Is data thinking the same as data science?

No. It connects problem framing and human needs with analytical work; data science provides some of the methods that may be used.

Must the outcome include AI?

No. The appropriate solution may be a process change, better measurement or a straightforward analysis.

Is there one official Data Thinking sequence?

The name is used for different frameworks. Identify the particular source before assuming its stages or terminology.

AI Glossary / Data engineering and architecture / Beginner

Data thinking

Definition

Data thinking is an approach to developing data-informed solutions that connects human needs, problem framing and data analysis. It brings ideas from design thinking into work with data: understand the situation, examine what evidence can support a useful decision, test possible solutions and learn from the results. Different authors use different frameworks under this name.

What it means

The term does not identify one universally standardized procedure. For example, researchers at CRIE/UFRJ describe a Data Thinking journey combining design thinking's Double Diamond with a data-science perspective. Other practitioners use the label for their own methods linking user benefit, business value and feasibility.

Their shared emphasis is that a technically successful analysis is not enough by itself. The work needs a meaningful question, appropriate evidence and a usable connection to action. This entry presents that broad approach rather than prescribing one author's sequence or suggesting that all named frameworks are interchangeable.

How it works

A practical exercise begins by understanding who faces a problem and what decision could improve their situation. Interviews, observation and existing records help distinguish the stated request from the underlying need.

The team then identifies which claims require evidence. It examines whether available data represent the relevant people, process and period, and whether missing information changes what can reasonably be concluded. Possible solutions may include a clearer report, a redesigned process, a simple rule or an AI model.

A small prototype makes assumptions visible. Testing considers both analytical behavior and actual use: can people understand the output, act on it and achieve the intended result? Findings can send the team back to the question or the data. This is an illustrative workflow, not a mandatory set of stages shared by every Data Thinking framework.

Three practical examples

1. Customer support: a team investigates why agents spend time reopening cases before proposing an AI classifier. It may discover that inconsistent case categories prevent meaningful analysis of repeat work.

2. Inventory planning: planners define which stocking decision needs support, then examine whether historical sales reflect demand or repeated stockouts. The distinction changes the information a forecasting project requires.

3. Everyday budgeting: a person categorizes expenses to understand a specific saving goal. They test whether the categories explain controllable spending rather than creating a complex dashboard with no practical next step.

Think of it this way

Think of investigating why a room feels uncomfortable before installing a new heating system. You talk to occupants, measure conditions and test a small change. The problem might involve timing, ventilation or uneven temperature rather than insufficient heating power.

The analogy explains problem framing and evidence. Data projects add issues such as sampling, measurement definitions and model evaluation, so observation and a prototype do not replace rigorous analysis when the decision requires it.

Worked example

Suppose a support team wants to reduce avoidable repeat contacts. It examines 200 randomly selected closed cases using a defined review procedure. Reviewers classify 50 as involving repeat contact, but only 30 of those appear linked to unclear instructions; the other 20 involve different causes.

Repeat contact occurs in 50 ÷ 200 = 25% of this sample. The instruction-related group represents 30 ÷ 200 = 15% of all sampled cases. That does not establish that rewriting instructions will eliminate 15% of cases or that the sample exactly represents future work.

A data-thinking response is to test clearer instructions for an appropriate subset and measure repeat contacts and unintended effects. The calculation narrows a hypothesis. It does not prove causation or justify deploying a large AI system before the simpler intervention is evaluated.

Illustrative case study

Fictional case: QuestionHarbour asks for a predictive dashboard to help coordinators manage training attendance. Interviews reveal that coordinators mainly need to know which sessions lack confirmed transport arrangements. The available attendance history does not record that factor.

The team prototypes a shared confirmation checklist and agrees a consistent way to record transport status. It then evaluates whether coordinators can identify unresolved arrangements earlier. A prediction model remains a possible later option if it addresses a demonstrated need with suitable evidence.

The lesson is that useful data work can change the proposed solution. This fictional case reports no measured attendance gain and does not imply that simple checklists always outperform predictive models.

Common misconceptions and mistakes

Do not confuse data thinking with accepting every number uncritically. Measurements can be incomplete, biased or unrelated to the decision.

Do not treat a convincing workshop prototype as proof of effectiveness. A prototype helps test assumptions; operational outcomes need appropriate evaluation.

Do not assume the name guarantees one fixed methodology. When working with a particular practitioner or publication, ask which framework, definitions and evaluation approach are being used.

Business use

Data thinking can help business and technical teams agree what success means before investing in collection or modeling. It is especially useful when a request starts with a technology rather than a clearly described need.

A practical project brief records the user, decision, expected benefit, required evidence and major uncertainties. It also states what finding would cause the team to change direction. For AI initiatives, separate model accuracy from workflow usefulness: a prediction can be statistically strong yet arrive too late, lack an actionable response or add work that outweighs its benefit.

Check your understanding

What does the 15% result establish in the worked example?

Answer: It describes the share of sampled cases classified as instruction-related repeat contacts. It does not establish the effect of a proposed intervention or prove why those contacts occurred.

Related terms

Hypothesis
Evaluation
Sampling bias
Design thinking
problem framing
data literacy
prototype
user research
measurement
data science
decision support

Learn next

Learn about measurement design and hypothesis testing next. Turn a request for a dashboard or AI feature into a specific decision question. List the evidence needed, what current records omit and one small test that could disprove the proposed solution's usefulness.

Hypothesis testing

Sources

Dias and colleagues, Data Thinking methodology, primary author abstract: https://inteligenciaempresarial.emnuvens.com.br/rie/article/view/100

Marc Weimer-Hablitzel, practitioner description of a distinct Data Thinking method: https://studioflyingoctopus.com/en/data-thinking/

The author abstract and practitioner account were consulted; the journal PDF retrieval failed. This entry distinguishes shared ideas from framework-specific claims. The workflow, arithmetic and fictional case are original teaching illustrations.

Previous: Data theft Next: Data Transformation