Machine Learning and Artificial Intelligence are changing the world of business as we know it. But for many business owners and their teams, the immediate connection between ML & AI for their businesses is not always clear. The business problems are often not realized and AI and ML solutions are not explored by executives and their teams. This article will aim to shed light on 12 use cases and demystify the use of ML and AI to solve real problems in the world.
Machine Learning and AI help create better products and services by:
- optimizing assets and services
- improving quality and reliability
- preventing the downtime of product maintenance
Used for existing products and services, ML & AI are helping executives and teams to focus on what matters most.
Cultural, educational, gender, race or other biases seeping into ML & AI is no secret. Transparency in AI is essential to building trust in these systems. On LinkedIn, high-paying jobs were not displayed as frequently for women as they were for men. The biases stem from the way the algorithms were trained. It is important to be aware of your training data because biased data will lead to biased ML models. We need to ensure there are experts in the loop that can comb through and analyze for bias within the data. These experts can see if there are enough examples of specific attributes within their model. Techniques and standards are also growing to reduce bias. Also, the algorithm itself can be altered by tweaking the optimization model to help the computer understand the importance of accuracy in certain cases such as choosing a tradeoff between avoiding Type 1 (false positive) or Type 2 (false negative) Errors.
Now, let’s explore some use cases of ML & AI in the real world:
1. Forecasting Inventory
Overstocked or understocked items are a classic inventory problem for retailers. For e-commerce companies, returned items are another problem that costs companies a significant amount of money.
Machine Learning looks at existing data and manages stock. It can see trends and patterns and prompt the business to take action. This could mean ordering items that are trending or forecasting demand for accurate quantities of specific items. Using ML, forecasting errors become decreased by 20-50%. ML can reduce lost sales up by to 65% and inventory reductions of 20-50% are achievable.
An interesting example of a business using ML & AI for forecasting inventory is major German e-commerce company, Otto. Returns cost the e-commerce company millions and their data analysis shows that customers tend not to return items if they are shipped within two days. Customers also dislike multiple and slow shipments. The problem becomes a bit more complicated given that Otto sells products from other brands and they don’t do stocking themselves. Otto is either waiting to ship until all orders are ready or shipping many boxes at different times.
Using a deep-learning algorithm originally designed for particle-physics experiments by CERN in Geneva, Otto uses ML to forecast their sales in the next 30 days with 90% accuracy! The algorithm analyzes about three billion past transactions and 200 variables, including site searches, past sales, and weather information to know what consumers will buy. With this algorithm, Otto can buy the products ahead of time so they can be ready to be shipped once the customer makes the purchase. Otto builds an inventory based on what their AI forecasts, ordering over 200,000 items a month with no human intervention. Contrary to popular belief of firing human employees, Otto did not fire anyone but they did hire more employees!
2. Electric Utilities
Machine Learning, applied to electrical utilities, is another case of matching demand and supply. Adjusting supply to meet anticipated demand in real-time leads to huge savings. Algorithms gather insight from weather-related variables and act upon those insights. It can switch off air conditioning when forecasts tend toward peak consumption. This can avoid the need to operate at peak generating capacity altogether.
3. Research and Development
Projecting can help with product research and development. Great efforts are spent on creating products and iterating them based on whether they will fail or succeed in the market. The speed of the design process can be increased with AI. Motivo, an AI startup, can complete chipset design processes that would normally take months in just four weeks. They’re saving companies time on testing and iterating designs.
There are many uses cases of farmers using AI and ML to improve productivity and increase profitability. Farmers can reap the benefits of projection based on demand and supply. They can use ML to expect crop yields and project the demand for certain crops to help optimize plant growth. The use of historical and real-time weather data helps farmers expect the number of crops that will be yielded, saving them time and money when anticipating crop yields.
There many other use cases of farmers using AI and ML to address various problems such as shortage of labour, weeds, and healthy livestock. About two-thirds of the strawberry industry in the US has invested in a robot that uses image processing to find the right berries to pick at the right time. Farming equipment giant John Deere has also been investing into AI and ML over the last decade. One of the technologies uses a “see and spray” technique to map fields and visual recognition software to target weeds and apply herbicides only where needed. It saves the farmers money and produces more crops with less herbicide.
Another great use case is by a dairy farm in Georgia that uses an app to detect their cows’ health. Ida, created by Dutch company Connecterra, uses TensorFlow, to analyze the data collected. The cows wear a collar that detects how much the cow is eating, drinking, ruminating, sleeping and walking. With that data, Ida uses TensorFlow to let the farmer know the general health of their cows. On a hot day, this can speed up the process of the farmer and their team checking their stock of 2,000 cows every day for overheating.
Teachers spend a great part of their time marking, going through hundreds of papers and outsourcing their work to assistants. But marking between teachers and teacher’s assistants (TAs) can greatly vary. Some TAs mark more strictly than others. Teachers themselves may not put the same amount of effort in marking the 99th paper as they did in the first paper.
Thanks to ML, we can remove much of this variance. Machine Learning algorithms can view and understand handwriting. Teachers can provide the algorithm with a few examples, then the algorithm learns how the teacher marks and applies their marking guidelines to other papers. Algorithms never tire and hold the same standard throughout the marking process. Algorithms like these have proven to be 85% accurate, giving the same marks as the teacher would. The algorithm can prove to exceptional in clear-cut answers such as Math and even more subjective courses like English. Using ML to mark work reduces teacher burnout and gives them more energy and time to focus on teaching.
