Like the double helix of DNA, innovation and economic productivity are complementary and inextricably linked. Crucial technological developments such as the steam engine and the internet preceded the Industrial Revolutions of the past, and the same might be said of the Fourth Industrial Revolution that is unravelling in the present.
Fundamentally, innovation brings new tools to bear for industries to enhance their productivity. In the latest expansion of the toolkit, artificial intelligence (AI) is becoming one of the most sought-after technologies by companies around the world. According to a Gartner report, the number of enterprises actively pursuing AI implementation grew by 270 percent in the last four years.
Despite the hype, Dr Pat Bajari, chief economist at Amazon and vice president of its Core AI division, noted that “only eight percent of organisations are really set up for success in deploying AI and machine learning,” citing the findings of a McKinsey survey.
He was speaking at the Singapore FinTech Festival and the Singapore Week of Innovation and TeCHnology (SFF x SWITCH), where he highlighted how AI can boost productivity in organisations and provided suggestions on how to effectively execute an AI strategy.
Keeping an AI on the prize
While most organisations manage to put systems in place to arrive at a desired level of output, oftentimes these systems are operating in a satisfactory but sub-optimal manner. “So when you look under the hood in most industries, you see big piles of waste,” Dr Bajari said, illustrating his point with the example of how 20 percent of the miles travelled by trucks in the US are without cargo.
Correcting such inefficiencies is a matter of changing the way decisions are made, and this is where data and AI become influential. By collecting data about the buying behaviours of consumers, for instance, Amazon has built an AI model that allows it to make better decisions about its inventory. “For each of those products, we need to determine how much of it to stock. That’s one of the most crucial decisions—if we get it wrong, we should just shut off the lights,” he said.
Getting to that AI model required a disciplined and systematic approach within Amazon, and Dr Bajari distilled this approach into five key steps:
1. Decide on a metric that you want to improve
“The first step you need to do is define what success means,” he said. In the context of inventory management, this could mean having items in stock to increase the probability of making a sale. But this needs to be tempered with the awareness that there is a cost associated with holding more inventory—cash gets tied up in addition to the opportunity cost of shelf space. “So the first thing you do is to articulate some notion of benefits minus costs to define a metric.”
2. Build a predictive model
Once the metric has been defined, the right types of data affecting the metric can be collected and used to create an AI model. Ideally, the model should be able to crunch the numbers and allow business owners to make predictions about a future state based on current circumstances.
3. Explore the relationship between model and metric
“Given your metric and your model, you then try to make a rational decision to improve that measure of success, given your model,” Dr Bajari said. This could mean reviewing which parameters of the model contribute significantly to a better ‘success score’, thereby helping organisations prioritise future actions.
4. Test the model
While the model may point out a theoretical path to success, it needs to be validated under real world conditions. “If you hang around statisticians, they’ll joke that all models are wrong, but some of them are useful,” Dr Bajari quipped. Hence, if a model does not produce tangible improvements in an organisation’s performance and productivity when deployed, then perhaps some data is missing, or the model needs to be adjusted to account for some previously unforeseen volatility.
5. Iterate and improve
Even after a model has been validated, Dr Bajari recommended that organisations return to step 1, if only to make sure that the original definition of success still stands. After all, changing business environments require adaptive strategies and models, not static ones. “If we can make [this five-step process] diffuse through industry, I think it’s going to improve material wellbeing of the world economy by replacing the forms of guesswork that exist in our decisions with science,” he concluded.
Jeremy Chan
Source: IMDA
