Did you know that Nagarro has five practices that are focused on different applications of AI (Artificial Intelligence)? Business Intelligence, Big Data, Machine Vision, Natural Languages, and Conversational AI make up Nagarro’s vast expertise in this growing field, which is our focus for the second edition of TNT (Think Nagarro Today) – a blog series where we put Nagarro’s experts in the spotlight.
Our colleagues from the Marketing team – Kerstin and Ionut – interviewed Nagarro’s AI experts Shallu Sarvari, Manaf Mahfouz, and Jan Noessner. They talked about what advice they would give companies who believe they are not ready for AI, commercial use cases for AI, and some stand out solutions they have implemented.
Shallu Sarvari from India, an architect at Nagarro. Her experience as a teaching assistant at the National Institute of Technology in Jalandhar has taught her a lot and shaped who she is today. She describes Nagarro as “an innovative, technology-oriented company with an open and inclusive mindset.”
Manaf Mahfouz from UAE, head of chatbot practice at Nagarro. Manaf has experienced first-hand a different kind of AI—artificial islands—having spent many years working in both the Netherlands and Dubai. Manaf enjoys the great variety of cultures present at Nagarro, as proven by his regular calls with colleagues from five different countries.
Jan Noessner from Austria, an AI & machine learning expert at Nagarro. Jan spent a few years working in Australia and studying in the US. If he had to choose, he would settle in Australia as his permanent home. Jan appreciates the non-hierarchical and entrepreneurial culture of Nagarro “because as soon as you have a cool idea, you can go for it.”
[Ionut Pop, Moderator]: Often businesses think that they are not ready for AI. What do you tell them?
[Jan Noessner]: AI is a huge area, so I'll just narrow it down to machine learning, as a starting point. What you need for machine learning is data. It operates on data and you need to have some sort of data ready to apply machine learning. But companies often think that acquiring this data and getting it ready takes a lot of work. But, actually, the companies have the data already. For example, if it's a production company, they have machines and these machines have sensors, they could use this data already for advanced analytics or for machine-learning algorithms to derive things like predictive maintenance. So, often when we speak to customers, data is already there, and they can very quickly utilize it. My main message is: People don't need to be afraid of machine learning because they feel they don't have the data ready. It is very easy to try, and, more often than not, the data is already there.
[Manaf Mahfouz]: I always suggest that they go for a PoC (Proof of Concept). On any AI solutions or intelligent solutions, the difference between normal software and intelligent software is that we need a pilot stage. Usually, when we deliver any software, we make sure that there are no bugs. In building these AI solutions, we need the pilot stage. That is essential. And in the pilot stage, the first day the machine will run, but will not have any good results. Maybe 50% or 75% good results, and the rest will fail. Then we go and do a visit. For example, I'm handling the chatbot and every week we do a visit to fix all the failed bugs. Now if you see the failed bugs chart you will see it's very healthy. The first time it runs it may have some starting problems then gradually it goes down and then it is healthy. We make sure in the pilot stage to have this healthy way and then we are ready to go to production. So, this is what I would advise them.
[Shallu Sarvari]: I totally agree with Manaf, this is exactly the process we follow. However, I would like to add one thing here: start with one business goal. Companies often think that AI is the product that we must achieve, that you have to implement AI somewhere. That is wrong thinking. You have business cases in which AI will help you deliver. Tech giants like Facebook or Apple are not AI products, but they have very strong business use cases assisted by AI.
Companies with high potential business values for AI are those with a lot of processes that are usually time-consuming and repetitive. For this scenario, we identify one problem that we want to solve and an impact area with a lot of repetitive work that we discover has a high business value, but the scope is minimized.
We start by finding the data, and ensuring its freshness. If not, we consult with the IT team, do a pilot phase to ingest the data from sources, like Manaf said. We work on understanding the context and deliver that first high-impact area by automating certain manual efforts or repetitive tasks.
[Jan]: Very good input! Of course, we do not have to use AI to solve every problem. The cool thing about AI is that you can solve problems with AI that you might not be able to solve with classic algorithms that existed before. But it is true that the first thing we search for is the business value that a certain problem can bring to the table and, based on that business value, we then choose the best technology. And one of those technologies is AI. Or machine learning, chatbots, big data.
[Kerstin Grüneis, Moderator]: What are the main commercial use cases for AI, and what are their business values?
