The Evolution of the Site Report

1985 was a milestone year for me. I wrote my first site reporting system for the new Intel 386 based computers which sported a massive 640k of memory and a 10mb hard drive. The design requirements were twofold: first, that we could execute a radius report within 30 seconds, and second, that it would fit on a 20 mb Bernoulli cartridge – which was a monstrous floppy disk without the floppiness. We easily beat the time standard by several seconds, and by writing our own compression routines and dropping a few variables, we managed the storage requirements as well. The second milestone was that as we were seriously into retail modeling and never really satisfied with the census data, we decided to build our own current year demographics – and have been building them ever since.

So why recite ancient history, other than to make me feel old?

Our latest engine, Snapshot, generates that typical radius PDF report – complete with graphics and fancy formatting, in about half a second. If someone just wants the data, we can generate those at a clip of about 20,000 per minute.

But what about the content of the report? Sure, the underlying current year demographics are much improved as data sources, techniques, and computing capability allow us to quickly spot changes to the housing stock, capturing growth and decline much more efficiently. We have branched out into geodemographic segmentation, consumer spending and behavior, crime and environmental risk, business/daytime data, and even such esoteric things as modeling tourist behavior and measuring quality of life.

And yet, the report itself is just a gussied-up recitation of a set of numbers.

For some users, the data is simply an input into a model or site scoring system and the improvements in speed and data quality are much appreciated. But for the normal user, a site report remains just a mind-numbing list of numbers. They are still left to figure out for themselves what those numbers mean to make their decisions or to craft a location story.

Nearly forty years of progress, and we were still emitting just the facts. So, we have set upon a journey to bring the site report into the modern era by transforming it into a geographic storytelling system. We are not far into the journey, and have ventured into some rough neighborhoods, but have made some progress on several areas which are largely implemented within our Snapshot engine –

• The distance adjusted trade area, or ‘data’, report, which attempts to take account of what is inside the radius or drive time by giving less weight to populations at the edges of the area and more to the center. The name? Admittedly horrid, and users scratch their heads at what the unfamiliar new statistics mean. But over time, we know this will become integral to the trade area report of the future.
• Sensitivity analysis, where we look at how minor changes in radius size or centroid location can affect the results. Basically, we wanted to build fuzzy trade areas. We took it a step further, where we are looking at statistical distributions of a variable within a trade area. The median income can often mask substantial variability within the trade area. This probably overly complicates things for most users, but again, we are helping to craft our story telling abilities, even if we will keep it hidden from most users.
• Site signature analysis, where we attempt to encapsulate the essence of a geographic area in a single chart. It is intended as a simple, effective site screening mechanism. More on this later, but it will be available as an analytic tool in a major platform come winter.
• Percentile and z-score benchmarks, that analyzed millions of trade areas of all sizes to establish what values are in normal range, and more importantly, which aren’t. Either score can be attached many variables from our library within a site report or data extraction. These also form the basis for our new Executive Summary report which we are announced in this week’s Profile.
• We are deep into work to bring our spatial density functions and contouring algorithms to trade area delineation problems – by combining some nifty spatial statistics that we developed a few decades back with even biased samples from foot traffic data, we can delineate accurate customer density trade areas for existing sites. But why stop there? Mobility data doesn’t help when the previous store was a bakery and you are selling building materials, so we are working on predictive trade area models based on locational context and retail sector.

So how will we tie this all together? We are using AI and machine learning to cut through the chatter of the lowly site report so that your report is interpreted in context and the data can truly tell the story behind each location. The road ahead is long, and I freely admit to having already retracted my usual ‘I mean, really, how hard can it be?’ quip that I foolishly utter at the beginning of any major project.

One of our younger employees recently called me old at a staff meeting. True, perhaps, but crafty old dinosaurs are often able to not only learn new tricks, but to invent them. We at AGS have never been so excited about the future of geospatial analytics, and the role that our small and merry band of spatial lunatics can play in shaping it.