Understanding the problem and purpose is important which helps in identifying the nature of the problem and the type of framework we can choose. Because each type of analytical work needs different approaches. for example when you are dealing with a health-related regression problem. don't jump into descriptive analytics without properly accessing the risk factors in that data.

I have an experience on this when working at Sesto Synergy where we customized analytics frameworks to add some sparkle to the jewelry manufacturing processes. We tailored predictive maintenance models to ensure our machines ran as smoothly as a diamond cut, preventing any 'brace-let' downs. By integrating real-time data and constantly polishing the framework with feedback from our gem of a team, we made sure our production was always dazzlingly efficient.

In addition to defining the problem, it is helpful to formulate a set of hypotheses. The goal of analyzing data would be to prove or disprove a hypothesis.

If you are into pure analytics (crunching data), the key would be to involve the main stakeholders in the defining phase. By this, you are avoiding a whole load of efforts being translated to "My expectations were different" or "thats not industry metric" or "that does not meet my business objectives". Involve stakeholders periodically.

While understanding the problem is the key to data analysis, it is also important to understand business context too. All this stems from asking the right questions. The more you can understand the problems and how and why certain things are done in business, the more you’re equipped to getting the right answers from your analysis.

DAR (Dashboard, Analysis, Reporting) is another framework that can be applied in building impactful reporting. This entails building the analytics presentation layer to start at the highest level of granularity (think high level KPIs), followed by singular views constructed around specific business processes or departments, followed by detailed, granular views represented in tabular formats.

Choosing the right framework for data analytics projects is crucial for better results. Key steps include: Define Objectives, Understand Context and Data, Consider Stakeholder Perspective and Feedback, Evaluate Complexity, and Prioritize Ease of Use. Incorporating these steps ensures alignment with goals, relevant insights, and effective stakeholder engagement. For instance, health sector data needs to handle the risk of predictions and classifications needs to be addressed. We can't use accuracy to determine the success in this domain. So choose your framework wisely.

I like the classic CRISP-DM, with its 6 iterative stages: 1) Business understanding – What does the business need? 2) Data understanding – What data do we have / need? Is it clean? 3) Data preparation – How do we organize the data for modeling? 4) Modeling – What modeling techniques should we apply? 5) Evaluation – Which model best meets the business objectives? 6) Deployment – How do stakeholders access the results? The insights discovered in each stage can either lead forward to the next stage, or backwards to the previous stage, for a more refined approach. The entire process is iterative, i.e. the feedback of the deployment phase should kick-off another cycle of exploration.

Selecting the right analytics framework is critical, especially in sectors where the stakes are high. The choice of framework — whether it be predictive models to forecast equipment failures, prescriptive models for risk management, or diagnostic tools for analyzing past events and learn — must align with the identified challenges' complexity. This choice should be influenced by the nature of the data available, the desired outcomes, and the organization's technological capabilities.

An appropriate framework helps in creating a structured approach towards analysing data which is designed to process and interpret vast quantities of information. What I have found useful is to define and document each step of the framework and organize the analysis around it with constant checks and controls whether it is during requirement gathering, data collection and acquisition, data processing and cleaning, data storage, analysis and mining, visualization, insight generation and decision making, and even the communication channels like dashboards, email reports, automations, alerts, and presentations. The formats during these stages can follow SMART, OKR, SWOT frameworks depending of nature of the analysis.

Its always helpful to have industry context to really understand the which data brings what value and really stands out. For example, while doing compensation analysis we found anomalies in a department since the compa-ratio seemed off. The compensation information is something that is specific to HR and thus the people working in HR will be able to identify it firsthand. So context really helps when understanding data quality.

Knowing the data is a must for analytics, including metadata, and the data quality needs for a specific function/ area determines how the source and its transformation must be adjusted. For example, Marketing needs data quickly for analyzing the impact a promotion and may need only 80% accuracy, whereas for a Sales Performance the data accuracy has to be more like 95% to assess how lead conversion to orders and actual sales happens. Same data used by both but Sales need further refinement before it gets analyzed for the outcome.

It's essential to adapt to your data sources and quality. at times you receive inconsistent and null data points. Handling inconsistent and null data points is crucial for accurate analysis. so handling these things requires careful handling. Simply replacing null values with 0 or mean may not suffice. Understand each variable individually and develop a specific approach for accurate analysis. This kind of careful analysis is required only if you aim have to better results.

In the 'Data Understanding' phase of CRISP-DM, we typically perform the following tasks: 1) We identify all the relevant data sources: database tables, files, streams, etc. 2) We profile and explore the data, to understand their basic characteristics, in terms of the requirements of our overall analysis. 3) We evaluate the data in terms of their quality 4) We create some sort of documentation, to share our findings with our collaborators. 5) Finally, we assess whether our planned objectives are aligned with the data at hand, and if not we either seek new data sources, or we iterate to the 'business understanding' step, to reevaluate the scope of the analytics exercise.

The diversity and quality of data, from real-time sensor data to historical studies, should dictate the analytics approach. Adapting the analytics framework to accommodate varying data sources, ensuring data quality, and integrating disparate and siloed data types are critical steps. This may involve refining data collection methods (digitalization), enhancing data cleaning processes, or employing sophisticated data integration techniques to create a comprehensive, high-quality dataset for analysis.

