Data Analytics Lifecycle
The Data Analytics Lifecycle is the structured, six-phase framework (popularised by EMC/Dell in Data Science and Big Data Analytics) that governs how a data analytics project is planned and executed. Unlike the generic analytics process, each phase here has explicit entry criteria, activities, deliverables and exit criteria, and the team can loop back to any earlier phase.
Phase 1 — Discovery
Goal: understand the business domain and frame the problem before touching data.
- Learn the business domain and its vocabulary
- Assess available resources — people, technology, time, data
- Frame the problem as an analytics problem
- Identify key stakeholders and sponsors
- Interview the analytics sponsor
- Develop initial hypotheses (IH) to be tested with data
- Identify potential data sources
Deliverable: problem statement, initial hypotheses, resource plan.
Phase 2 — Data Preparation
Goal: get the data into a usable state. Usually the longest phase.
- Prepare an analytic sandbox — an isolated workspace where the team can freely explore and transform data without touching production systems
- Perform ETLT — Extract, Transform, Load, and Transform again (load raw into the sandbox, then transform inside it)
- Familiarise with the data (data dictionary, ranges, distributions)
- Condition the data — clean, handle missing values, normalise, deduplicate
- Survey and visualize the data
Deliverable: clean, documented, analysis-ready dataset in the sandbox.
ETL vs ELT vs ETLT — ETL transforms before loading (classic warehouses); ELT loads raw and transforms inside the target (modern cloud/big-data); ETLT does light transforms before load and heavy ones after, which is what an analytic sandbox usually needs.
Phase 3 — Model Planning
Goal: decide what to model and how.
- Explore relationships between variables (correlation, EDA)
- Select key variables / features
- Choose candidate techniques — regression, classification, clustering, association rules, time series
- Choose tools — Python/R, SAS, SQL, Spark
- Split strategy for training and testing data
Deliverable: shortlist of candidate models with justification.
Phase 4 — Model Building
Goal: actually build, train, and test the models.
- Create training, validation and test datasets
- Fit the candidate models
- Tune parameters and compare performance
- Verify whether the environment (hardware, memory, runtime) is sufficient for production workloads
Deliverable: trained models with measured performance on unseen data.
Phase 5 — Communicate Results
Goal: compare outcomes against the initial hypotheses and tell the story.
- Determine whether the results are statistically significant and practically useful
- Identify the key findings, quantify the business value
- Build a narrative and visualizations for each audience (sponsor, technical team, end users)
- State assumptions and limitations honestly
Deliverable: final report, presentation, and executive summary.
Phase 6 — Operationalize
Goal: move from a project to a running capability.
- Run a pilot on a limited scope before full rollout
- Deliver final code, technical documents, dashboards, and runbooks
- Deploy the model into production systems
- Set up monitoring for performance decay and data drift
- Plan periodic retraining
Deliverable: deployed model/dashboard plus monitoring and maintenance plan.
Phase Summary Table
| Phase | Core question | Key output | Typical time |
|---|---|---|---|
| 1. Discovery | Do we understand the problem? | Hypotheses + scope | 10% |
| 2. Data Preparation | Is the data usable? | Analytic sandbox | 40–50% |
| 3. Model Planning | Which technique fits? | Candidate models | 10% |
| 4. Model Building | Does it work on unseen data? | Trained, tested model | 20% |
| 5. Communicate Results | What does it mean for the business? | Report/presentation | 10% |
| 6. Operationalize | Can it run reliably in production? | Deployed system | 10% |
Comparison with Other Frameworks
| Framework | Phases | Origin |
|---|---|---|
| EMC/Dell Data Analytics Lifecycle | Discovery → Data Preparation → Model Planning → Model Building → Communicate Results → Operationalize | EMC Education Services |
| CRISP-DM | Business Understanding → Data Understanding → Data Preparation → Modeling → Evaluation → Deployment | Cross-Industry Standard Process for Data Mining (1999); still the most widely used in industry |
| SEMMA | Sample → Explore → Modify → Model → Assess | SAS Institute |
| KDD | Selection → Preprocessing → Transformation → Data Mining → Interpretation/Evaluation | Knowledge Discovery in Databases |
All four frameworks encode the same underlying truth: understand the problem, prepare the data, model it, evaluate honestly, and deploy with monitoring — with iteration built in at every step.