Story image

Why businesses need to start implementing AI

29 Apr 2019
Twitter
Facebook

Article by DataRobot APAC GM Tim Young

By next year, AI-driven companies will be taking US$1.2 trillion of business revenue away from their rivals. They will be, as the saying goes, eating their competitors’ lunch.  

Losing out to rivals is usually enough motivation for most companies to change strategy. Competition, however, is not just about P&Ls – there are far-reaching ramifications to not keeping pace with the market.

Many organisations are grappling with this right now as they consider the rationale for adopting artificial intelligence against the investment required and other resources needed to make the change. 

AI, specifically machine learning, is already being widely embraced by businesses and organisations across the spectrum. It helps companies produce better products and services using a continuous feedback loop from customer behaviour and data to help machines learn at a rate faster than ever before.

This helps them bring upgrades and new products to market at vastly improved speeds compared to traditional methods. First-mover advantage has significant revenue implications in competitive industries.

Machine learning can also improve internal business processes, such as CRM or credit rating assessments. For example, banks are using AI to evaluate the creditworthiness of potential customers, using machines that have learned from the data about the behaviour of thousands of other customers and transactions.

This helps lenders make quick loan decisions that are supported by a data review that no human credit officer could ever match.

AI is all about data, increasingly the most valuable asset in any company, large or small. On its own, however, data has little utility unless an organisation knows the best way to extract its benefits.

Optimising data insights hinges on two key factors: quality, well-structured data and robust cradle-to-grave processes - from recording and collecting data to its ultimate disposal - that are aligned with best practice. The good news is that all companies have been collecting data for the last 50 years. Now they can put it to greater use!

Artificial intelligence has been a game changer in tapping data’s riches, but it comes at a cost. It takes time and money to code AI and build predictive and analytical models. In a world where time-to-market has never been more important for companies fighting for market share, speed in delivering an insights-driven product or service to consumers is self-evident. 

This is where automated machine learning comes in. Traditional AI requires time-consuming, manual processes to build analytical models.

 It involves multiple tasks – from feature selection and engineering, trawling through thousands of possible algorithms, to the eventual evaluation and comparison of results. For many companies, most of these activities are carried out by data scientists who often have more than they can handle on their plate -- assuming companies can find, attract and retain qualified data scientists in the first place!

Automated machine learning lets companies replace this slow and arduous human activity with a rapid, automated process that scales the delivery of AI.

By automating time-consuming and repetitive tasks, (citizen) data scientists spend far more time-solving problems and delivering real value back to the business. This approach also improves job satisfaction and reduces churn.

Automated machine learning users have reported that tasks that once took several months have been reduced to days and in some cases, just hours.  The resulting advantages are blindingly obvious.  

In traditional enterprise AI models, businesses invest in data teams staffed with skilled analysts and engineers.  This capability doesn’t come cheap: data scientists are in hot demand and the good ones can write their own cheques. You also can’t get by with just one. Typically, a team is required to carry out the labyrinthine tasks in building AI.

If a team is among the best in the business, they’ll be poached with offers of fantastical salaries. The company that can’t afford to keep them is then faced with two problems: finding a new team and losing the IP amassed under their roof. Neither are replaced easily.

This leads to my next point. Once a business accepts that it’s going to embed AI across its business, it is faced with a choice of either hiring the skills (we’ve already established the risks here) or adopting creative ways to build their internal AI capabilities organically. Traditional AI has been the domain of very highly skilled individuals.

Not only are they expensive to find and employ, but they can sometimes operate in silos, using jargon, calculations and coding languages that are confusing to the average business person.

The Harvard Business Review recently pointed out that this kind of specialisation can increase costs due to the need for coordination.

“Coordination costs act as a tax on iteration, making it more difficult and expensive, and more likely to dissuade exploration. That can hamper learning…Data science roles need to be made more general, with broad responsibilities agnostic to technical function.”

With new automated machine learning tools, building AI capability across a company’s various business units is now a reality. Automating some of the more complex processes, such as identifying the right algorithms to build a model AI, makes it easier for employees who don’t necessarily have a technical background to become involved. 

This allows the smarter analysts you have in place to work as data scientists and, thereby, create a new class of user -- a trend Gartner refers to as “citizen data scientists.” Tools like DataRobot put the potential of artificial intelligence in the hands of any business user, empowering teams in every division to build and deploy highly accurate models in a fraction of the time of traditional methods.

Built-in guardrails also ensure that users don’t make foolish mistakes or introduce bias into their models. For example, detecting target leakage -- training your algorithm on a dataset that includes information that would not be available at the time of prediction and so distort actual world predictions.

A final argument for why businesses need to adopt AI is related to transparency. The recent Australian Royal Commission into the financial services industry has driven home the need for greater transparency when dealing with customers. Financial institutions, like all highly regulated customer-facing industries, must comply with strict data regulations.

Automated machine learning includes model and feature explanation, which offers transparency around how data insights are derived. This cost and time-effective function will become increasingly critical as organisations face ever-tougher regulatory requirements.