Dimensional Constraint Analysis

Sounds a little scary doesn’t it?  It’s a fancy term that means you can "constrain the data" any way you want to get your analysis.  This architecture is used for analyzing data to whatever depth you want based on what you found from your last analysis or when your are just searching for something.

As an example, I was analyzing fast food ordering data for what type of modifiers people ask for at what times of the day, or on which days of the week.  In fast food restaurants, you get rush business as everyone gets out of work for lunch or off for the day.  A hoard of people all show up at once and they all want to eat as soon as possible and get back to work or on with their evening.

When people modify a sandwich it means it has to be prepared after they order it, and not "pre-made".  When you see a hundred people converging on a restaurant, believe me, you want to start cooking ahead!

As we ran the analysis, we started with the bursts.  We analyzed the data and discovered that the burst happens between 11:30 AM and 1:30 PM and again between 4:00 PM and 6:30 PM on weekdays.  This became the first constraint, why look at any data outside of that?

The second analysis looked at which of these periods were the highest for requested modifiers (no tomato, extra sauce ect.).  It turns out that Fridays were the highest day for requested modifiers.  So we now use that as a constraint.  Let's look at modifier requests on Friday. 

We examined what products (sandwiches) were modified the most on Friday during the lunch and evening rush periods. Nothing jumped out, so we decided to see if there was anything about WHAT they were asking for and BINGO. 

The overwhelming modification was to remove onions….   We could actually create a product mix that modeled the requested modifications. 

The results?  Create a "pre-make plan" that cooks a percentage of the orders without onion so that you can cook ahead and handle the rush more efficiently.