On stage at a SAS Canada event hosted in Toronto, Creative Destruction Lab founder Ajay Agrawal invites the audience to engage in a simple “science-fictioning” exercise – imagine your business had a dial that could turn up the accuracy of your ability to predict things.

At what number would turning up that knob suddenly change the entire way you do business. Agrawal points to a familiar product-recommending ecommerce giant to illustrate the effect this would have.

“For Amazon, they don’t have to get to Spinal Tap levels of accuracy, if they can get to six out of 10, then someone at Amazon says ‘why are we waiting for people to even order, why don’t we just ship it?” he says. “Now you just return what you don’t want and the competition is preempted that you might buy the products you do keep elsewhere.”

Amazon has already filed the patent for “anticipatory shopping,” so this may not remain sci-fi for long. Shoppers that find they suddenly have items they needed on their doorstep, with the only nuisance in exchange being to leave unwanted products in a box for pickup, may see it as a convenient service. But in order to deliver it, Amazon has to get even better at predicting the products you want to buy.

The good news for Amazon is that the cost of prediction is getting cheaper. In effect, the sci-fi “accuracy knob” Agrawal asks us to imagine is simply the passage of time. As the cost of arithmetic has fallen over the past two centuries thanks to increased processing capabilities, predicting outcomes for many things has become a simple matter of applying the right models to the right data. That’s why applicable artificial intelligence (AI) is suddenly within reach for so many businesses.

“One way to think of AI is to think of it as a drop in the cost of prediction,” Agrawal says. “Everywhere it says ‘AI,’ just replace it with ‘cheap prediction.'”

The results of cheap prediction will have three major effects, he says:

  1. We’ll use more prediction and convert existing business problems into “prediction problems.”
  2. Complements to prediction will increase in value. AI can’t provide judgment, Agrawal says, just predictions based on data. So humans that can offer experience that lends itself to good judgment will be more valued. Also, the value of data will go up since it can inform predictions.
  3. Substitutes to AI prediction will fall in value. People that are asked to make predictions for their work now may want to look for some new training.

While getting better at predictions is the top priority of Silicon Valley tech giants like Amazon and Google, other more traditional industries are also buying into the trend towards free predictions. Among the top clients of event host SAS Institute Inc.’s predictive analytics software in Canada are banks, including Scotiabank.

On a panel, Vishal Gossain, vice-president of risk management at Scotiabank, says that the bank is using AI in three main ways. By predicting customer lifecycle models, improving customer experience with chatbots and better fraud detection, and automation of internal processes such as evaluation cases. Improved abilities in these are so powerful, Gossain expects $100 million in risk adjusted returns to the bank, he says.

“We have many proofs of concept that we’re scaling. The collections department has a deep learning model that’s been in production for about a year,” he said on the panel. “It has materialized into a multi-million dollar savings on the portfolio that it was applied.”

SAS also offers an Anti-Money Laundering solution to banks, which helps to understand if the risk coming from a new customer is too high, says Amanda Holden, solution executive of fraud and security intelligence at SAS Canada. It’s now common for criminals to attempt to defraud banks by applying to lines of credit.

“A lot of fraud is moving to the application process,” she says. “There’s analytics that boils it through some information processes and comes out with a number. The higher the number, the higher the risk.”

Typically, banks expect traditional methods of anti-money laundering detection to be a false positive nine out of 10 times – meaning that most instances are caught, but for every instance caught, nine legitimate instances are wrongly flagged.

SAS solves that problem by taking much more data into consideration when determining the fraud risk. Instead of the usual five to 10 pieces of information to indicate fraud, it’s looking at 300 to 1,000 pieces of information, says Mike Ames, senior director of product management at SAS Canada. It also uses natural language processing to put the items that flagged it into a narrative that is easy to understand.

“There’s that balancing act of between the good customer experience and protecting the bank from risk,” he says. “You want to make sure it’s truly necessary to conduct that investigation.”

The ability to explain such alerting decisions is also making it easier for banks to explain their decision-making to auditors, he adds.