The Clearly Podcast

Andy: Welcome to the final episode of Season 2. Today, we’re discussing Power BI and machine learning. This is the third episode in our series. In the first two episodes, we covered "What is ML?" and "Are you ready for ML?". Now, we’ll focus on how Power BI and ML work together. Tom, can you explain their relationship?

Tom: Sure. Power BI isn’t a machine learning platform, but it helps visualize machine learning results. The easiest way is using Power BI’s data flows with built-in ML components, which might require a premium subscription. You can also connect Power BI to Azure Machine Learning with M code or use Python and R in Power Query, but that gets more complex.

Benefits and Use Cases

Andy: What benefits do pre-built components offer versus custom coding?

Tom: Pre-built components save time and leverage Microsoft's extensive resources, offering reliable results faster. Custom coding provides more control but requires expertise and more time. Pre-built tools are excellent for tasks like text processing, sentiment analysis, and key phrase extraction.

Shailan: We haven’t seen many customers using these advanced features yet, but key features like NLP and sentiment analysis are becoming more accessible. Language detection and image tagging also show potential but require clear business needs to justify their use.

Andy: Are customers finding practical uses for these tools?

Shailan: Not widely yet, but potential use cases include text analysis and sentiment analysis. Key phrase extraction and language detection can be particularly useful for global reports. Image tagging is promising but needs clear business needs.

Tom: It’s important to consider the impact on Power BI’s performance, especially with image tagging. Some organizations might use other tools for visualization while running machine learning models in the background.

Starting an ML Project

Andy: What should customers do to start an ML project?

Tom: Identify a clear business need first. Projects often fail if they lack a defined goal. Machine learning works best for specific problems like analyzing sales patterns. Without a clear purpose, projects can become costly and unproductive.

Andy: There’s a risk of projects failing if they’re driven by trends rather than needs.

Tom: Yes, like with data cubes years ago. Projects driven by buzzwords rather than needs are likely to fail.

Long-Term Considerations

Andy: What do long-term ML projects look like?

Tom: They involve ongoing maintenance and frequent adjustments. Initial phases involve testing different algorithms to find the best fit. These projects need continuous monitoring and updating to ensure data quality and model accuracy.

Shailan: Maintenance is crucial. Changes in source data can impact results, so ongoing updates are necessary. Budgeting must account for fluid costs and time, considering the potential for unexpected changes and maintenance needs.

Top Tips

Andy: Any final tips?

Shailan: Start with small, specific requirements. Use Power BI’s new features like the key influencers visual. Test with a subset of data to explore potential.

Tom: Ensure you have a clear business case. Use AI-driven visuals and cognitive services in Power BI for initial exploration. Treat ML projects as R&D, not just implementations.

Andy: Remember, these projects need flexibility in budgeting and time. Think of them as R&D efforts rather than standard implementations.

Tom: Agreed. The bulk of the work in machine learning is in the R&D phase. The implementation follows once you’ve validated your model.

Andy: Great insights. Thanks, everyone. That wraps up Season 2. We’ll start Season 3 soon with more topics on technology.