Value Driven Analytics

05/19/2024

If you've ever needed to calculate the return on investment of a long-term investment, you may be familiar with the concept of net present value (NPV).

Calculating the net present value of cash flows is especially important when the payoff of an investment occurs over a long period. This is due to the time value of money, which suggests that a dollar today is generally worth more than a dollar in the future.

Learn more about net present value, the time value of money, and discount rates and how these concepts work together to enable effective ROI analysis in the article below!

Calculating the Net Present Value (NPV) of cash flows for an investment is important because of the time value of money, which suggests…

05/11/2024

One of the most frustrating aspects of data analytics and data science, from both a stakeholder and data analyst’s perspective, is bad data.

In our experience, a typical organization has database issues that impact data availability or data quality about 5% of the time, but this varies significantly across organizations and it’s possible to reduce this even further through improving the business intelligence / ETL processes that update your database.

Downtime and data quality issues can, at worst, cause stakeholders who don’t know that the data is inaccurate to make bad decisions based on bad data. It can also make your teams less efficient and less able to make timely, data-driven decisions. If it happens often enough, it may even impact stakeholders’ trust in the data, making them less likely to make data-driven decisions even when the data is correct.

Sometimes validation and attention-to-detail processes need to be put into place to reduce the risk of this happening. We recommend that organizations put in place daily automated data validation checks to proactively alert teams when unexpected data scenarios occur. This allows your team to alert stakeholders and identify the root cause quicker, lowering the risk of making bad decisions on bad data.

For more data validation tips, consider checking out this video: https://youtu.be/wGU3mZDnZmo?si=yUlonEOw1mDN7jLC

In this video, I'll provide 4 tips to lower the risk of errors in your analysis and data science projects. This is critical for your career growth as attenti...

05/04/2024

Most leaders and stakeholders have a desire to understand how they’re tracking to the current month or quarter’s goal. A robust index model can help them do just that.

Imagine trying to assess current period performance without an index model. Perhaps…

- Stakeholders compare period-to-date actuals to the entire period’s goal; but, unless the company’s performance has dramatically outperformed, it’s likely this will always create a false sense of lagging behind until the period is almost completed. Perhaps some stakeholders will prefer this approach that creates a sense of urgency, but, more likely, viewers will come to just expect that they’re in the red because it’s early in the period and never really know if they’re on pace to hit their goal

- Recognizing the pitfall of the former approach, stakeholders might decide to just wait until the end of the period to assess performance for the most recently completed period. While the actuals and goal would be “apples-to-apples” at that point, it’s too late to respond to performance when the time period has already completed. Most stakeholders prefer to know ahead of time if they’re on track or off track to hit their current goal. That being said, should a company’s strategy really change depending on whether they’re on track to under or over deliver vs. a goal? Shouldn’t a company keep investing in positive ROI investments and activities regardless of whether they’re on track to hit a goal or not?

- Stakeholders might seek a simple compromise between the 2 approaches above that takes the current period goal and divides it by the number of days in the time period. The current period-to-date actuals are then compared to the prorated goal. For example, if a quarter has 91 days in it and we’re 10 days into the quarter, the quarter-to-date actuals would be compared to 0.11 (10/91) x the goal. By comparing to this prorated goal, the company can get a better idea of whether they’re on track to hit the current period goal based on their performance so far.

- In some cases, stakeholders may take a similar approach to the one above, but only prorate the goal based on days that the company is open for business or days the company generally does a significant amount of business.

While these last 2 approaches are usually better than the first 2 (making no adjustment to the goal at all or waiting until the end of the month to get a read on performance), these simpler approaches don’t consider that certain days may generally perform better than others. They also don’t consider historical momentum within a quarter or month or the impact of holidays.

That’s where a data science-driven index model comes in. You can robustly see how you’re performing against your goal at any point within the month by using an index model. An index model uses historical performance to estimate the impact of day of week, week of month, month of year, and holidays on sales. It ultimately suggests what % of a time period’s achievement is expected to come in on day 1, 2, 3, etc….This index model can be multiplied by the organization’s goal for a given time period to parse that time period’s goal out by day. With this, an organization can track period-to-date actuals against the period-to-date prorated goal to see how it’s tracking on any day within a period, proactively alerting stakeholders on performance issues before it’s too late.

Call now to connect with business.

04/27/2024

If you were managing a retail company with multiple locations, how would you determine where to build your next location?

Would you consider available real estate in what seems like a good location for a good price? Perhaps you’d assess the price of real estate compared to how many people in your target demographics are nearby?

