Must all raw records be published?

No. The arrangement can use controlled access or selected outputs appropriate to its purpose.

Is a collaboration successful once data has been transferred?

Not necessarily. Transfer, useful analysis and public impact are different milestones.

Does a company's dataset represent everyone in a community?

No. Coverage must be examined rather than assumed.

AI Glossary / Data engineering and architecture / Advanced

Data collaboratives

Definition

Data collaboratives are arrangements in which participants from different sectors work together to make data and related expertise useful for a public purpose. Participants may include companies, public bodies, researchers and civil-society organizations. The collaboration involves more than transferring files: it needs a defined problem, an agreed way of using the information and responsibilities for producing and assessing the intended benefit.

What it means

Useful information and the ability to act on it often sit in different organizations. A company may hold relevant observations, researchers may offer analytical expertise and a public organization may understand the service problem. A collaborative connects those contributions. The term does not imply that every participant contributes identical resources or that all underlying records become publicly available. What matters is how the arrangement supports a defined public objective and whether the evidence justifies the conclusions drawn from the shared information.

How it works

Partners identify a problem and assess whether available data can help answer it. They define the contributions, permitted uses, access arrangements and measures of success. The GovLab's practical guidance emphasizes matching data supply with demand, governance and evaluation of impact. Technical arrangements vary with the information and task. A project might work with selected summaries rather than a common pool of raw records. The collaboration should distinguish obtaining data, generating an analysis and producing a useful public outcome, since success at one stage does not establish success at the others.

Three practical examples

A fictional transport partnership combines operator summaries with local service knowledge to investigate gaps in route coverage.
Researchers and a utility examine aggregated demand patterns to support a clearly defined public planning question.
Community organizations help interpret a shared dataset so an analysis does not mistake the provider's customers for the whole local population.

Think of it this way

Imagine several teams assembling a map for a rescue exercise. One knows the roads, another knows available facilities and another understands local access constraints. Their contributions become useful through a shared question and a coordinated interpretation. Simply piling the maps together would not ensure a good plan. A data collaborative has a similar challenge: complementary information and expertise need an agreed purpose, and the resulting picture can still have gaps that the participants must explain.

Worked example

Use an original hypothetical transport study. A provider supplies 6,000 observed journeys from District A and 4,000 from District B. Within this dataset, A accounts for 60 percent of the 10,000 observations. That does not establish that A generates 60 percent of all journeys in the two districts. If provider coverage differs, the sample shares may differ from population shares. The collaborative therefore needs additional evidence before using the ratio to allocate services. The arithmetic is straightforward; the important collaborative work is understanding coverage and deciding what the observations can legitimately support. No population estimate can be calculated from these counts alone.

Illustrative case study

A fictional city, university and delivery company explore whether shipment summaries can help identify underserved areas. Community representatives point out that areas with few recorded deliveries may reflect limited platform use rather than low need. The partners refine the question, compare independent information and report the coverage limitations. They also define how any proposed intervention would be evaluated. The scenario illustrates the value of combining data with local knowledge and scrutiny. It does not claim that sharing a commercial dataset automatically produces a representative account of a community or a measurable improvement in public services.

Common misconceptions and mistakes

A collaborative is not automatically an open-data release, and public value is not established simply by calling a project collaborative. More data may add little if it does not address the question or represents the wrong population. Another mistake is assuming that technical access resolves differences in incentives, definitions and responsibility. A partnership announcement is also not evidence of impact. Finally, organizations contributing data and people described by that data are not necessarily the same participants; the design should consider whose knowledge and concerns are absent from the discussion.

Business use

For a company considering participation, define the contribution and expected public use before committing to an exchange. Assess the relevance and limitations of the information, the effort required to prepare it and who will interpret the results. Clarify responsibilities for ongoing maintenance and for communicating findings. For an AI project, examine whether the data supports the intended population and task, rather than treating volume as proof of suitability. A useful collaboration makes its assumptions and outcomes assessable by the partners and the people expected to benefit.

Check your understanding

What proportion of the toy provider's observed journeys came from District A?

Answer: Sixty percent.

Why could that not be used directly as the share of all journeys?

Answer: The provider's coverage of each district was unknown.

What additional contribution might a partner make besides data?

Answer: Expertise about the problem, the population or the interpretation of results.

Related terms

Data governance
Data sharing
data stewardship
public value
representativeness
impact evaluation

Learn next

Describe one public problem and list the information and expertise needed to address it. Identify which organizations hold each contribution and which affected groups should inform interpretation. Separate an access milestone from an analysis milestone and an outcome measure so the project can evaluate more than the volume of data exchanged.

Sources

The GovLab, New Tool to Design Data Collaboratives: https://blog.thegovlab.org/new-tool-to-design-data-collaboratives The transport calculation, map analogy and city-delivery scenario are original educational illustrations. The reference supplies the cross-sector public-value concept and design context; none of the fictional examples claims a documented real-world outcome.

Previous: Data cleaning Next: Data collection