6. Medicine & Healthcare
Diagnosing ailments and identifying diseases is at the center of ML & AI in medicine and healthcare. Patients continue to receive diagnoses from images like x-rays. Thanks to advances in computer vision, it is becoming more possible for AI to take on a part of the role of radiologists. The deep learning algorithms are fed thousands of images and diagnoses to improve themselves and can examine images faster than a human. Enlitic, a startup based in San Francisco, sent software engineers to approximately 80 medical imaging centers in Australia and Asia to use a deep learning algorithm designed for use on PACS (Picture Archive Communication Systems for digitalized radiological images and reports). The company hopes the algorithm will eventually become smart enough to identify diseases in every imaging modality in the centres, including magnetic resonance (MR), computed tomography (CT), ultrasound, X-ray and nuclear medicine.
It’s common to go to a doctor’s office with a cough and leave with a prescription for cough medicine. Machine Learning algorithms can consider outside-the-box factors such as eating, sleep, and exercise habits from the data we have available from smart devices (like a Fitbit) to modify prescriptions for sick people. Instead of broad pharmaceuticals targeted to people with colds, there will be an increasing amount of highly-specific prescriptions tailored towards individuals.
7. Call Centers
Hiring and training staff takes time and money. For AI-powered automation customer service, you only have to train the model once. Call centers can also use ML to track the sentiment of current callers and intervene when necessary. Companies can use ML models to create chatbots to help with FAQs and train them to learn from and answer new inquiries. The model could even be trained in knowing when to switch over to a human agent. These models can also be used to train telemarketers to identify situations where clients become upset and give tips to alleviate the situation. This is all possible thanks to advances to Natural Language Processing.
AI-powered automation for customer service is predicted to be the norm by 2020. Autodesk, an American software company, uses IBM’s Watson to take on repetitive non-value calls, provide 24/7/365 support, and provide better service at a lower cost. Using Watson, Autodesk built Ava, a virtual agent designed to resolve the most common support issues. With Ava, they have brought down resolution time from 38 hours to a mere 5 minutes!
Machine Learning, applied properly, can significantly affect the bottom line. Many large financial institutions are diving into ML to find solutions to their common problems such as fraud prevention, risk management, investment predictions, network security, and much more.
For example, ML has become imperative in creating Credit Risk Models. ML models large volumes of data to provide a granular view and reveal hidden trends of customers. This creates a more robust and flexible credit score for the individual in less time. With the data provided, the model can better understand how likely someone is to pay back a loan. Based off millions of loan-paying cases, this ML model has the capacity to process this information in a way that humans cannot do.
9. Mobile Use
It is undeniable how much our smartphones have penetrated our lives. Adobe research shows that people spend the same amount of time sleeping as they do on their smartphones. With AI, brands can finally get one step ahead of the consumer with personalization, predictive analytics, and content consumption.
Warby Parker, an eyeglass maker, has created an iPhone app using Apple’s face-mapping technology to measure users’ faces and make recommendations for the best glasses for every user’s face shape and features. Starbucks’ mobile app uses existing user information, like the type of coffee they drink or the time of day of their usual visit, to create offers, coupons, and discounts.
10. Social Media
Social media channels such as Twitter and Facebook shape our views on many different topics. Besides showing us viral tweets and posts, these companies are using AI a lot deeper to weed out the humans vs the fake bots. Cybercriminals can use various tricks when opening many false accounts to trick the social network companies.
But, the cybercriminals cannot fake human movement. For example, Facebook can use your phone to measure subtle movements from breathing, the angle you are holding your phone, to how quickly you tap on the screen.
Fact checking is another big problem for Facebook. Misinformation can spread quickly from images and memes with the social algorithms. The social media giant has deployed fact-checking program utilizing third-parties including AI and humans. These tools including using the AI of AdVerif.ai which finds flagged images and conducts reverse image lookup to see where else it’s been posted, and if it’s been tampered to show something different. For example, Facebook caught a false image of an NFL player burning the American flag. Facebook then uses this data as training data for its machine learning classifiers.
Recently, Twitter is using AI to take action on false accounts that cause chaos in public trust and politics. Even Google is using AI to track down harassing trolls on Youtube. AI will continue to be at the heart of social media networks even as the times and needs change.
11. Travel and Leisure
Travel agencies use ML to understand the types of travellers on their websites. With the small user interactions on a website, companies can determine whether they are a business or leisure traveller, if they’re picky about their meals, or whether they favour certain hotels. This data helps these agencies make recommendations to travellers like where to travel and which airlines to take.
For example, KLM is using an AI system that is dealing with 50% of its inquiries. Dorchester Collections, a luxury hotel operator, changed its breakfast menu after AI analyzed guest reviews and came up with customizing options. Lola, an iPhone travel app, uses AI with humans to provide help for hotel bookings, flight schedules, and restaurant advice.
12. Recruiting Agencies
Matching employers to employees is complex. There is often a large turnover rate that’s a strain on human and financial resources. There is a big challenge for hiring managers to find the right candidate for the job. Let’s consider the role of a web developer. Some employers place more weight on academics and fail to identify important skills. Candidates with a Masters degree in Computer Science pass through the screening process. But the fact that they lack important job skills goes unnoticed.
At times, ML can better understand what the employers need than they can themselves. Employers can misread variables. They can confuse priorities and relevance due to personal bias. They can underestimate the value of senior developers teaching junior developers. A personalized matching service learned from previous matches can help companies increase employee and employer satisfaction.
Machine Learning is the next step we need to make sense of all the data that we’ve all been collecting, and continue to amass. It is here, now, and will continue to change the way businesses will operate forever.
“Machine Learning. This is the next transformation…the programming paradigm is changing. Instead of programming a computer, you teach a computer to learn something and it does what you want.”
- Eric Schmidt, Executive Chairman of the Board, Google