These methods could be a good start. The downside is they fall short of actually estimating whether a prospective location would be profitable or not. They may also fail to identify the demographics that the company has historically performed the best around and fail to weight each demographic properly (v. weighting all target characteristics the same). Based on these drawbacks, your organization could easily invest in some non-profitable locations and miss out on highly profitable locations.

Especially if your organization has opened quite a few locations in the past, your organization would highly benefit from a data science model that estimates how many sales a potential new store location would generate based on the surrounding population, demographics, and competitors and how these factors have historically related to sales.

You could apply this model to all real estate opportunities available and compare the estimated sales and profit margin to the location’s cost.

Imagine having an interactive dashboard where you could rank these prospective sites by expected profitability and, for any given location, be able to see what your expected profitability would be at any offer being considered.

If your company doesn’t have access to something like this today, perhaps you’re just a little data science knowledge away from making this a reality!

If you haven’t already, consider watching our free Python-based data science training that teaches modeling principals through a college football game prediction use case. These same principals could be used to build a prospective retail store sales prediction model instead!

https://youtu.be/vrYhBEjHBcQ?si=9tbMBqGW30hiOof1

In this 6th session of the 7-session Definitive Guide to Data Analytics series, we explain and demonstrate in Python many of the core aspects of data science...

04/20/2024

Have you ever wondered how deep learning neural networks are different from other types of statistical and machine learning models?

Deep learning neural networks have dramatically improved over the last 80 years and are often the most effective way to perform complex prediction tasks like image detection, voice recognition, natural language processing, and beyond.

However, deep learning neural networks can often also be used to improve upon the performance of more classical classification and regression use cases, particularly when the dataset is larger and there are more complex relationships between variables.

We recommend testing deep learning neural networks on your predictive model use cases to see if they can improve upon the performance (on a holdout dataset) of simpler statistical and machine learning models, especially if explainability of the model is not required.

Deep learning neural networks are also behind large language models like ChatGPT. If you’d like to learn more about these, check out the video below that explores how these are created!

https://youtu.be/U6CmM9wlaVg?si=PsauUD5wQvWiDZDX

How Was Chat GPT Created | Learn Data Science Through AI🚀 Dive deep into AI with our latest video: "How Was Chat GPT Created | Learn Data Science Through AI...

04/13/2024

Last week, we posted about what analytical rigor really means, including a couple examples. This week, we wanted to follow that up with another example and, most importantly, recommend a few ways analysts can further develop their analytical rigor.

Imagine an analyst looking at quarter over quarter ice cream sales in the winter time, noticing that sales in the winter have been lower than the fall and that was lower than the summer, and then suggesting the company is seeing a severe decline in performance and forecasting that to continue.

This analysis would lack rigor because it failed to recognize and take into account the seasonality of the business; so a better analysis might be to look at year over year growth over time.

Accuracy and analytical rigor are important because, without them, an analyst could perform an analysis and get stakeholder buy in for something that could actually make the company worse off.

With that said, there is not necessarily a straightforward path for an analyst to grow in analytical rigor. There are plenty of coding tutorials and practice platforms and a fair share of leadership and communication trainings too, but instruction on analytical rigor seems to be much rarer.

Most often, analytical rigor seems to be built over the course of time performing analysis and data science.

But analysts shouldn’t assume their analytical rigor will simply grow with time without intentionality. And perhaps there could be a way to speed up this process!

In our experience, the best way an analyst can grow in analytical rigor is through collaboration with other analysts.

Seeing the methodology behind other analyst’s analysis, even analysts supporting other functional areas or industries, and then thinking about how a similar methodology could be adopted to the challenges they’ve been tasked with solving is a great start!

The mentorship of an experienced analyst to provide feedback on an analyst’s initial approach can also be helpful! Add in initiatives like peer review and the occasional hackathon/one-day solution and your analysts will dramatically accelerate their analytical rigor development!

In conclusion, analytical rigor is a critical aspect of any analysis and analysts can generally improve in this area through collaboration with other analysts. Remember, “the impact will be bigger if you apply analytical rigor”.

Call now to connect with business.

04/06/2024

One of Value Driven Analytics’ differentiators (perhaps the most important one!) is the level of rigor behind our analytics and data science solutions that comes with more than a dozen years of experience; but what exactly is analytical rigor?

It’s a fair question and we realize it can be a confusing term! The dictionary has several definitions. The one most related to analytics is “strict precision”, but this is still a little abstract.

So we thought we’d expound upon what analytical rigor means to us at Value Driven Analytics with a few specific examples.

While coding skills can be incredibly helpful for an analyst’s efficiency and flexibility, analytical rigor (in addition to communication, leadership, project management, and attention-to-detail) can really differentiate an analyst!