[Manaf]: I am doing conversational AI and chatbots. To select one as the main business case, I would say, a chatbot as it helps people to search for what they are looking for. So, we built one nice accelerator, which helps people if they are on an e-commerce website, if we are on any industry like a service industry or a manufacturing industry. The chatbot will understand what the user is saying and then look at available products. For example, if you are looking for an iPhone on Amazon, you will get, maybe, more than 1,000 search results. Now, which is the best one for you? The solution we built helps us see all the return results of the query and then, based on the data returned, it can start a conversation asking: “Hey Mr. User! What kind of screen are you looking for? Are you looking for a 7- or a 10-inch screen?” Based on the reply we can narrow down the 1,000 search results into 50. Then we can narrow that down more by asking about the price or color preference. So, at the end the chatbot exchange, we will be able to identify, by going back and forth between the user and the catalog, the best product the user is looking for.
[Shallu]: The latest queries we had are in the sales and marketing areas. People ask to integrate AI with CRM software for lead generation, to find out which leads will get converted, or what is the sales forecasting for this year. Marketing mix is another important topic as a lot of companies are asking for the right channels to invest in. And, of course, customer 360 and customer experience is essential in retail or in any kind of area where we must use some advanced analytics to check the loyalty of the customers, or any kind of personalization that we can offer via AI. It could be, like Manaf said, an intelligent solution that could leverage just five results, those results that the potential customer is looking for.
[Jan]: One big area where AI can help is process automation. Processes that are very hard to automate with classic techniques, can be automated with AI. I can take a manufacturing example: you have certain parts where you have to adjust, where you must check if the path is manufactured correctly, at the end of the manufacturing line. With AI, you can install a camera and AI-based algorithms can easily check if that part is correct, or if there was a mistake, or if it needs to be sorted out. Another big area for AI is decision support, where you take data from the past. And you create some sort of decisions for forecasting the demand of certain products, or forecasting in general value than just using machine learning to get new insights from your past data. Potentially, predict the future or just make decisions for the current here and now.
[Ionut]: What are two AI applications that Nagarro has created that stand out to you and why?
[Manaf]: We have built many chatbots and one of them is Ginger, Nagarro's AI-driven virtual assistant. I think everybody in our company has used it.
[Jan]: I would go with the "used car dealer" use case. Our customer buys used cars and sells them again. The challenge is that they need a standardized process of buying these cars and evaluating how expensive it is to repair them before they can sell the car again. We built an AI-based solution, with very advanced machine learning techniques, that have image recognition classification. You can take pictures of the car from all angles and then the algorithm will first extract certain parts of the car. For example, the door. And from that door it will then detect if it has scratches or dents. And from that, it will also tell how expensive it would be to repair the dents and stretches. The AI algorithm will do an estimation to see if it's worth buying the car. Then, you will know that you need to buy the car for €10,000 and sell it for €15,000. It's essential to know what margin you can expect when you still have to repair the car.
Another use case that stands out to me is "customer opinion mining" for airlines. It uses advanced machine learning techniques and natural language processing. As an airline gets many customer reviews on social media, it must go through all reviews, order and answer them, especially if they are negative. Therefore, we built an AI-based solution to split each review into parts, and the parts are assigned to so-called "clusters." For example, if you say, "the breakfast was not very good," the solution will assign it to the category "meal," and, in the second algorithm, it will evaluate this sentence as positive or negative feedback. In this case, it would be negative feedback. The company can then analyze the feedback to know where they need to improve to increase customer satisfaction.
[Shallu]: The current application that I am working on is a multi-tenant SAP project for facilities knowledge management. In this project, we ingest a lot of data from a tenant in the form of videos, images, PDFs, structured, semi-structured, and unstructured data into workflows through many custom-made algorithms. And we are using cloud services, too. Our search portal is enabled to show the complete information (that has been extracted and mapped to the relevant assets at the back end) to the facilities management technician. So, it is a very relevant use case and includes many AI concepts. For example, we use image rendering. We extract the layout from the image to suggest which assets are placed at which part or which area of the building by ingesting the layout plans of the tenant.
Another interesting use case is the “fault detection” project for a European telecom partner. It was a great project where we were ingesting a lot of data from network devices in real-time and this data was then fed through many analytical processes. And, ultimately, fed through a predictive modeling technique, where it predicted if any kind of network device would fail or not, if there were any anomalies or patterns detected.It would tell you if you need to fix those and if they were directly affecting the customer's experience. So, it is a fascinating predictive maintenance case study for Nagarro.
Thanks to our artificial intelligence experts for providing such excellent insights to leverage AI and successfully embark on the journey to an AI-enabled future!
In our next TNT group discussion, we'll discuss another hot topic: Innovation. Stay tuned!
Machine Learning, Artificial Intelligence, Transform, AI and ML