According to an article in The Economic Times, nearly 90% of the world relies on algorithms daily — from ordering pizza or an Uber to finding a show on Netflix. We are living in a world where artificial intelligence (AI) and machine learning (ML) are all around us. And out of all the economic and technology changes and breakthroughs that are underway, AI may have the greatest economic impact on the world, according to Matt Wolodarsky at The Wealthy Owl, an investment blog.

Very few technologies throughout history have had this level of societal and economic impact. We’re talking about the internet, electricity, steam engine and printing press-level impact. In 2018, The McKinsey Global Institute forecasted that AI would deliver USD13 trillion of incremental economic activity by 2030. Their forecasts call for an additional 1.2% of additional gross domestic product growth (GDP) per year through 2030. Twice the growth, by comparison, to the steam engine’s productivity growth of 0.3% per year between 1850 and 1910, and information technology’s 0.6% boost in the 2000s. 

This projection is quite possible when you think about the applications that AI technologies, such as computer vision, natural language, virtual assistants, robotic process automation and machine learning, will have across industries. The opportunities can be classified into four main areas:

  • Augmenting how processes or tasks get done today
  • Automating repetitive manual labor tasks
  • Extending or innovating products and services via personal customization, as an example
  • Increasing the competitiveness of a company (or country) due to better AI and/or data

Pandemic forced transformation

The pandemic has fundamentally accelerated the process of digital transformation across industries. As a result, companies equipped with digital technology are more resilient and able to adapt faster. Due to the impact of COVID-19, International Data Corporation (IDC) forecasts that 65% of the world’s GDP would be digitized by 2022. What we were going to think about during 2030 is probably going to be true in 2025, Satya Nadella, Microsoft CEO, stated during a fireside chat with Flipkart Group CEO Kalyan Krishnamurthy. As a result, IDC forecasts direct investment in digital transformation will grow at a healthy compound annual growth rate (CAGR) of 15.5% globally between 2020 and 2023.

The time is now to start investing in AI. Certain industries have already been and will continue to be entirely disrupted by new or innovative incumbents that apply AI in transformative ways. If you need further proof, Apple just announced it is building a new campus and engineering hub in North Carolina. The move would create at least 3,000 jobs in ML, AI and software engineering.

Let’s take the example of a new insurance disruptor and startup, Lemonade. The company went public in 2020 and is upending the entire insurance industry by using AI algorithms to more precisely quantify risk when underwriting insurance policies. Their technology is leading to substantial profitability gains that allow them to easily undercut traditional insurance companies. And, as Lemonade collects more data to feed its algorithms, the traditional insurance providers are left in the dust.

Many corporations are exploring processes that would enable them to start testing and exploring the potential of AI to deliver improved business value (such as increasing fraud detection, providing better customer recommendations, etc.) or a proof of concept in the short term. Use cases tend to be narrow, and developers are typically leveraging exploratory data analysis (EDA) tools and ready-to-use AI and ML services for proofs of concept and prototyping. Two examples are using a prebuilt computer vision service to detect printed and handwritten text and using descriptive analytics to create a customer segmentation model.

Organizations are investing in the improvement of business processes and creating new value with machine learning. Companies can get started by exploring machine learning use cases implemented by other businesses.

It’s also important to assess the feasibility of machine learning use cases, to be able to identify the requirements to build, train and evaluate an ML model and to define data characteristics and biases that affect the quality of ML models. After identifying key considerations for managing ML projects, you can begin to create a custom ML use case that can meaningfully impact your business.

Learn how to start deploying AI applications that can unlock sustained value —hundreds of millions to billions of dollars per year — from reduced costs, increased revenue and higher margins. 

Identifying business value for using ML  

ML is a way to use standard algorithms to analyze data to derive predictive insights and make repeated decisions. When discussing business value with your executive team, answering the following questions will help you identify the value that machine learning will have on your business processes.

  • Benefit analysis: How would solving this problem improve or benefit the business, customers, and/or people in general?
  • Value-add: How would you classify the project: quick win, long-term development, or full transformation?
  • Resources and buy-in: Do we have any existing budget, expertise, and/or leadership support?

Who will be using and how will they be using data and processes facilitated by AI and ML? Assess ML use cases on two criteria:

  • Difficulty goals – should be challenging, but not impossible
  • Specificity goals – should be clear but not too specific  
Image Credit: Google Cloud Platform

It’s important for managers and executives to learn how to translate business problems into machine learning use cases and vet them for feasibility and impact.

Defining ML as a practice

The key to a successful ML model is lots of data. Big data cloud frameworks and technologies make it easier for businesses to adopt ML. Business professionals in non-technical roles have a unique opportunity to lead or influence machine learning projects. There are several types of machine learning problems. It’s important to learn how to differentiate between the most common ones; develop the key vocabulary to support yourself when working with ML experts; and to be able to identify the short- and long-term benefits when solving those ML problems.  

The greatest impact machine learning can have is by using data at scale. The Google Cloud’s AI Adoption Framework PDF cites an example of how an auction site reinvents its car valuation process with ML. A process that used to take 20 minutes, now using machine learning, is cut down to two-to-three minutes.