Analytical rigor is taking a business challenge and turning it into an analytical approach that would then be implemented with analytics tools (Python, SQL, Power BI, etc…) in a robust way. We often use rigor and robustness as interchangeable terms to describe the same “strict precision”. Another way of describing analytical rigor is analyzing something in a way that truly achieves the business goal.

Analytical rigor is sometimes thought about as accuracy, although we tend to think about accuracy as the numbers being right and rigor as taking those numbers and analyzing them using a methodology that assesses the business use case properly.

Oftentimes, there can be a few valid approaches, but it requires a strong analytical thought process and often a good amount of creativity to come up with a rigorous analytics approach.

When it comes to building predictive/prescriptive data science models, an example of applying analytical rigor could be making sure the model doesn’t have any “data leakage”. This means that the features in your model should only consider information that would have been available at the time of prediction.

Another example could be controlling for all the necessary factors in an analysis, not assuming causality between 2 related variables when it could really be pure correlation.

For instance, an analysis that finds people buy more sweaters when they turn on their fireplace and then recommends that the sweater company do an advertising campaign to encourage more people to turn on their fireplaces expecting this will lead to more sales would not be rigorous!

This analysis would need to control for weather / time of year and then find that is actually what’s driving sweater purchases and, after controlling for these, would probably find no correlation between fireplace use and purchasing sweaters; so that would be an example of a more rigorous or robust analysis!

If you found this post to be helpful, look out for a future post where we'll plan on sharing more examples of analytical rigor!

03/24/2024

In the process of building a data science model, have you ever skipped the step of assessing your model on a holdout dataset?

There is a concept in data science, one that should be avoided, known as ‘overfitting’ a model; this happens when a model is trained too closely to a training dataset to the point where all the predictive power essentially goes away when it’s applied to a new dataset with observations the model has never seen before.

Data scientists should test their model on a “holdout” dataset (or sometimes called “validation” or “test” dataset) before deploying to ensure that its predictive power stands even when it’s applied to data it has never seen before. This essentially ensures that the model is not “overfit”.

Beyond simply setting aside a randomly selected % of the modeling dataset as a holdout to test a trained model, analysts and data scientists may be able to get an even better, less variable read on overfitting and model performance by using k-fold cross-validation. This essentially creates k (often set to 5 or 10) different holdout datasets to validate the model against, which reduces the variance of your holdout dataset model performance compared to just taking a single randomly selected cut.

It’s also a good idea to train your model on several different time snapshots and test it on other time snapshots. This can ensure that your model is built and validated on trends and relationships that stand the test of time; in other words, it’s not built on trends that stand in one time period and not in another. This is particularly important when variable relationships and trends may be changing over time, such as in the stock market.

Testing your model on an independent holdout dataset is part of building data science models in a robust manner that will drive value for your organization. You can learn about this concept and many others in our data science training YouTube video!

https://youtu.be/vrYhBEjHBcQ?si=MX2gO-WaQg9FzsB7

In this 6th session of the 7-session Definitive Guide to Data Analytics series, we explain and demonstrate in Python many of the core aspects of data science...

03/17/2024

Have you been thinking of learning data science?

We would highly recommend that any data analyst learn to build statistical and machine learning models and apply them to high value use cases.

Data science algorithms can be used in every functional area of your organization to better predict or understand something of interest. This can help your organization be more efficient and effective in a variety of ways:

💡A propensity-to-buy or sales effort impact model could be built to provide guidance to your sales team on which customers in their territory they should prioritize contacting.

💡A simple time series model that predicts future sales could help your manufacturing team plan their production schedule in anticipation of expected demand.

💡A media mix model could help your marketing team understand which channels and programs have the highest v. lowest ROI to determine what to invest in more or less in the future.

These are just a few ways that data science modeling could help your organization.

Even though data science models can create so much value, there can be several reasons why data science modeling may not be happening at an organization including 1) analysts not having time and 2) analysts not having data science skills.

If you’ve been contemplating learning data science, join more than a thousand others and start by watching our free hands-on data science training video on YouTube. In this video, we teach core data science concepts through the creation of a college football game prediction model. You can create your own model alongside the training!

Learning data science is a worthwhile investment as it can help you drive more value for your organization through the actionable insights that come out of the models you create.

🔗 Watch the full video here: https://youtu.be/vrYhBEjHBcQ?si=taNwi41bK78ynBmM

In this 6th session of the 7-session Definitive Guide to Data Analytics series, we explain and demonstrate in Python many of the core aspects of data science...

