The power of precision: AI-driven predictive analysis versus Amazon’s HQ2 decision

Making sense of massive amounts of unstructured data from news and other sources is complicated, but artificial intelligence can be leveraged to process large datasets to make intelligent predictions.

In September 2017, Amazon announced that they had issued a request for proposals (RFPs) across Canada, Mexico and the U.S. to establish a second headquarters to complement their main location in Seattle. More than 200 cities heeded the call and submitted proposals filled with promises of tax, construction and community betterment incentives. Amazon ended up announcing two new locations in November 2018.

Bitvore decided to use this huge corporate decision, which generated significant news coverage as well as business, industry and government speculation for more than a year, as a way to showcase how precision intelligence can be used in predictive analysis. The experiment had two goals:

  • To determine if, at each stage of the competition, it could be predicted which cities would advance to the next round.
  • To determine if the final location(s) of Amazon’s HQ2 could be predicted, before it became public information.

Backtesting and data

The Amazon HQ2 predictive experiment makes use of what data scientists call backtesting. This is where a predictive model is used on historic data to test its validity. Such testing is heavily relied on in the financial industry to give informed guidance on investments and strategy. Normally, backtesting relies on quantitative data and not quantified, organized and normalized news data.

Bitvore set out to see if it could predict the finalists and chosen cities in Amazon’s HQ2 decision using all the data available from their various products. Bitvore’s platform consumes massive amounts of world business news and runs that data through a proprietary AI platform that identifies only the most important business material events. The resulting data helps businesses make better decisions, faster.

AI tools enable businesses to make sense of the overabundance of information and news in real time. Using the Bitvore AI platform to monitor and analyze hundreds of thousands of articles a day can cut down on fruitless searches that yield completely unrelated and useless business data.

For an example, 10 hours of internet searching and going through tracked data (such as Google Alerts) can be narrowed down to 10 minutes of searching through intelligently analyzed information from Bitvore.

Analyzing the data

The Amazon HQ2 study was reviewed at three critical stages:

  • Amazon’s call for headquarter RFPs in September 2017
  • The shortlisting of 20 cities in October 2017
  • The final headquarters announcement in November 2018

The resulting white paper, highlighting the predictive value of AI-processed news datasets, includes the following:

  • Predictive experiment description
  • Experimental design
  • Calibration & prediction methods
  • Scoring
  • Final results
  • Predictive model validation

Download the white paper, Using AI Processed News Datasets to Perform Predictive Analytics, for a closer look.


Greg Bolcer is chief data officer at Bitvore, an Irvine, Calif.-based provider of precision intelligence derived from world business news and information. Its products are deployed in more than 60 of the world’s largest financial institutions, allowing them to rapidly create augmented intelligence solutions to address their unique business requirements.