Machine Learning (ML)
Machine Learning (ML) explained
In marketing, ML can help organise customer feedback, estimate demand or suggest relevant products. Its value depends on the specific task, the data and how it fits into the working process. A model identifies statistical relationships; its prediction does not automatically explain why a customer acts or which intervention would change that behaviour.
Supervised learning uses examples with known target values or labels. Unsupervised learning looks for structures in data, such as groups of similar cases. Reinforcement learning adjusts strategies for taking actions based on feedback about their results. Generative AI also uses ML and produces content such as text, images or audio. The task determines which method is appropriate.
A useful starting point is a clearly defined decision with a measurable objective. First, establish how well the current approach works. Then prepare suitable data, train models and evaluate them on separate validation and test data. Information that would not be available at the time of a real decision must not inadvertently enter the prediction. Otherwise, measured performance can look better than performance in practice.
Better can mean more relevant recommendations, fewer incorrectly routed enquiries or more reliably identified themes in customer feedback. These quality goals should be assessed alongside the effort involved, including data preparation, integration, operation and review. Comparison with the existing approach shows whether the additional value justifies the effort. It does not establish a universal conversion uplift or cost saving.
At Davies Meyer, AI is part of connecting strategy, creativity and technology. We take responsibility for concept and quality. When evaluating an application, this means defining the objective and quality criteria, reviewing results with subject expertise and making limitations clear. A model score alone does not authorise a customer communication.
Data and behaviour can change during operation. Regular monitoring, checks for systematic disadvantage and a clear process for uncertain results are therefore necessary. When personal data is involved, permitted purposes and legal requirements must be established for the specific use. The following scenarios describe possible applications.
Examples
Illustrative scenario: product recommendations
A fictional online shop could trial recommendations using approved product and interaction data. Comparison with its existing recommendations would assess relevance, additional purchases and operational effort.
Illustrative scenario: lead prioritisation
A B2B team could test a model for prioritising enquiries. Evaluation would cover actual sales outcomes and suitable enquiries that were missed. A high score would support a sales decision.
Illustrative scenario: customer feedback
A brand could have comments sorted by topic and sentiment. Specialists would review samples, irony and ambiguous statements before the analysis informs planning.
Illustrative scenario: service enquiries
A service provider could use a model to suggest the appropriate team for incoming enquiries. Low-confidence results would go to manual review. Measures would include correct routing and rework.
Key Points
- ML learns patterns from data; a prediction does not establish a cause.
- Start with a clear task and a comparison with the existing approach.
- Evaluate model performance on separate, suitable test data.
- Assess relevance and reliability alongside cost effectiveness.
- Monitor data quality, bias and changes during operation.
- We take responsibility for concept and quality.
Sources and context
- Google: What is Machine Learning?
Introduction to learning methods, predictions and generative AI.
- Google: Dividing the original dataset
Guidance on separate training, validation and test data and the validity of evaluation.
Frequently Asked Questions about Machine Learning (ML)
Machine learning (ML) is a branch of artificial intelligence. Models learn patterns from data to classify new content, predict values or make recommendations without every decision rule being programmed individually.
Possible tasks include product recommendations, sorting customer feedback and forecasting. A useful starting point requires a clear objective, suitable data and a reliable comparison with the existing approach.
A quality improvement can mean more relevant recommendations, fewer incorrect classifications or more reliable analysis. It must be evaluated against predefined criteria. More automation alone does not establish better quality.
Models are evaluated on separate, suitable test data. Results in use are compared with the existing approach. Costs include data preparation, integration, operation and professional review.
We take responsibility for concept and quality. Objectives, quality criteria, approvals and the handling of uncertain results must be clear. Model performance, possible bias and changes in data are reviewed.
Related links
Loading related terms…
All TermsArticles about Machine Learning (ML)
AI with Brand DNA: Why Generic Bots Are a Brand Risk
Off-the-shelf AI assistants can dilute your brand. Learn how strategic calibration, guardrails, and red-teaming can transform a generic bot into a powerful, on-brand ambassador that positively impacts business outcomes.
Grok Bot Skills: How to Truly Scale AI in Your Marketing
Reusable task instructions help teams organise AI work consistently. Learn how to connect briefs, data, quality checks and version control, and measure the benefits in your own workflow.
Prompt Ops: The Operating System for AI in Marketing
The uncontrolled use of AI prompts leads to chaos. Prompt Ops provides a structured approach to manage prompts like software, ensuring efficiency, quality, and scalability in marketing.