When gut feel isn’t enough: planning your modern fuel network with confidence

8 October, 2019

Share our post:

Kalibrate has been at the vanguard of network planning since the discipline emerged around the late 1970s. While the tools and techniques involved have changed, the three organising principles remain largely the same: mapping, data and predictive science. 

I wanted to take a look back at this evolution, discussing how science has increasingly blended with art, and what will come next. We talk a lot about data and automation. Both are huge, transformational innovations, but there’s always a role for human expertise. Data always leaves a gap that - for the most effective decision making - should be filled with the art of local knowledge and human interpretation. 

Then… limited science, lots of gut-feel

Mapping: print outs, dots on the map and radius rings. 

Data: census information and limited demographic detail. 

Predictive science: finger in the air, and kicking the dirt. 

Looking back to the late 1970s and 1980s, network planners relied on a lot of paper. Huge print outs of trade area maps, with dots applied for potential sites. Radius rings to define trade areas. Customer demographics came from census information, fairly limited in scope - only detailing home addresses, offering no insight into places of work or movement patterns. Not to mention, it could be nine years out of date by the time it was used. 

Real physical work was next. Getting on the road to ‘kick the dirt’ and get a gut-feel for the site. Even then, Kalibrate was working to complement these practices with science: generating maps, compiling surveys, and even using static aerial photography over remote sites to assess positioning and potential. 

As we moved into the 1980s and computers became mainstream, spreadsheets were introduced. Next, mapping evolved. The launch of online maps meant network planners could explore without leaving the office. This PC and connectivity revolution was probably the biggest step change in the network planning evolution. 

There was also then a huge progression in predictive science. Predictive modeling moved from simple spreadsheets to analog models. Gravity models helped planners understand where demand would come from and where volume could be taken.

Data availability changed more gradually, but steadily. We’ve consistently been able to capture better and better understanding of customers’ travel origins and movement patterns, and today, data volumes grow exponentially. 

Now... blending art and science

Mapping: everywhere online.

Data: traffic counts, shopping habits and loyalty cards.

Predictive science: capital optimization.

Data science is constantly and consistently evolving, but one of the biggest steps towards knowing consumers more intimately has come with loyalty cards and retailer apps, which paint a very comprehensive picture of a customer base. 

As drive time data has become more readily available, forecasting accuracy has improved - we can understand true routes and journey times. It will be interesting to see how these routes evolve with further growth in electric vehicles (EV) and alternative fuels, which require different predictive models. User behaviour is different, and the drivers for electric sales are different, so other approaches are required to improve EV predictive models.

Today, data abounds. But science only gets you so far. An experienced network planner still adds significant value. Data’s data, after all. It’s how you read it that makes all the difference. At Kalibrate, our specialty is predictive science - but we’re always looking to take it a step further. Capital optimization science is a good example of how we do this, running hundreds of simulations to identify the optimum combination of simulations that generates an outcome that meets our client’s objectives for any given market.

Next… diving deep, going mobile

Mapping: real-time info, geo-fencing.

Data: mobile and telematics.

Predictive science: deep learning.

The future is very exciting. We’re already in the realms of machine learning, but are now taking tentative steps towards deep learning, a very advanced subset of machine learning. The applications of deep learning techniques will enable us to effectively use unstructured data. For example, being able to extract site information from photographs to speed up the survey process, or learning to make automated network recommendations for the user. 

We’re always looking for ways in which to streamline surveys, and obtain data more quickly. Taking GoPros or similar image capturing devices to each site is one of the options we’re exploring to gather information. Coupling this approach with deep learning site analysis, and you’ve got the potential for highly accurate, rapid feedback - which in turn allows us to report back to clients even more quickly.

But the most exciting development is mobile data. It gives us a deep knowledge of the customer journey: where they’re going, how long they stay on site, how often they visit the same location, and maybe even what they’re buying. Mobile location services are so precise, they open up new levels of insight, both on a macro level and at a really granular level too. 

Built-in car navigation and telematics will offer further insight, and it will be interesting to see how movement patterns change with electric vehicles and self-driving cars, or even just the trend towards home working.

The exponential growth of data and increasing sophistication of predictive science will only continue to improve accuracy and reliability in network planning, but there will always be a role for human interpretation. The volume and depth of data available means the human element should represent the added value, though. Gut-feel alone isn’t enough to plan a modern fuel network with confidence, but a great network planner and analyst can still make a difference over a good analyst.

Add a Comment