How do you identify and quantify key drivers and factors for your business outcomes?

There can also be variables that have complex relationships with other variables and affect the outcome or dependent variable. Such variables are called interaction variables

"Data" is the key element in evaluating the factors influencing an organization's sales, revenue, and customer satisfaction. While data availability is not a challenge nowadays, establishing an effective data ecosystem is crucial. This involves collecting various types of data, including transactions, interactions, web logs, and external sources, and employing a Lake house architecture to gain detailed insights into each business process. By utilizing this data, Machine Learning (ML) models can be created to predict sales. The data can also be analyzed to create dashboards and Business Intelligence (BI) reports, allowing the organization to identify pain points and areas needing improvement.

70% of Key drivers can be found from the end-user's requirement, where as remaining 30% can be found from the other stakeholders. E.g. let's talk about an oncology drug, which is ready to enter the market. If you try to figure out key drivers, total number of patients, their stages, LoT, their persistency, compliance control a huge part of market share. Where as the remaining depends upon Prescribers, Payers, govt. Policies etc.

Correlation analysis in quantifying the key drivers in a business, the ability to measure the relationship between the variable enables a business to make informed decisions. Trends analysis and time series analysis gives impactful insights on projections and ability to predict future outcomes and expectations.

I find it incredibly effective to think about the concept of key drivers and factors by placing them within the context of a desired business outcome that aligns with the broader business vision, goals, and strategic direction. For instance, the dependent variable "sales" is extremely broad. If you are a company within the electronics industry, aiming to increase sales, the key drivers and factors will naturally differ from a company in the fashion industry pursuing the same goal.

Engage with key stakeholders and decision makers in your business. They will know what key questions need answering, and what key metrics are relevant to them. You can also apply a method similar to a Balanced Scorecard. Where you can define key strategic goals, and what influences their success. Specifying definitive metrics relevant to each area.

I completely agree with what Jacob stated below. The first step I will always take is going to the Gemba! See where the work gets done, how it gets done, and why it gets done. Then dive deeper into it by communicating with the people performing the work to understand where they see complications and their perspective on where defects arise. From there, I look into the data. Often times, the two methods will show different perspectives leading to a similar issue. By having a wider lens into the variables of a problem, it allows a team to use collective intelligence to analyze the issue and formulate a strategic approach to improve upon.

From an analytics PoV, identifying them is an easier process when you have a meaningful data, and more importantly a complete data. I would rather go back to my forecast model, to check and compare my market performance to identify key factors. Before a product launch, we always do a detailed forecasting model. We consider many interpolation and extrapolations to make the data fit to reality. Cubic Splines, Poisson and normal distributions and competency curves with a proper measure of inflation we take into count. So, best way to identify key factors of market performance could be compare it to a past forecast model. Sometimes, that might not work, if the model isn't built properly. But, definitely you will get a lot of information worthy.

Narrative-based Analysis: Engage with stakeholders, clients, and frontline employees for qualitative insights. Often, ground-level insights can highlight factors that are overlooked in purely data-driven approaches.

To identify key drivers and factors we use regression analysis in addition to correlation and causation. It is also important to have subject matter expertise. For regression we use linear/multiple regression, but logistic regression is actually a classification model and not used for regression.

In my experience, quantifying key drivers involves a combination of exploratory data analysis, statistical methods, and domain knowledge. Correlation analysis, as mentioned, is a valuable tool. However, it's worth noting that correlation does not imply causation. Therefore, it's crucial to integrate domain knowledge to interpret the relationships between variables better. While correlation analysis can identify relationships, it might not capture the story. A variable may not have a strong correlation but could be crucial when combined with others. Techniques like regression analysis, decision trees, and machine learning can provide more insights into complex relationships. Visuals can also reveal patterns, trends, and outliers.

Quantifying them is an easy process yet lengthy to understand. A wide variety of analysis can quantify them as mentioned e.g. Corr. Analysis. But, domain expertise, simple and flexible calculations are more important. A lot of graphical vis., numeric comparisons are made to quantify them in different scales and units. But making that simple and suitable for every perspective, is difficult. For example, if we talk about persistency in a drug market, you will find hundreds of tools, methods to show curves and directions. But defining them in a ideal way and making sense of that unit in all cases is something crucial, which can be solved if we focus on domain knowledge.

