Exercise 1

Types of Data Products

Audi AI in production

Data as Insights: Supports auto production, only internal
Value proposition to customer: High tech manufacturing
Impact business model: Error prevention in production, lowers costs

OpenAI API

Data as a Product: Without data and algorithms no product
Value proposition to customer: using general llm's as a everyday helper
Impact business model: Data is the business model, collecting user data for better data product

John Deere: Precision AG Technology

Data as Enhancement: Directly advertised to consumers, as a enhancement for existing hardware and infra
Value proposition to customer: reduce costs, increase yield, more efficiency
Impact business model: better utilization of your product, higher value for consumer, usage data collection

CRISP-DM Case Study

Please read the CRISP-DM case study “E-Retail Example - A Web-Mining Scenario Using CRISP-DM” carefully. It is uploaded together with this file on teams. The following questions relate to the case study.

[[zhaw-dps-CRISP-DM-case-study.pdf]]

1. Business Understanding, Modeling and Evaluation:
Which are valid business objectives for the data product described in the case study?

• Improve cross-sales by making better recommendations.
• Increase customer loyalty with a more personalized service.

Which modeling techniques were used? Explain how they relate to the business objectives of the project.

• Kohonen network clustering techniques: Creating better recommendations by clustering together products that are purchased often, personalize the recommendation by adding customer data
• C5.0 ruleset: determine which recommendations are most fitting at any point during session
• Sequencing algorithm: identifying pages often used but too deeply nested, makes shopping a easier experience

Were the business objectives achieved? Explain for each business objective and algorithmic approach.
There was no direct mentioning in the study if the project goals listed below have been met.

• Cross-sales increase by 10%.
• Customers spend more time and see more pages on the site per visit.
• The study finishes on time and under budget
  But summarizing the studies result per business objective shows a clear success.
• Improve cross-sales by making better recommendations: Using Kohonen network clustering techniques and C5.0 ruleset recommendations are clustered and context aware
• Increase customer loyalty with a more personalized service: Using Kohonen network clustering techniques recommendations also contain personalizations according to customer data
• Thanks to the Sequencing algorithm popular actions are now better accessible

2. Data Understanding: Where does the implemented data mining process deviate from the suggested CRISP-DM standard?

• Not exploring web logs before processing due the big size and complexness

3. Deployment: Explain the activities that are included in Maintenance.

• Monitor to determine whether improved recommendations worked, preferable automatic measures
• Setup process to automatically include new registered users
• Deciding when to update rulesets (have to be done manually)