind of data lives.

2. You have a data warehouse or data lake

A data scientist needs queryable data in one place, not scattered across production databases. If you don't yet have Snowflake, BigQuery or Redshift, budget for it and build it before you hire. A data scientist can help set one up, but starting without one wastes their time.

3. Your data is reasonably clean

It doesn't have to be perfect, but you should have run basic validation. Are user IDs consistent? Are timestamps in a known format? Are there columns that are mostly null? Document the major quality issues and either fix them or flag them up front.

4. You log events and behavioural data

Predictive models need the underlying events: sign-ups, page views, purchases, cancellations. If your product or backend doesn't log them, no data scientist can make up for it. Make sure your product team has the instrumentation in place.

5. Your retention policy is clear

How long do you keep data: 30 days, 1 year, forever? The data scientist needs to know the window they are working with. If you delete data after 60 days but need 2 years of history to train a model, fix that before they start.

6. You have defined the target variable

For a churn model, what counts as churn: 30 days of inactivity, or a cancelled subscription? For a forecast, are you predicting revenue or users? Be specific. A vague target produces a vague model.

7. Someone owns the database

A data scientist builds models, not databases. If the warehouse needs ongoing maintenance, backups, access management or tuning, you need a database admin or data engineer handling it, even part-time. Otherwise the data scientist ends up doing it.

8. Access and permissions are sorted

Can the data scientist query the warehouse? Can they read raw logs? Who approves access to sensitive customer data? Settle this in advance so the first week isn't spent waiting for permissions.

9. You have defined success metrics

What does a good model look like: 90% accuracy, or 70%? Are you optimising for precision or recall? Does latency matter? Agree on this before modelling starts, not after a month spent building something you don't want.

10. You have budget for compute

Building models needs compute: cloud credits for training, GPUs and storage. For small projects, budget ₹18,800–94,000 ($200–1,000)/month. Without resources, a data scientist can't experiment.

If you are missing some of these

Missing 1-2 items is recoverable: fix the most important gaps first. Missing more than 3? Hire a data engineer first to build the pipeline, then hire the data scientist. That is faster than hiring both at once and leaving the scientist with nothing to model.

Hire in the right order

The usual order is a data engineer first, to build the warehouse, clean the data and set up instrumentation, then a data scientist to build the models. Reverse it and your data scientist does engineering work, which is expensive and frustrating for everyone.

Finding the people

CompanyBench supplies vetted contract data scientists and data engineers, with a shortlist of at least 3 candidates within 48 hours of confirming your requirement. See data scientist profiles and data engineers, including warehouse specialists. For a step-by-step plan to build your first data team, read the data engineering team checklist. For the different roles, see data analyst vs data engineer vs data scientist.

Frequently asked questions

Can a data scientist build the pipeline themselves?

Technically yes, but it is wasteful. A data scientist's time is expensive, and it costs less to have an engineer build the infrastructure while the scientist focuses on modelling.

How clean does the data need to be?

Clean enough that basic queries work and don't return strange results. You don't need perfection, but obvious issues should be fixed or documented.

What if we don't have a year of historical data?

That is fine for some projects, such as churn or forecasting, and hard for others, such as rare-event prediction. Understand the constraint before you start.

Can a small company use a data scientist as a data engineer?

For the first month, maybe. As your data needs grow, they will need dedicated engineering support, or the data scientist becomes a bottleneck.

How much compute budget is realistic?

Start small: ₹18,800–28,200 ($200–300)/month for experimentation. Expect ₹47,000–1,41,000 ($500–1,500)/month or more once a model is in production and serving inference.