Where there’s data, there are ML opportunities. A financial institution can use ML to help evaluate: 

  • Is this credit card transaction fraudulent?
  • Should I offer this customer a loan or a savings account?
  • How much will this current customer deposit over the next 10 years?
  • Why is this customer calling the bank now?

Barriers to entry have fallen

  • Increasing power and availability of computing hardware and software
  • Increasing maturity and sophistication of ML algorithms
  • Increasing availability of data

Another example of ML solving a problem is how do we help to stop COVID-19 using ML?  

Image Credit: Google Cloud Platform

Building and evaluating ML models

To prepare data for ML you will need labeled data sets, or examples. The Google Cloud whitepaper uses the attributes of a leaf as an example.

Image Credit: Google Cloud Platform

Labels are the outcomes we look for.  

Image Credit: Google Cloud Platform

Label types can be either numbers or categories or phrases.    

Image Credit: Google Cloud Platform

Sometimes, labeled data is not as readily available it is suggested that you: 

  • Use labels from historical (joined) data
  • Use a proxy label
  • Build a labelling system
  • Use a labelling service

3 steps to formulating the ML problem

  1. Choose an objective
  2. Choose input features
  3. Get labels

The Google Cloud report states that “training the model is usually the easiest step in machine learning.” The best-performing models are continuously trained with new data. Always evaluate a model before using it in production. Google has a few best practices to consider:

  • Involve experimentation
  • Start simple
  • Don’t use your test data during experimentation
  • Test your ML projects with end users

Using ML responsibly and ethically

Google has a set of AI principles but has recently come under criticism after firing two AI ethics researchers. Timnit Gebru, a leader of the Ethical AI team at Google, said that she had been fired by the company in December 2020 after criticizing its approach (in a research paper) to minority hiring as well as its approach to bias in AI. The New York Times reported in February 2021 that another AI researcher, Margaret Mitchell, said she was let go after criticizing the way it has treated employees who were working on ways to address bias and toxicity in its AI systems. The Time’s article also highlighted a growing conflict in the tech industry over bias in AI, which is entwined with questions involving hiring from underrepresented communities.

Today’s AI systems can carry human biases because the researchers and engineers building these systems are often white men, and many worry that researchers are not giving this issue the attention it needs. According to an article by Josh Feast, published in the Harvard Business Review, many incidences of AI adopting gender bias from humans exist. The HBR article cites an example of natural language processing (NLP) that is present in Amazon’s Alexa and Apple’s Siri. An article in International Women’s Day about Addressing Gender Bias in AI, suggests that “AI companies need to attract more women in tech jobs, to diversify the pipeline and the workforce creating these new technologies.”

As a result of the growing need for women tech leaders, I launched a vetted private networking and peer learning group called Women Leaders in Data and AI (WLDA) for senior executive women leaders working in technology, data and AI to help mentor more women and help advance their journeys into C-suite technology positions.

Discovering ML use cases in day-to-day business

How does a company get started with finding use cases where machine learning can be used in their business? Every business has problems, and luckily, machine learning is another tool in your problem-solving toolkit. So, what kinds of problems are good candidates for machine learning? Typically, problems that require a series of decisions and have relevant data are best. After all, that’s what machine learning does — it automates decisions using relevant data. There will be different business use cases for machine learning depending on what industry your company is in. (Check out the industries and examples identified in a post on the blog Squadex.)  

 Image Credit: McKinsey

Managing ML projects successfully

Practical use cases of AI will define your business. You need the right expertise, data, tools, technology, partner ecosystem and economics to develop and operationalize your artificial intelligence. Every machine learning use case is different in design, architecture, tooling and optimization. To top it off, data availability, data quality, data storage, processing power and many other factors play a role.  

Key considerations for successfully managing ML projects include: 

  • Business Value – How would solving this problem improve or benefit the business, customers and/or people in general? How would you classify the project: quick win, long-term development, or full transformation? Do we have any existing budget, expertise, and/or leadership support?
  • Data Strategy – Are systems designed so that you will have more data next year? Can you break down the data silos? Have you made the transition from data lakes to data warehouses? You will want to run ML models on real-time data to extract the most value.
  • Governance – Data access must always be balanced against security. Train models on subsets or anonymized data. Sometimes, the most important information is also the most private. Privacy concerns should be addressed before data is shared with data science teams. The goals of ML and privacy are: identify sensitive data, protect it, create public governance documentation.
  • Team (expertise) – The most important data science roles are: data engineers, ML engineers, data analysts. When hiring, make sure the data scientists know your domain. Take advantage of learning platforms to up-skill your team (e.g. Coursera, Qwiklabs, or Pluralsight).
  • Culture – A few key principles: focus on the user, think 10x, launch and iterate.

AI and ML is ever-evolving. One thing this pandemic has taught all of us, is that digital transformation is a must.

Has your company discussed the use of AI to improve your processes, products or customer experience? What issues have you come across in your field?