Skip to content

SourceX Research · public sources

How much time data teams spend on data preparation: a review of public surveys

Quick answer

This page summarizes public surveys, not a SourceX study. Anaconda's 2020 State of Data Science survey of 2,360 respondents found that data loading and cleaning together consumed close to half of data professionals' time, before any modeling work begins, which illustrates why raw, unprepared data is hard to use even internally, let alone license externally.

By Noah Loul · Updated

This report summarizes published public sources, listed at the end. It does not contain SourceX survey or deal data.

Checklist

Much time data teams spend

This page covers how much time data teams spend on data preparation: a review of public surveys.Illustrative

The source survey

Anaconda's 2020 State of Data Science survey collected responses from 2,360 people across more than 100 countries, including data scientists, developers, analysts, data engineers and business managers [1].

What the survey found

According to the survey results, data loading took up about 19% of respondents' time and data cleaning took up about 26%, meaning the two preparation steps combined accounted for close to half of a typical data professional's working time, before model selection, training or deployment [1]. Model selection, training and scoring, and deployment together accounted for roughly another third of time, at about 11% each [1].

Why this matters beyond data science teams

If trained data professionals spend close to half their time just loading and cleaning data before they can use it, it is a reasonable inference that unprepared business records (customer records, logs, operational files accumulated over years without a licensing use in mind) require significant work before they could be handed to an outside party, let alone an AI developer with its own format requirements.

What later Anaconda surveys cover

Anaconda has continued its annual survey in subsequent years, each with its own methodology and respondent pool, covering topics like open-source security practices and AI tool adoption; this page relies specifically on the 2020 edition for the time-allocation figures above because that is the edition where Anaconda published the breakdown [1].

What this means for a business

A U.S. business thinking about licensing its own records should expect that a meaningful share of any project's effort will go toward the unglamorous work of loading, cleaning, standardizing and removing personal details from data, not toward negotiating a deal. Budgeting time for that step before approaching any Acquisition Partner will avoid false starts.

Frequently asked questions

Is this SourceX's own survey?

No. This page summarizes Anaconda's publicly published State of Data Science survey results; SourceX has no survey data of its own.

Does this figure apply to every industry?

The survey respondents were data science and machine learning practitioners across industries and countries; results for a specific company's own data preparation effort will vary.

Is more recent data available?

Anaconda has published newer editions of its State of Data Science survey, but the specific time-allocation breakdown cited here comes from the 2020 edition.

Sources

Public sources, checked 2026-10-09. SourceX is not affiliated with the organizations named.

  1. Anaconda's 2020 State of Data Science survey of 2,360 respondents found data loading took 19% and cleaning took 26% of time, with model work taking about 11% each for selection, training/scoring and deployment techcentral.ie

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies