The Four Stages of Analytical Growth

At the start of your data journey, you may dream of when AI will guide your business.  A long stretch of work must happen before you can get to that level of data capabilities.  Gartner developed the Analytic Ascendancy Model, which helps show you where you are on your data journey.  Initially, your data is about getting information on your business. After time and effort, you can use that data to optimize your business.  There are a lot of articles on the web discussing the model, but they tend to have technical jargon that makes it hard to understand what it all means. I will provide you with my take as someone who has helped companies move along the model.  I can't say I have ever seen a company I have worked with ascend to the peak, but who wants to hit the peak when there is so much fun in the journey?

Descriptive Analytics: What Happened?

Your journey begins with hindsight data.  I have sat in interviews for companies where they want someone to come in and optimize labor and build regression models on day one.  When I asked them about their reporting capabilities, they said they were interested in a data science manager and didn't need reporting.  That is an immediate red flag for this job's long-term opportunities. Reporting is the foundation of any data culture, and descriptive analytics helps you understand where you are and who you have become. You truly must understand your history before you can predict the future.  Machine learning models are built on historical data. While you clean and prepare that data for models, your company will have a more significant ROI by building your reporting capabilities and understanding what is happening.

I am constantly surprised by the number of companies that need a reliable foundation to know where their money is coming from!  As a small business, this stage of your journey is usually led by a Business Intelligence (BI) or Data Science manager and, if lucky, a data analyst with some SQL or data engineering skills.  This is enough for a small company to invest in and get immediate actionable insights from your data.  I recommend investing in a reporting tool like Power BI or Tableau during this time.  A small but mighty team can help build out KPIs and measures to start tracking your business while you prepare for deep analytics later.

Diagnostic Analytics: Why did it happen?

Once your company understands what is happening, you can begin to ask why it happened.  This step is critical in moving to modelings because you must understand what to put in a model.  Data scientists take data points to predict the future, but choosing the wrong data can cause models to be overly focused on certain parts of the company and overlook others.  If you don't understand why something happened, how do you predict it in the future?  This stage of the journey is usually where data cultures begin to fall apart.  Many of the answers to "Why did this happen?" are built around the gut instincts of those who have been around the company for a while and not around the data.  I have worked with companies that want to begin predicting the future before they understand why the company is performing the way they are.  This stage of the data journey will never end, but it needs to be fruitful before you can move on.  I have personally built statistically sound models to predict traffic and revenue. Still, they always had more variance than a reliable model should have.  The variance happened because I didn't understand the data or the company before building the models.    At this journey stage, the BI team needs to reach out to department leaders and see what data they have stored away.  Also, it would help if you began looking for additional outside data to support your theories.  Invest in getting weather data, or see if your payroll processor can provide you with your labor data.  Always remember that vendor companies hold your data, by the way. Make sure you keep that in mind when signing contracts. I worked with a company that took a slightly lower cost fee from a vendor to find out later that the contract prevented them from getting their data from the vendor. Start collecting new data with the same zeal as Thanos collecting infinity stones!   Storing data is relatively cheap nowadays.  Please take advantage of that, grab everything you can, and centrally store it.  By now, you should have a data warehouse to connect all this data and create feeds to Excel or your reporting tool for departmental analysts.  This will ensure that everyone uses the same data without making mistakes.

Predictive Analytics: What will happen?

Now, your journey moves from hindsight to foresight.  This stage of the data journey is where most companies want to be the day they decide to invest in their data culture.  Where can we open a new location?  What will traffic be like in two weeks so we can flex part-time hours?  Predictive analytics is where you can begin to see the EBITDA impacts of your investment.  You have been reporting and helping people understand what is happening while relying on field leaders and senior support staff to guide the conversation.  At this point, your BI team will guide the discussion.  This stage is very shaky, however.  Your team's first few models will probably not be that great. It is going to take time and patience and allow data scientists to understand where their shortfalls are.  It sounds wild to accept mediocrity, but it is your best option for a while.  Analysts need to learn, but the key is to allow them to fail fast.  Focus on short-term models and predict less impactful measures.  When the model turns out to be wrong, you fall back on your diagnostic analysts.  Why did the model fail?  What piece of data is preventing you from getting over the hump and driving real impact on the business?  This is why you have to develop diagnostic analytics before moving on.

Models are never one-and-done. They take time to develop, and whenever you find a new data source, you find a new opportunity to increase your R squared!  If you don't know what R squared is, this is the point you begin to find out.  I recommend hiring a junior-level data scientist at his end to help out.  These applicants usually have a master's degree or have a couple of years of experience.  They demand a higher salary but are worth it if the hire is right.  I love giving data scientists the chance to present their work early on.  They usually crash and burn the same way I have several times.  They will present a model with all the bells and whistles and want to ensure you know it!  They will use terms no one in the room understands and be super frustrated that no one "gets it."  This is a huge growth opportunity for them, and if you hire the right BI lead, they can help develop your new data scientist into a critical member of your company.

Prescriptive Analytics: How can we make it happen?

At this point, your BI team will have proven to be a reliable insights provider.  Your company will call on them to make recommendations, not just predictions.  The data scientists can prescribe actions to overcome headwinds in their predictive models.  The machine learning models are built by taking data points and seeing their impacts on the ultimate result.  Now that your data scientists have spent time making their models accurate and understanding the data that goes into them, they can see which factors are the most controllable.  They can recommend how the different measures interact and where you can get the best ROI.  This is the stage where you can do fun things like pricing strategies and optimizing labor allocation to maximize profits.  The BI team and data scientists are now fully integrated with corporate leadership and providing actional insights. I don't have much to say in this area because many small and medium companies will force themselves to this stage without the foundation to sustain it. I intentionally keep this section short because the other stages are where your focus needs to be!