03/12/2024

Have you been thinking about learning an advanced analytics tool like Python, R, or SAS?

We’d highly recommend learning Python in particular as an efficient (i.e. fast to run on larger datasets), general purpose (i.e. flexible), and open source (i.e. free) coding language that has become more and more popular for data analysis and data science.

It can be very helpful for analysts and data scientists to know at least 1 of these 3 tools for advanced data analysis, automation, and data science use cases. While SQL can be quite useful in the process of these tasks and can be used directly within each of these tools, sometimes SQL alone can’t entirely accomplish these more advanced use cases.

Using these advanced tools to code out a data process, analysis, or model build can lead to efficiency gains as the code can easily be tweaked, scheduled to run automatically, and reused for similar projects in the future. They also enable more flexible data manipulation, analysis, and data science than Excel and even SQL sometimes through operations like API calls and looping/iterating (although some forms of SQL enable loops!). If you can dream it, you can build it!

While the idea of learning a coding language may seem daunting to many analysts who are used to using point-and-click tools for analysis, all 3 of these advanced analytics tools are widely considered to be much easier to learn than other lower level coding languages like Java, C, etc…

Consider finding out for yourself by starting with our hands-on introductory course on Python, and start boosting your efficiency and flexibility today!

https://youtu.be/LWOoW269pwc?si=uQuqIjORyCDMZXtU

Python Course for Beginners | Best Way to Learn PythonIn this 4th session of the 7-session Definitive Guide to Data Analytics series, we demonstrate some of ...

03/11/2024

If you’re an aspiring or practicing data analyst and haven’t learned SQL already, we would highly recommend investing in learning it!

SQL is truly an analytics staple. It’s evident in the fact that nearly every analytics platform including Python, R, SAS, Alteryx, Power BI, Tableau, dozens of database platforms, and more have ways of running SQL code within their platform. SQL is one of the best ways to select, merge, modify, and aggregate data known to analysts.

Perhaps you’re not using SQL today because you’re relying more on other coding languages like Python, R, or SAS, which are great analytics coding tools as well; hopefully, it’s not because you’re relying on manual Excel clicks to manipulate data.

While using Excel may seem easy and efficient the first time the data is manipulated, it can become inefficient and disengaging when needing to repeat these clicks over and over again in the future to refresh an analysis or report based on the latest data. It can also limit what kind of data manipulation is possible as more complex data manipulation may not be practical in Excel.

Instead, we’d highly recommend coding out data manipulation with a tool like SQL. The code only needs to be written once and can be scheduled to run/refresh automatically from there. Just about any kind of data manipulation is possible and relatively easy with SQL.

While the idea of learning a coding language may seem daunting to many analysts who are used to using point-and-click tools for analysis, SQL is one of the easiest coding languages to learn, nothing like Java, C, etc…

We encourage you to find out for yourself by giving it a try. Start with our SQL training playlist that assumes no prior knowledge and begin your journey to efficient and flexible data manipulation: https://youtu.be/CGDdCyI_3co?si=SpgO1yft8xJvL2yl

In this 5th session of the 7-session Definitive Guide to Data Analytics series, we demonstrate some of the main parts of a SQL query by walking through an an...

03/09/2024

Is your organization’s reporting updated manually and static in nature (no filtering capabilities)? If so, we would highly recommend investing in an interactive visualization tool like Power BI or Tableau.

Not only can these tools automate the refresh of your reporting (saving your analytics team time and constantly giving you access to the latest data), but they also add interactivity to your reporting, allowing stakeholders to immediately drill into time periods, geographies, and business segments. This is a better experience for stakeholders (allowing them to act on data quicker) and can lead to fewer follow-up requests for analysts, giving them more time to pursue advanced analytics projects.

These tools are fairly easy to learn and require little to no coding, although SQL can be used to manipulate the data imported in. If your analytics team is used to using Excel for reporting, the formula style in Power BI, another Microsoft product, may seem quite familiar to them. From a cost perspective, downloading Power BI desktop is free and, if your organization is already purchasing Microsoft E5 licenses for its employees, you already have access to use Power BI online (aka the Power BI service) at no additional expense; it’s just a matter of your analytics team learning to use it and putting that knowledge into practice by migrating old reporting and building new reporting in Power BI.

If this sounds good to you, getting started today is easy. Download Power BI Desktop for free, watch our comprehensive hands-on Power BI course on YouTube (uses a college football use case), and start experiencing the value of interactive dashboarding today!

In this 3rd session of the 7-session Definitive Guide to Data Analytics series, we provide a fairly comprehensive overview of Power BI, all while walking thr...