How Can Companies Use Pervasive AI/ML Applications in Their Daily Business

*“Artificial Intelligence and Machine learning are looked upon by most people as esoteric and life changing. But are the technologies ready for normal business applications?” *questions Ramesh Narasimhan, Managing Director, Accion Labs at the Accion Labs Global Innovation Summit 2019. Accion Labs Global Innovation Summit is a flagship event organized by Accion every year in Goa for its prospective clients, existing customers, investors and Accionites.

In this session, Mr Narasimhan talks about how AI and ML are used day in and day out and yet most companies consider these technologies as beyond their abilities to implement them in their systems and processes.

He illustrates this observation by referring to a study that talks about, ‘How ready is your organization in ML?’ Based on the results of the study, a mere 4% companies say its being used broadly and it’s improving the outcome. While the other, 96%, say ‘it's being spoken in pockets, someone knows it, but nobody else knows it. Or this is too difficult, my business does not support it and similar vague responses.

What these companies fail to realize is that AI and ML are already involved in their daily life! Here are some examples:

1. You wake up with your phone every morning, you do a fingerprint scan or use facial recognition to open it.
2. Facebook recognizes faces, and identifies someone you met today or last week as your friend with the help of its recommendation engine.
3. Check how the recommendation system of Netflix works and surprisingly 85% of the movies we watch are recommended. People don’t search for them.
4. Take a look at Uber app; which is the nearest car, the time taken for the car to arrive, they’re all implemented by ML libraries, recently Uber also announced an open-source ML platform Ludwig, to make training and testing of the deep learning models easier for non-experts.
5. Last but not the least, Google which implements ML everywhere, in its search engine recommendation to Google Ads, Gmail features, translations, YouTube recommendation, etc.

Mr Narasimhan goes on to explain that these are some of the common use cases of AI and ML that apply to everyone, while most people relate the technologies to the more extraordinary applications like flying drones, driverless cars, quantum computing.

The underlying question is, “How can I apply machine learning and AI into my everyday business applications? What are those low hanging fruits and how can I think about AI ML in them?”

Six Accion Machine learning levers

We introduce the 6 Accion ML levers:

Classification and Scoring: Classification involves predicting the unknown Class/Label or Category for each individual in the population, to know which class does this individual will belong to. It will involve ranking or classifying your customers into various groups for example; which customer will not pay my bills? Which customer does not have the money?

Regression: Regression attempts to determine the strength and characteristic of the relationship between one dependent variable (a real numerical variable) and a series of other known variables (independent variables). Under Regression we can find relationships between multiple dimensions or variables that govern a particular process. For example,dynamically predicting the bid prices for a product or service based on the availability, demand & popularity.

Segmentation: You’ve got a thousand customers, so how do I segment them in a way to perform targeted campaigns or reach out to them in a specific way. For example, you have a prospective customer, look at the size of the customer, or the level of engagement you have, the potential growth that the company, the technologies they are using, how is the emerging tech adoption, their competitors and so on. There are multiple such factors and if you list them up, you may have fifty to hundred odd variables that you can use to create segments and use them in a way that it makes sense for you to pitch the customers in, to offer more relevant product or service.

Forecasting: This is all about forecasting and predicting what is going to be my next quarter, how will be my sales and revenue etc. Forecasting helps with prediction of future values merely based on past performance.

Recommendation: We earlier saw the Netflix example, where the recommendations suggest which product to offer to a particular customer, what kind of services, should I do a price discounting, should I upsell or cross sell, etc. Recommendations are based on finding behavioral patterns.

NLP: We have a number of NLP applications such as Chatbots, translation services, etc. NLP technologies have matured to a large extent, providing accurate analysis of not only text but of voice, and that too across multiple languages and accents. Using voice as an interface to applications has become extremely feasible and all applications need to leverage Conversational Interfaces.

Computer Vision: Computer vision includes methods for acquiring, processing, analyzing and understanding digital images & videos in order to perform some very specialised tasks including Object detection, change detection, facial and biometric based recognition, feature extraction, image classification and many more. For example, earlier we saw how Facebook uses facial recognition for their business.

These 6 Accion ML levers can be used to implement business solutions in most industries and use cases. Mr Narasimhan illustrates how these levers can be used by giving examples of use cases and applications.

Healthcare sector

The healthcare industry is an ideal industry for machine learning solutions to solve specific problems, such as identifying the risk of developing diseases like sepsis or diagnosing breast cancer. In addition, machine learning in medicine can also be used for disease identification and diagnosis of ailments.

Personalized medicine, or more effective treatment based on individual health data paired with predictive analytics, is also a hot research area and closely related to better disease assessment. ML and AI technologies are also being applied to monitoring and predicting epidemic outbreaks around the world.

