How Can I Do Advanced Analytics in the Face of Uncertainty?

(Everyone is uncertain, whether they know it or not)

One thing the Covid-19 epidemic has done is teach us all a lesson about uncertainty. Everyone doing Business Intelligence (BI) analytics is wondering if old data even has meaning. They are uncertain. A good example is airline pricing. Algorithms were tuned on years of data. But a look at TSA checkpoints shows that since March, there hasn’t been a day when passenger counts were 50% of 2019.

TSA 2020 passenger compare 2019

Auto production swung even more violently. April 2020 saw the lowest car production in the United States in over 50 years; not just a little worse than the prior records, dramatically worse. Covid isn’t the only uncertainty. Covid only highlighted how uncertain we should be about many things.

So, what can we do to perform advanced analytics in the face of uncertainty? Lone Star has three answers.

1. We start with how humans react to uncertainty. Our species is bad at this. We are bad at recognizing uncertainty around us, bad at estimating it, and bad at making decisions as we face it. We need humility. There are other species, like pigeons who are naturally better at it than we are. Pigeons routinely beat grad students at games requiring uncertain guesses.

pigeons deal with uncertainty

To cope with this, Lone Star builds-in benchmarked best practices for uncertainty. That alone puts our solutions in the top 10%. Without these steps, you can’t handle mildly odd circumstances, much less a black swan, like an epidemic.

An example of best practice is helping customers avoid predictable traps. When anyone uses a single number (an average) to represent a span of uncertainty, it’s a red flag called the Flaw of Averages.

Another example is found in our strategic pricing software, TruPredict®. The Pro version of this software walks users through questions about the economic behavior of competitors. Prospect Theory and other behavioral economics show humans don’t respond “rationally.“ What seems like logical pricing in one firm won’t look logical to another firm.

We help users think about uncertainties with a method we call Little Questions®. We need to get them away from the uncertainty they usually consider, like a SarBox checklist, and think about all the risks they face. Humans need help with uncertainty. So, this may be the most important thing we do. Oddly, few others do it.

2. We treat uncertainty with respect. Uncertainty can be good. It can mean there are upside opportunities. It can be bad when we call it “risk.”  Either way, unless we treat it respectfully, we might miss something important. How can we treat uncertainty with respect?

slide rules analogTo start, we need many ways to find it and move it into our analytics. A disrespectful way is to assume the entire world is made of uncertainty, as described by the bell curve someone told you about. The “Normal” or “Gaussian” distribution is handy for doing math in the days of quill pens and slide rules. But the odds are you are reading this on a computer.

A respectful way to treat uncertainty is to accept it into modern computing in ways real users and real data easily accommodate. In hundreds of engagements and more than half a million data inputs, we have never seen a client who could tell us which dead mathematician described the shape of their data distribution.

We have many respectful data incorporation tools in our bag of tricks. Some software, like Lone Star’s MaxUp™ line, automatically handles it from live data feeds in real-time. Some software, line MRO2™ (software for maintenance centers), can rely on data pulls from customer databases or from expert estimates using TruCast®.

There are other respectful methods we use. We admire the SIPmath™ Standard (and sit on the standards body). We like Metalog distributions. These both allow us to use data directly and avoid antiquated methods.

3. We perform advanced analytics in the face of uncertainty using robust computational integrity delivered by our no-code tools. Applications built in our platforms, TruNavigator™ and AnaltyicsOS™ as well as our end solutions, like TruPredict™ and MaxUp Fleet™ have key features for dealing with uncertainty.

  • They maintain the full span of uncertainty across the entire computational thread
  • They are screaming fast in statistical and AI processing

“Full Span” matters because the uncertainty math simply doesn’t work without it. Some other solutions give up on this because it’s not easy. Memory management, processing times, and odd computational conditions are all easier when you give up on this.

“Screaming Fast” matters in diverse ways. If you need edge processing on a light IIoT platform, you can’t use fat, slow software. If you are in the cloud with unlimited potential resources, you probably don’t really want to pay unlimited processing and memory charges.


The picture of a sign at the start of this blog is from Uncertain, Texas. It’s a real town in a pine forest by a lake with about 100 inhabitants. Although they live in an Uncertain place, they seem to do quite well. You can live with uncertainty, too. But you may need a partner who knows their way around. If so, Lone Star would be happy to chat.