The Clearly Podcast
Time for a Rebuild?
Summary
This week's discussion focuses on recognizing when it's time for a rebuild in Power BI or data analytics models. Changes in business needs, data inaccuracies, performance degradation, and the introduction of new systems are common reasons to consider a rebuild. Rebuilding should not be seen negatively; it's an opportunity to improve and adapt to new requirements.
Key indicators for a rebuild include declining performance and the need to awkwardly integrate new data elements. Rebuilds help maintain data integrity, performance, and alignment with business objectives. It's essential to communicate the necessity and benefits of a rebuild to clients and business owners.
Rebuilds should be approached with caution if they seem too sudden or unnecessary. A model should generally be robust for at least a year or two. A phased approach to rebuilding, prioritizing critical reports and retiring outdated ones, can make the process more manageable. Clear communication and training on changes are vital for a successful rebuild.
Budgeting for potential rebuilds is important as business needs evolve. Addressing issues promptly and strategically planning rebuilds can prevent problems from escalating. Rebuilds often progress faster than anticipated, emphasizing the importance of early intervention and careful planning.
Next week, the discussion will likely cover Fabric, a current hot topic, with a focus on providing a well-considered perspective.
You can download Power BI Desktop from here.
If you already use Power BI, or are considering it, we strongly recommend you join your local Power BI user group here.
Transcript
Andy: This week, we're discussing when it's time for a rebuild in your Power BI or data analytics models. Things change, and occasionally, you need to rethink and rebuild. We'll explore the reasons why and the circumstances that necessitate a rebuild. It's an important part of maintaining effective models, especially if performance degrades, numbers are incorrect, or the model has been patched too much.
Tom: The biggest warning sign is when a model that used to perform well no longer does. Rebuilding shouldn't be considered a dirty word. It's not about throwing away the original setup but rather taking what you've learned and improving. Performance degradation and the need to shoehorn new elements into your data models are clear indicators that it's time for a rebuild. This applies to data warehouses, Power BI models, and more.
Andy: It's crucial to set expectations with clients and business owners that rebuilds are normal over time. A rebuild ensures data integrity, performance, and that business needs are met. Tom, how do you justify a rebuild to a business user in their terms?
Tom: Businesses and their goals change over time. A model built a few years ago might no longer be suitable. For instance, if you originally aimed to grow revenue and have achieved that, your new goal might be to sustain that growth, requiring different data handling. New systems and data streams can also necessitate a rebuild to incorporate new technology effectively.
Andy: Adding new data streams meaningfully might require a rebuild. While you can't build infinitely extensible models, there's a balance between usability and extensibility. When do you risk what you've already got?
Tom: If you have a model with many reports, changes might risk those reports. Extensive testing is crucial to ensure accuracy. Sometimes a phased approach is better, transitioning some reports to a new model gradually.
Andy: What are the hallmarks of a rebuild project for you?
Tom: Consider the scale of reports off any model. Extensive regression testing might be needed, making a phased approach more practical. Retire old, irrelevant reports and manage the transition carefully. Communication is key to ensure everyone understands the changes and why they are necessary.
Andy: When should you urge caution with a rebuild?
Tom: If a rebuild seems too sudden, it might indicate something significant was missed or there's an issue with implementation. Data models should be robust for a year or two. Rebuilding within the first year suggests a reactive approach rather than a strategic one. Also, don't rebuild just because a model is a few years old if it's still functioning well.
Andy: Any tips for a successful rebuild?
Tom: Communicate the business justification clearly. If you suspect an issue, address it promptly to avoid exacerbating the problem. Phased rebuilds can be less daunting and more manageable.
Andy: Yes, ignoring issues can make things worse. Rebuilds often go faster than anticipated. Budget for potential rebuilds as part of your data provision plan.
Tom: Definitely. Business needs evolve, so reporting should too. Budgeting for rebuilds ensures you're prepared.
Andy: Thanks, Tom. Next week, we'll likely discuss Fabric, something everyone's talking about.
Tom: Yes, we're taking our time to give a considered view. It'll be better than everyone else's.
Andy: Great. Thanks, Tom. Until next week, everyone. Thanks for listening.
Tom: Cheers. Thank you.