More signal, less noise: How PredictHQ’s ranking works
Our ranking systems are world-firsts – here’s the role they play and how they work
When we first started building software to make sense of millions of real-world events so we could predict catalysts of demand, people thought we were crazy.
Event data is notoriously messy. There were few reliable sources and most cover only one or two categories. This meant we would have to aggregate a huge quantity of data from hundreds of sources to get the depth and diversity required to verify events. And then we would have to create smart systems to cleanse and enrich each event to create reliable intelligence for demand forecasting.
Once we were successful in creating forecast grade data, how on earth would someone know which of the thousands of events each week to focus on? We didn’t want to hit people with a firehose of data, but instead ready-to-action demand intelligence to inform and improved their planning immediately.
We realized swiftly it was critical to our PredictHQ vision to create a way to reliably rank events by predicted impact. But this had never been done before. It had never even been possible before because it requires our high quality data to build the machine learning models.
We have launched our first two ranking systems, with our first industry specific system – Aviation Rank™ – launching soon. We have detailed guides for our customers on our ranking systems, but we get a lot of questions about them, so here’s our introduction to how they work.
The TL;DR version:
PredictHQ assigns every event a numeric value on a logarithmic scale between 0 to 100.
These numeric values communicate predicted impact. More than 20,000 events each month are ranked 70 or above.
Accurate ranking is complex but essential for scalable demand intelligence and forecasting.
The highest ranked events have the biggest per-event impact, but clusters of lower ranked events occurring close together can have major impact. We call these perfect storms of demand. These can not be identified without PredictHQ’s ranking.
How PredictHQ Rank™ works
We launched our first rank, the PredictHQ Rank™ in 2016. PredictHQ Rank™ was designed to be a quick, scalable and reliable indicator of how significant an event will be for demand forecasting teams.
It is a general rank of event impact that draws on a wide range of factors beyond the obvious ones such as venue and location. Models built based purely on venue and location will inaccurately predict impact because not every event sells out. For example, a large sports field will be used for major games and concerts, but will also be used for far smaller crowds drawn to community fun runs or smaller acts and leagues.
PredictHQ’s ranking draws on a broad and deep set of additional factors including our proprietary entities, customized labels and unique repositories of historical data. We then process this with our custom-built machine-learning models.
To arrive at reliable event rankings, we have created six algorithms to power different ranking protocols for different event categories. Our categories span conferences to concerts to public holidays to natural disasters, all requiring different ways of gauging their impact.
How Local Rank™ works
Local Rank™ takes the general rank’s foundation and adds additional intelligence layers to identify the impact of an event on a specific location.
Local Rank™ matters because an event of 5000 people in New York City or Tokyo is barely a blip, but it’s a major demand distortion in places such as Aspen or Bora Bora. It’s used a lot by our customers because it can identify the difference between an event in different parts of the same city or location. It enables teams to identify local high impact events which may appear insignificant on a global stage.
Local Rank™ is not just the PredictHQ Rank™ factored by the population total of the location. It draws on its own unique algorithms combining a wide range of data about locations such as how built up an area is and its accessibility, as well as many others.
Opportunities hiding in plain sight and perfect storms
Both ranks are essential tools in your demand intelligence tool kit. For companies that have operations in multiple countries, picking up the signal of an event on the other side of the world is going to send accommodation, transport or retail demand through the roof is near impossible.
Both ranking systems—but particularly Local Rank™— are powerful tools to identify potential perfect storms of demand. A perfect storm of demand is when multiple events, large or small, occur near each other around the same time, say over a week or weekend. They also rarely repeat each year, as many recurring events change locations and timing each year. These clusters are frequent but fiendishly hard to see coming without the right technology. We wrote a guide about how to forecast and make the most of these recently.
What’s next for PredictHQ’s ranking
Our ranking systems are used by global companies across aviation, accommodation, transport, retail and many more industries. But each industry treats the insights from ranking in slightly different ways based on their business.
We work closely with our customers and have learned a lot watching how they integrate the PredictHQ API into production and their demand forecasting. And that’s why we’re so excited to be launching our first industry specific rank shortly.
Aviation Rank™ will gauge the impact of an event on air travel. It’s possible through access to unique historical data and extensive data science experimentation by our Chief Data Officer Dr Xuxu Wang and team. Aviation Rank will inform provide targeted demand intelligence for airlines.
This is significant because many major impact events for airlines are smaller than people realize. For example, a large professional conference may not be plucked out of the month’s event line up by demand forecasting teams because it has only 25,000 attendees, but our data indicates it is a significant catalyst for tens of millions in additional flight bookings.