Data science / ML

Use Dawn for data-science work when a question needs more than a descriptive analysis: forecasting, classification, segmentation, optimisation, or a model that supports a repeatable decision.

Start with the decision

Frame the work in business terms before choosing a method:

• What decision will the result improve?
• What outcome are you trying to predict or optimise?
• What would a useful result change in practice?
• How will you measure whether the approach is better than the current one?

This prevents a technically impressive model from solving the wrong problem.

Prepare a reliable dataset

Use a versioned, well-understood dataset. Check for missing values, leakage, biased coverage, and fields that would not be available at the moment a prediction is made. Document target definitions and important assumptions in Context.

See Data cleaning for a practical preparation workflow.

Develop in stages

Ask Dawn to help you explore the data, propose a baseline, and compare candidate approaches. Begin with the simplest credible baseline so improvements are meaningful.

We want to forecast weekly demand by product and region for the next eight weeks.
Review the dataset for suitability, recommend a baseline and evaluation approach,
then explain which features are appropriate at forecast time.

Validate before operationalising

Evaluate performance on data that was not used to develop the model. Check the metrics that matter to the decision, inspect important errors, and look for performance differences across relevant groups or periods.

Before relying on a model, agree on its owner, review cadence, inputs, outputs, and the condition that should trigger retraining or escalation.

Put the result to work

Save preparation and scoring steps in a pipeline, expose decision-ready output in a dashboard or report, and use a job when the workflow needs to run again. Continue to monitor outcomes; data science is an ongoing product, not a one-time analysis.