Skills application
Instruction and application
CompleteIdentifying ML solutions to solve business problems
In this skills application, you will practice selecting appropriate ML models for different business scenarios.
Learning this skill brings value to the workplace by enabling you to understand the necessary considerations when determining the right ML approach. You'll also be able to frame business problems correctly to match business objectives to the appropriate ML method and model.
Context
In this skills application, you'll need to select the most appropriate ML solution for three different business scenarios.
Instructions and materials
Work through the steps below to complete the skills application; then, see how you did by comparing your work with our solution.
Initial analysis
Review the three business problem scenarios thoroughly, taking notes about:
- Key business objectives
- Critical constraints
- Data characteristics
- Performance requirements
- Technical limitations
Scenario A: Customer churn prediction for MultiConnect Telecom
Business context:
MultiConnect Telecom is a telecommunications company offering mobile, internet and cable TV services. It has been experiencing an increasing customer churn rate over the past year, currently at 15% annually, which is above the industry average. The company has a customer base of approximately 100,000 subscribers.
Business constraints:
- Need to predict churn risk 60 days in advance to allow for intervention.
- Must maintain customer privacy compliance (GDPR/CCPA).
- Solution needs to work with the existing CRM system.
- Limited budget for customer retention actions.
- Need interpretable results to design retention strategies.
Available data:
| Data type | Data |
|---|---|
| Customer demographics | - Age, gender, location - Customer segment (residential/business) - Length of relationship - Contract type (monthly, annual, two-year) |
| Service usage | - Monthly bill amount - Payment history (last 12 months) - Services subscribed - Data usage patterns - Call patterns (mobile) |
| Customer interaction | - Number of support tickets - Average resolution time - Customer satisfaction scores - Last upgrade/downgrade date - Website/app usage frequency |
| Historical churn data | - Previous churn events - Successful retention actions - Reasons for leaving (when available) |
Scenario B: Supply chain optimisation for Freshly Mart Group
Business context:
Freshly Mart Group is a regional grocery chain with 70 stores. It is facing challenges with inventory management, particularly with perishable goods, leading to approximately 8% waste and frequent stockouts. The company aims to optimise its supply chain to reduce waste, improve stock availability and better manage cold chain logistics.
Business constraints:
- Perishable goods require quick turnover (shelf life 2-14 days).
- Limited storage capacity in each store location.
- Different demand patterns across urban/suburban locations.
- Weather-dependent demand fluctuations.
- Lead time constraints from suppliers (24-72 hours).
- Temperature-controlled transportation requirements.
Available data:
| Data type | Data |
|---|---|
| Sales data | - Historical sales by product, store and date - Promotion impact on sales - Seasonal patterns - Price points and margins - Waste/spoilage records |
| Inventory data | - Current stock levels - Storage capacity by location - Product shelf life - Reorder points - Safety stock levels - Storage temperature requirements |
| External factors | - Local weather data - Local events calendar - Competitor locations - Regional demographics - Holiday schedules |
| Supply chain operations | - Supplier lead times - Transportation costs - Warehouse capacity - Labour availability - Cold chain monitoring data |
Scenario C: Chatbot implementation for PurpleZone Health
Business context:
PurpleZone Health is a health insurance provider serving two million customers. Its customer service department handles approximately 60,000 inquiries per month. The company aims to implement an AI chatbot to handle routine inquiries, reducing wait times and allowing human agents to focus on complex cases.
Business constraints:
- Must maintain healthcare data privacy (HIPAA for US, UK GDPR/DSPT for UK).
- Language requirements (English/Spanish for US, English for UK).
- Must seamlessly integrate with the existing customer portal.
- Required 24/7 availability.
- Need clear escalation paths to human agents.
- Must maintain 95% accuracy in responses.
- Response time under five seconds.
Available data:
| Data type | Data |
|---|---|
| Historical customer inquiries | - Chat transcripts (last two years) - Email communications - Call centre logs - Common questions and answers - Resolution categories |
| Customer information | - Policy types - Claim history - Customer segment - Language preference - Digital platform usage |
| Knowledge base | - FAQ documentation - Policy documents - Procedure manuals - Benefits guides - Provider network information |
| Interaction metrics | - Average handling time - Customer satisfaction scores - Resolution rates - Escalation patterns - Peak inquiry times |
Model selection
Review the Model Selection Guide and diagrams in the materials section below. Then:
- Identify the top three candidate ML models.
- Evaluate, select and recommend the most appropriate ML model(s) using the Selection Matrix Template.
Justification
Document your selection reasoning, including:
- Why is this model the best fit for the scenario?
- What key advantages does the chosen model have over alternatives?
Final review
- Cross-check selections across scenarios.
- Ensure justifications are complete.
- Prepare a quick bulleted list of recommendations.
Best practices
- Always follow a structured process, such as the Model Selection Guide, to ensure you're considering all available options.
- Keep notes that describe how you came to your final solution, as this helps when getting stakeholder approval.