“Marrying IoT data with traditional customer-generated data could lead to more exciting and unique services,” says Cathy Feng, Associate Vice President, ITC Forward Innovation & Data Analytics at Evalueserve
Cathy Feng leads Evalueserve’s AI practice, guiding the company into the next era of data analytics. Passionate about educating people on the possibilities of AI and its business applications, her efforts led to creating domain-specific AI applications for companies like McDonald’s, Intel, and Syngenta. Evalueserve’s AI platforms now have over 25k users. A Forrester Wave report recognised one of Evalueserve’s AI-based products, Insightsfirst.
Excerpts from the interview
How can IoT solutions become a differentiating factor for companies?
IoT solutions allow companies to get real-time data on the conditions of their assets/devices, the environment around them, locations, and other types of diversified data. With proper analytics, companies can take advantage of such data and gain a competitive advantage.
Condition monitoring, anomaly detection, predictive maintenance, and energy optimisation are well-known use cases. They could not only help the company optimise its operation, but data-backed insights could also help customers understand the performance of their assets, the lifecycle phases, optimisation, and what precautions to take.
Marrying the IoT data with traditional customer-generated data could lead to more exciting and unique services. In this way, the company can better engage with their customers, increase stickiness, and change the relationship from supplier/vendor to partner.
What key aspects should one keep in mind while building a future ML roadmap?
Different companies have different businesses. They may also be at various stages of AI adoption.
Stage 1 is the very initial stage, in which a company starts to explore AI. It is pretty standard in Stage 1 for people to have very high expectations and even fantasies that AI can do anything.
Later, in Stage 2, as a company gains more knowledge about the field, running through more POCs – some of which can be successful, while others could fail – people come to understand the realities of AI.
There could be some post-Stage 3 companies that conclude AI is not their thing and give up. Others learn the capabilities and drawbacks of AI through trial and error and understand the actual challenge of putting it into real practice. Based on their experiences, these companies could find balance and identify areas where the technology can bring value to the business, grow an ecosystem, and scale-up.
Very few in the race make it to Stage 4 – successful adoption – and have an ecosystem that can enable self-evolution to maximise the impact and value of AI.
People should keep in mind a few aspects when building their AI roadmap:
- AI has reached a stage where it can certainly help many areas, but not all. Be realistic, manage expectations, and focus on bringing measurable impact.
- Start small and get the low-hanging fruit first. For a successful implementation of AI, you need support and buy-in from stakeholders across the organisation. Having your own success story is essential.
- Leverage external partner expertise, especially in the initial stage. They are the ones who have achieved AI adaption and learned the lessons (sometimes the hard way). The experience and expertise they have can help hasten your learning curve and speed up the process.
- Your roadmap should keep evolving, especially in this fast-changing domain. It will be difficult if you decide on a three-year roadmap, start execution, and your plan remains inflexible. Keep an open mind.
Can you elaborate on the tech stacks used by Evalueserve?
At Evalueserve, we have more cloud-based products (AIRA) and those that are more on edge ( AiLENZ).
AIRA stands for AI for research and analytics, a one-stop platform providing access to Evalueserve’s AI-related use cases/models/algorithms, covering Natural Language Processing (NLP), Machine Learning (ML), and Deep Learning (DL). It sits on Azure AKS, with all services/use cases containerised and deployed. These modules are then exposed via API Gateway to allow easy integration with other Evalueserve digital products for AI enablement. In this way, we ensure the platform’s high availability, scalability, and security.
The primary language we use is Python when it comes to model development. We are open to using open-source libraries such as TensorFlow, PyTorch, Keras, and Spacy. Still, we are developing our own software development kits (SDKs) and domain-specific models to fit our business requirements better.
Some of our products, such as AiLENZ, a computer vision-centric product series, are also designed for the edge. These products focus on helping our customers in traditional industries speed up their digital transformations. A model optimiser and inference engine, such as OpenVINO, is used to improve inference speed at the edge.
How do your use cases help your clients with decision-making?
Research and analytics have been one of our company’s primary services in various sectors over the past 20 years. All our projects provide insights to our customers and support their decision-making strategies. This is in our company’s DNA. The knowledge accumulated over the years on what matters to our customers and what doesn’t is the asset that allows us to design our AI solutions/use cases better.
For example, in Data Analytics, we have use cases like demand forecasting, customer segmentation and analytics, and opportunity identification.
With proper demand forecasting, customers can make better decisions and plan their assets and resources well, thus improving operating income and customer experience and reducing costs. With customer segmentation, churn, and LTV prediction, promotions, discounts, retention, and upsell and cross-sell campaign decisions can be made and offered. With opportunity identification to estimate market potential and product affinity, people can identify upselling and cross-selling opportunities. And there could be many more.
How are you helping companies gain competitive and market intelligence?
We support multiple platforms with our AI engines to enable more efficient processing, digesting, and analysing incoming data, be it text, speech, image, or videos. One of these platforms is Insightsfirst, and its core modules are Market Intelligence, Competitive Intelligence, and Opportunity Radar (ORAD). So, via this architecture, i.e., AI engine layer plus integration and consumption layer, we help customers gain insights more effectively.
How vital is algorithmic fairness in today’s AI?
With more and more adoption of AI in various systems, especially those involving automated decision-making, algorithmic fairness becomes essential. People use AI for resume screening in the hiring process, monitoring employee sentiment, deciding on marketing approaches for targeted groups, checking insurance eligibility, and government services in public sectors. In all these examples, the biases brought in via models themselves or input data could lead to discriminatory practices. No one wants that. It also compromises the initial intention of applying AI in these fields.
Which data science domain will come out on top in the next ten years?
Data science has immense scope. If I talk about NLP and computer vision – the areas that I’m more focused on and the topics that matter for applying this technology in the real world, here are a few:
- Bridging the gap between lab experiments and actual industrial implementation, including the landing of large language models.
- Model training based on smaller datasets to lower the bar of actual business application.
- Low-code, no-code ML/DL engine allows more people to leverage the technology.
- Advancement in MLOps makes model deployment much more effortless and standardises maintenance across life cycles.
