Reading Predictive Tea Leaves With Smart, Big Data

The old adage that a butterfly flaps its wings and causes a hurricane is especially apt in business. After all, ripple effects cascade across firms’ top and bottom lines all the time.

Demand for a hot product in one part of the world, stockouts or even the coronavirus can all impact corporate fortunes, both short- and long-term. Missing the opportunity to gauge demand accurately can result in lost sales, and ramping up production or staffing while waiting for foot traffic or clicks on buy buttons that never materialize can ding operating profits. Demand forecasting is as much an art as a science, but it could use a bit more science.

Predictive analytics can help companies pinpoint what might lie ahead, and plan for contingencies. However, as Cam Brown, CEO of PredictHQ, told Karen Webster, the key is to find “smart” data buried within the Big Data.

The conversation came against a backdrop where the coronavirus has caused retailers and tech giants to scramble, readjusting supply chains, closing locations and idling staff. Against an even larger backdrop, the rise of the gig economy has brought urgency to platform-focused firms — operating internationally — to pinpoint where demand waxes and wanes in real time.

As Webster noted, only slightly tongue-in-cheek, there’s data on top of data: flight data, tracking data, pricing data, currency fluctuations and even pollution data. Making sense of it all can be a challenge.

“When people look at the data they have internally, it doesn’t enable them to understand anomalies,” said Brown, and managers are unable to uncover and discover cause and effect as quickly as they might with better insight.

The PredictHQ platform aggregates and verifies about 2 billion data points across a single “global events” application programming interface (API). That data, in turn, can help millions of events across 30,000 cities globally, impacting demand for companies’ goods and services, and enabling them to pinpoint why consumers’ desires fluctuate across retail, hospitality and transportation verticals.

Clients include Accenture, Booking.com, Uber and others, the company noted.

Brown explained that the genesis for the PredictHQ offering sprang from running a car rental booking platform — where seeing the peaks and troughs in demand across the globe led to retroactive reflection that certain events, such as concerts or school holidays, had influenced such swings.

“We decided to flip this on its head and ask, ‘What if we knew what was coming, and what if we could plan for it?’” he said.

The Way It’s Been Done

Such data analysis is not scalable, or perhaps even possible, with the entrenched method of creating spreadsheets — forecasting demand off those spreadsheets and sending those forecasts across far-flung, possibly global locations within a company.

“The way in which forecast models used to work was all based on historical data,” said Brown, a foundation that is no longer as useful as it was once.

As he told Webster, picture the hematology conference in Austin — which can draw 25,000 attendees — taking place on the same day as a huge sports rivalry in town. Another example: an event that used to be held in Boston that has now shifted to San Diego, sucking all the attendant commerce to a new location across the coast.

Brown said there can also be “perfect storms of demand” that would be missed by manual, piecemeal data collection.

Delving Into The Process — And The API

In an illustration of how it all works, he said the global events API can enable a café to identify an event (say, a concert) occurring 0.6 miles away — which, with proper analysis, could lead that café to bring more staff or inventory on-premise. It might even start a short-term promotional campaign to target maximum foot traffic.

Drilling down a bit, he noted that there are several types of events, and data, regardless of vertical. There are scheduled events, which can be one to two years in advance, and there are unscheduled events, such as natural disasters or airport delays. There is also historical data, which Brown said is used to assist in forecasting. Amid the thousands (or even hundreds of thousands) of data points that flow through the API, the goal is to boil it down into actionable insight for enterprise clients through a “knowledge graph” — tied to other, alternative sources of information that can also take into account events that feature celebrities, sports venues and the aforementioned industry conferences.

Such granularity and classification of data, he said, can help Uber forecast the demand materializing in real time on, say, 6th Avenue, and gauge what demand may look like tomorrow or in three weeks’ time. As he noted, for the Ubers, Grubhubs and Geminis of the world, the challenge is getting the drivers into the right place at the right time.

Alternatively, for the coffee shop chain facing a positive swell in demand, having four staffers on hand rather than two at a particular location in, say, Boston can make a huge difference in that outlet’s efficiency and sales.

Brown noted that of the hundreds of thousands of events tracked by his firm, 15 percent are recurring events. Of that 15 percent, as much as 85 percent of the events are changing locations or dates every year.

“If you get a difference of a month and a half [for an event], your old-school forecasting algorithms wouldn’t have picked up on that shift,” Brown said, across demand’s peaks and valleys, where firms can recalibrate what is needed, even when events do shift and they need to offset lost revenues.

Use of the PredictHQ offering, he added, improves the accuracy of client firms’ forecasting models by 25 percent to 32 percent.

The Black Swan Events

Delving into forecasting, of course, begs the question of how — and whether even if — firms can gird for black swan events. That’s especially timely amid the coronavirus epidemic. Brown maintained that the data might not be able to predict the coronavirus, but it could help firms mitigate the impact much more quickly.

For an airline, Brown said, “the mitigation could include a change in pricing or a reduction in flights, or perhaps they should put more flights into other regions where they might not be impacted by the coronavirus.”

The company, founded in New Zealand, recently raised $22 million in a Series B funding round, led by Sutter Hill Ventures and others. Brown said the funding has been earmarked to grow staff internationally, including in the U.S., and the firm has opened an office in London.

“On the surface, people believe that Big Data is relatively simplistic, but it’s just so complex, and the dynamic nature of that means you’ve just got to grow this vast kind of knowledge,” he told Webster. In an effort to understand: “How big is an event, and how does it change over time?”