Take a look at scenarios where healthcare focused applications can leverage AI and ML:

Digital consultation: Computer vision can be used to analyze videos, whereby the user takes a video or a picture of their pet, with automatic recognition signs of sickness tagged to the videos and sent to the doctor. The doctor can use the videos to fix appointments if required, or suggest an immediate remedy for the ailment. Another scenario could be there is something critical or serious, the user takes the video, uploads it, and the application could recognize possible symptoms of various standard ailments and to provide recommendations for care or seriousness, there by providing real time assessment of the ailment.

Medication management: An interesting use case for medication management is to provide elderly care to senior communities. Elders often face problems in reading or remembering their medication schedules. An application could provide guidance for reminders for scheduled medication, or even better, recognize when the elder fails to adhere to the schedule based on voice conversations.

Health Monitoring: Patients admitted in the ICU are monitored constantly for checking vital signs such as BP, pulse, oxygen saturation etc. With real time monitoring of such data, AI and ML can be used to recognize anomalies in the vital measurements, and can alert the staff when such anomalies occur. This can increase the effectiveness and capacity of ICU wards, and potentially extend such facilities to patients who are not recognized to be critical.

Media and Entertainment

In Media and Entertainment, there are many applications of Machine Learning and AI hiding in plain sight. We already spoke about Netflix in the earlier section, but there are many other ways where in ML can be used to turbo charge the user experience. When it comes to the future of AI in Media and Entertainment, experts say; it’s all about content, content, content. Recommender systems surface relevant content to readers, listeners, watchers. As creators develop new content, embedding based recommender systems can autosuggest image assets for articles, or help surface image assets without copyright limitations and so on.

Sampling of audio video: There can be a social videography project for a college sports event where you condense long hours of video to only showcase the highlights. Hence, you need not watch the whole video, just get the highlights to show the focal points.

Objects and scene recognition in video: the video feed from a CCTV can be monitored to measure how many hours an employee, a doctor, or other person sits on a seat. This is at the first level where you identify if someone is sitting or not. Then, depending on the quality of the video that is recorded, the application could also recognize the actual person in the video. That's next level. But in the first stage we determine what is your occupancy and how much is your office filled, what is the max mean usage of official assets etc.

Content search: In this you can develop a Shazam like app, where in you find out which song is playing? Every time you listen to songs and say ‘hey what song is that?’. And this is on the content search in entertainment.

Services industry

In the services industry there are many applications where AI ML is being used widely for example in cybersecurity, banking, education, telecom etc. Here are some examples:

Appointment Scheduling: You can develop a solution that tries to solve the problem, for example, you're driving your car, want to get a haircut, your application tells you, “hey in the speed that you’re driving, you’ll take about ten minutes. If you reach in ten minutes you have to wait for three minutes for a haircut. ‘okay I can’t wait for three minutes,’ but if you drive for five minutes on the left side, you'll find a barber shop where you need not even wait for more than a minute.” So if we think about appointment managing across services sectors, it’s everywhere. You've got people waiting at the hospital, could be in my gym, in tennis court, etc. That's an interesting way to think on how to handle or solve an appointment management problem.

Customer satisfaction: Under this you perform a sentimental analysis from the reviews provided by your employees on Glassdoor or by customers on social media sites. It could be positive or negative, and it's important that you understand the sentiments either to solve the problem for your customer or to target them other services that they are looking for.

Content validation: At times we have processes where in form filling is an essential step, for example you need to upload a passport copy, or upload a particular file and the system checks it, if it is a JPEG file or a GIF file or something, and you're done. That's your validation. Suddenly it comes back after two days and says by the way you put your driving license instead of your passport. Here comes our solution inplay where in you upload the passport and it immediately says ‘Hey dude this is not your passport’. It already sees it, validates it, and says this is not what I want. Go and change it. And it's done in a minute.

Data extraction and creation: Imagine a day when you say ‘Hey this is the form, here are my inputs, go fill it up.’ In today’s day and time nobody wants to fill in those large forms and they want to automate as much as possible. Think of the number of days that will be released from human labour in data entry. One because you're extracting the data using OCR and other techniques, filling in the form and coupling it with a chatbot kind of feature which will automatically fill up all the required fields in your form.

Employee engagement: You can predict an employee’s satisfaction in a company like Accion. What we're talking about is, take a few scenarios. Attendance, time to come to office, the JIRA tickets, performance in the daily stand up, in the retrospective story grooming, attending the various events, various festivals etc. If you just put all these things together and the participation of an employee, you can figure out if he's checked out or is in the system. Is he a performer or a non-performer? Was he a performer and now his performance has come down or not? That's the trend that the team is working on to figure out the employee engagement retention rate.

If you are interested to watch the whole session, check out this link. Consult our AI/ML Center of Excellence to help implement Artificial Intelligence or Machine learning in your day to day business processes.