Understanding the audience of your visualization is key to creating a high quality and effective product. When the ask is for something high level, a few KPIs paired with graphs or a map will generally fit the bill. As the need for detail increases, the need for tables, tooltips and creative visualization increases. When developing, do your best to put yourself in the stakeholder's position: what level of detail would you need in order to make data-driven decisions with this product?

I recommend this often to my team members 1. Dont get into analysis paralysis 2. Use the right infographic. Fancy graph doesnt mean great inforgraphic. Keep it simple and consumable

Story telling with Data is key to highlight how a business problem could be or was solved with Data. The journey from what the problem was and how the data driven decision helped with the outcome is a compelling way to convince the stakeholders. Various visualization techniques are key to answer the questions and sub-questions one may have with a specific scenario leading to business outcome.

Contrary to common belief, the visualization piece of work is probably the most important, as it immediately communicates insights to decision makers. It is partially art and partially science to produce impactful visualizations. Some books I've enjoyed reading on the subject: - Storytelling with Data by Cole Knafic. - The Visual Display of Quantitative Information by Edward Tufte. - Better Data Visualizations: A Guide for Scholars, Researchers, and Wonks by Jonathan Schwabish

Furthermore, I believe that the most important thing is “how you can” develop Storytelling. When you need to create analysis, this information will be for other people to read, understand and make action plans using the data. This is why "storytelling" in a way that the other person can read is important

The art of translating your findings and insights into compelling stories is increasingly important. The art of data storytelling will need you to understand your audience, and their needs and a way to create a memorable way to remember findings and follow action items.

Right now – you can choose the best path after understanding “What is your target audience” this is 80% of the way. After that you can choose between "tools"

Effective communication is what gets you paid! Consider this - you do all due diligence and sift through complex data, using various models and tools to arrive at a viable solution. Now, if this solution is not effectively communicated to the target audience, or if the solution is too complex to even understand by some non-technical audience, then the entire effort would go to naught. One peculiar observation I have made in many cases is that once we communicate the results of our analysis to the key stakeholder, which are most of the times not the implementors of the solution, then most often the analysis goes t waste, as the people who can actually do something with the insight are unaware, which is a loss-loss both ways.

Adopting a one-size-fits-all solution for analytical reporting is a recipe for failure. Understanding and resolving a business problem cannot be achieved with a linear mindset and approach. It's crucial for an analytics team to adapt based on the project's needs and to present findings tailored to the audience's interests and the value they seek. While having a general approach to a project is beneficial, the willingness to adapt is crucial for success.

Here is a LinkedIn article I wrote back in 2017 -- Analytical Context. https://www.linkedin.com/pulse/analytical-context-chris-westphal/ You need to put everything into a context. There are no right-answers and no wrong-answers; only an interpretation within a specific context.

Adapting an analytics framework to different contexts involves understanding the problem and purpose, choosing an appropriate framework, and considering data quality and sources. Tailor your analysis and visualizations to match audience needs, ensuring insights are relevant and clear. Communicate findings effectively with concise, actionable messages. Continually review and refine your framework based on feedback and new insights to keep it robust and aligned with evolving needs. This flexible approach ensures your analytics framework remains effective across diverse scenarios.

Data is alive - there are always new perspectives to put old analyzes into, essentially because business changes all the time... Furthermore, it is necessary to improve the KPI with the current business situation.

Analytics framework should be customized based on the breadth & depth of business questions asked, output goals to be achieved, availability of data, analytical resources & capabilities. Not all analysis needs an expensive dashboard if it can be answered in a light-weight spreadsheet view. Similarly, not all analysis needs to be automated especially if there is limited bandwidth on the team or it’s a one-time data ask. The ability to foresee what should be automated is crucial & for that close alignment with business stakeholders is important. For questions that are asked on a consistent basis automation is key so that the teams can scale & focus their time on generating valuable insights which is a hallmark of effective analytics teams.

In a project for a regulated industry (banking), I had to ensure that the analytics models were validated against strict regulatory requirements, including explainability & bias detection. Standard validation methods were insufficient, so I incorporated advanced techniques like SHAP (Shapley Additive Explanations) to explain model decisions & ensure compliance. This adaptation was crucial for regulatory approval of the framework. Adapting model validation techniques to industry-specific regulations ensures that the framework not only delivers accurate results but also meets compliance standards, which is essential in highly regulated sectors.

Your analytics teams should understand your business area and your objectives. The risk with centralised analytics teams that don’t understand their customers is that they produce outputs that don’t meet the business needs and they don’t get used, and stakeholders revert to excel! If your organisation has a centralised team, consider embedded relationship managers in the various business areas that your team serves. They should understand the objectives of your customers and help craft the requirements together, showcasing the art of the possible.

Understanding the specific context & defining clear objectives aligned with business goals is crucial. Identifying relevant data sources and integrating them into the framework while ensuring data quality is essential. Selecting or developing appropriate analytical models and customizing them to fit the context by incorporating relevant features & addressing domain-specific considerations is important. Customizing visualization and reporting components helps effectively communicate insights. Continuous evaluation, improvement, & scalability considerations are vital. Incorporating advanced techniques, fostering a data-driven culture, collaborating with stakeholders, & staying updated on industry trends contribute to successful customization.

In embracing analytics, consider the integration of IoT for real-time data, and AI/ML for deep insights. Leverage open innovation, collaborating with startups and academia to infuse creativity and speed. Cultivate a data-driven culture, prioritize data security, and navigate ethical data use. Aim not just to solve immediate challenges but to anticipate future trends for sustainable, agile operations.