I would be very cautious utilizing these tools without specialized training in statistics. In a Lean Six Sigma project, a black belt would be utilizing this tool and a master black belt would be coaching and reviewing. You don’t want to draw the wrong conclusion and find yourself not only going in the wrong direction, but also losing support from your process owner and project champion.

Conjoint Analysis: This technique helps in quantifying the value that people place on different features of a product or service. By presenting choices and analyzing trade-offs, you can deduce the relative importance of each feature.

Sensitivity Analysis: By altering one variable at a time while keeping others constant, you can understand the impact of each variable on the outcome. This approach quantifies how sensitive the outcome is to changes in a particular driver or factor.

A few views could be especially helful for this task once you have your model locked in place: 1. A contribution analysis. Once your model is set, project your input features with estimated coefficients. Which drivers have the largest proportional effect, which smallest? This is helpful to understand the relative importance of various marketing drivers in driving outcomes. 2. Year over year changes in contribution. Look at what has changed in the projections between this year and last. What were top tangible drivers driving growth or decline in the business. This can have some key insights about marketing lever performance in driving growth. 3. ROI comparison. If we vie all inputs as investments, what has given most return?

Technically speaking, using tools like R squared, statistical significance, effect size measures like Cohen's, etc. On top of that I would add some contextual analysis, domain expertise, etc.

I have always found it helpful when presenting a complex analysis to break it down into 3 parts: 1. Explain the analysis you are going to show the stakeholders and the methodology used 2. Actually show the analysis with charts, graphs, etc. 3. Reiterate what you just showed them and end with your recommendations/next steps

This is very near to my heart - as I advocate simplicity! My advise is: (a) Max three metrics pr page. (b) All data is to be labeled - no discussion. (c) All pages/views must include a link to brief documentation on what I see. This documentation should provide the context of the view/report/page. 'Why does it exist'. 'What limitations has been chosen, and why'. 'Any obvious weaknesses in data/visuals' - and always write, what your assumptions has been, for the calculations/visuals. Reports and visuals are great - but what we really are looking for, are insights. "An insight is a concise, standalone sentence or paragraph that holds meaning for its intended audience - data and visuals are not insights on their own"

Your story should highlight the current business loss or the potential business benefit that could be won if you action your key drivers and factors. Then you should estimate the implementation cost (tangible and intangible) to impact the business outcome and demonstrate the added value of your recommendations. In order to fully engage deciders, it is always better to come with multiple proposals and their associated pros and cons.

I always try to adjust my language to my audience's level of understanding and present a compelling narrative that highlights key points

For a data analysis project, you can use frameworks, visualization tools, and storytelling techniques. Frameworks are methods of stating the problem, analyzing the data, sharing the results, and deciding based on the data. For example, SMART, SWOT, PESTEL, etc. Visualization tools are methods of making visuals, such as charts, graphs, maps, or dashboards that display your data clearly. Visualization tools can help you understand and share your data better. For example, Tableau, Power BI, Google Data Studio, Plotly, etc. Storytelling techniques are methods of writing stories that explain your data results. Storytelling techniques can help you connect and persuade your audience with the data. For example, STAR, SCQA, AIDA, SOAR, etc.

Despite being highly analytical, I discovered all the facts, figures and graphs mean little to the people doing the work. Those analyses merely allowed me to demonstrate to “higher ups” that I had successfully identified levers that encouraged improvement. And an area of focus where I could encourage said improvement. My advice? Do the analysis but try to never get so embedded in the data you forget these are people who contributed, not robots. My most effective program relies on tags in ServiceNow. The tags are shared with everyone in the department, as is the documentation that tells them exactly how their contribution is assessed. I tell my people how I think they are doing on a daily basis but remain open to pushback and explanation.

During a product development process, "design of experiment" is also a systematical approach, where we can identify the root cause, make relevant analysis to see cause and effect for each parameters. After these analysis, we can determine, which parameter has the bigger impact, in case if we decide to modify it. In addition, quality, cost and timing impact can be taken into account for a total value effect for each parameter change. Finally, we would have the complete analysis and by helping the visual tools to demonstrate a complete overview, findings, technical/economical impacts and finally decision. Statistical analysis are always good tools and techniques for the complex decision processes.

What’s missing is that at the start your identification process will be qualitative. Ask the stakeholders what they think they are. Certainly ask the process owner! Believe me, they will at least have some ideas, and you won’t be able to effectively proceed without getting their engagement and buy-in. Using the tools outlined here may yield some unexpected results that will require discussion and ultimately acceptance from these people!