Breaking the barrier between Machine Learning (ML) prototype and production

Leverage MLOps to scale and realize the ML use cases faster

Most businesses in the ‘Connectedness’ industry have started embracing Machine Learning (ML) technology to provide effective customer service to the customers. However, managing these ML projects and putting them into action is challenging. For service providers who strive to move beyond ideation and embed ML into their business processes, Machine Learning Operations (MLOps) will be a game-changer. According to Gartner, “Launching ML pilots is deceptively easy but deploying them into production is notoriously challenging”. Listed below are a few challenges that make it hard to scale ML initiatives.

  • Lack of automated mechanism to address the change request in ML pipeline
  • Inefficient ways of retraining and deploying the ML models to accommodate the data changes
  • Lack of in-depth visibility of the model’s performance

Service providers need to implement the MLOps approach to overcome these challenges, which automates and monitors the entire machine learning life cycle. It enables consistent improvement in the baseline accuracy and accelerates the production time of ML models.

Launching ML pilots is deceptively easy but deploying them into production is notoriously challenging.

The successful implementation of the MLOps approach requires the right set of enablers such as de-coupled architecture, standard change management process, automated retraining and deployment of ML models, and continuous monitoring.

Operational Excellence

Fiber is fast, but rollout needs to keep up

AI/ML can forecast delays before they occur, making the service delivery predictable and fast

The global pandemic has highlighted the fact that high-speed broadband is a necessity, not a luxury. And fiber is one of the ways to faster broadband. This appetite for fiber means that service providers need to roll out fiber-based connectivity services faster. However, with the rising complexities in the order management process, delivering the service within the specified timeline is becoming a nightmare. The main business issue is unpredictability, which may be as important as speed. Its absence means frustration for service providers and their customers.

The main cause of this lack of predictability stems from the structure of the process. In many cases, the enterprise service delivery process has evolved and grown organically. The most common causes of dysfunction are:

  • Multiple teams operating in silos prevent a clear view of the process and a single source of truth
  • Manual hand-offs leading to errors and delays
  • Dependency on external vendors, resulting in vendors operational issues being transferred to the service provider
  • Lack of strategies to forecast order delays
  • Lack of mechanisms for real-time tracking of service delivery flow

To overcome these challenges and tap into the next wave of opportunities, service delivery operations will require an advanced vision. AI/ML is at the heart of that vision. With AI/ML in service delivery, enterprises can predict and address delays before they impact the business. Enterprise AI can, over time, improve the prediction of potential delays and delivery dates at all points of the order journey. Over time, enterprises can achieve faster processing of orders with improved predictions.


The appetite for high-speed broadband demands a faster rollout of fiber-based connectivity services.


To treat, or not to treat: Increase marketing ROI with targeted campaigns, through uplift modelling

While running direct marketing campaigns, businesses must map the right customers to a given promotional offer to maximize the campaign effect. For example, which customers should receive a discount on subscription, to minimize the business overall churn rate.

Different methods can be used to identify the right set of target customers for campaigns, such as, manual spreadsheet-based statistical modelling and outcome modelling. These methods, however, have some limitations like:

  • Randomized and inaccurate list of target customers
  • Lack of granular details such as which customers are most likely to respond to marketing campaigns
  • Low marketing ROI due to poor response rate from customers

Machine Learning (ML)-based uplift modelling is a promising approach to overcome the above limitations. It allows businesses to categorize customers as the ones who are likely to respond positively to a campaign and those who would remain neutral or even react negatively.


An uplift model increases marketing ROI by determining the right target customers.

A well-executed uplift model would improve a business marketing efficiency and help in driving higher incremental revenue. The successful implementation of the model requires the right set of enablers such as raw data acquisition, feature engineering, and AI/ML model development.

Operational Excellence

Accelerating Digital Transformation with Hyperautomation

Leverage the power of RPA, process mining and AI for end-to-end process automation to increase automation rate, reduce operational expenditures and improve customer experience

‘Hyperautomation’ is one of Gartner’s Top Strategic Technology Trends for 2022. Hyperautomation aims to identify, analyze, and automate business processes to the greatest extent possible. It involves orchestrating the use of multiple technologies, tools, and platforms to streamline business processes.

Legacy infrastructure and outdated processes can hinder an organization’s ability to compete. Automation of only task-based processes will not deliver the cross-functional results needed to drive business decisions and outcomes. By automating as many processes and tasks as possible, hyperautomation transforms an organization.

Increase connectivity, efficiency, and agility in business operations with hyperautomation.

As per Gartner, hyperautomation will lower operating costs by 30 percent or more by 2024, thereby increasing connectivity, efficiency, and agility of business operations. The businesses in the connectedness vertical can achieve end-to-end process automation and scale up the automation rate by building and implementing a hyperautomation framework that includes four key components:

  • Intelligent Process Orchestrator: Orchestrates bots, people, and IT applications for end-to-end integration of any business process.
  • Conversational AI: Automates all sub-processes that requires a conversation with humans. Conversational AI understands natural language and converses with the customer.
  • Low-code Applications: Helps to automate the sub-processes that require aggregating data from humans by building applications/interfaces rapidly.
  • Unified Hybrid Dashboard: Provides a real-time integrated view of the order completion process, resolution time, automation success rate, and many other KPIs. It also highlights the actionable insights.

Explainable Machine Learning (ML) models demystified

Enable 5X transparency in AIOps, achieving a more reliable and accurate business outcome

Service providers in the connectedness vertical embrace Artificial Intelligence for IT Operations (AIOps) to transform their businesses, but the users are hesitant in entrusting their operations to a complexly driven platform that provides no clarity and visibility into its functionality. Due to the lack of transparency, service providers are concerned about making bad decisions based on AI recommendations and the liability of such decisions and actions.

In their quest for autonomous operations, service providers seek to be more proactive with predictive analytics, where the machines make most of the decisions and help engineers take preemptive actions. However, the engineers need to have complete visibility into the underlying logic used by the AIOps and the ability to validate if the outcome is reliable.

Figure1: Assisted Artificial Intelligence and Machine Learning Framework

To accelerate AI/ML model development with enhanced transparency, enterprises must switch from existing auto-machine learning to assisted AI/ML framework-based solutions.

Explainable Machine Learning (ML) models aim to solve this problem by explaining the logic of the AIOps solutions so that the users can easily understand the outcome. The model explains the application of the AI solution and its result to the users in a way that they can clearly understand, rely on, and trust the outcome. Explanation in the ML model can be viewed as a means to transforming a black-box AIOps into a glass-box AIOps, by precisely lifting the veil on its computing and logic.

Digital Customer Experience

Plotting the future of customer care through an effective Virtual Agent (VA) rollout strategy

Improve the VA’s ability to engage with customers confidently and more accurately

The Virtual Agent’s (VA) market is at an all-time high and is garnering more and more interest with each passing day. It is beginning to establish as “the must-have” solution for the businesses in the connectedness industry, seeking to improve customer experience, reduce call center costs, optimize time to serve, etc.

But are these virtual agents living up to the hype?

Gartner has placed them in a “trough of disillusionment” in its hype cycle, meaning the technology is struggling to meet the envisioned expectations. When faced with complex and unknown scenarios, VAs tend to react in an unexpected way. One often comes across instances on social media where VAs are humiliated for their out-of-context interactions.

The primary reason for this shortcoming is that many VAs are launched without the right implementation strategy. As a result, they don’t reach the required confidence levels and cannot capture the right customer intent.

Virtual Assistants (VAs) use semantic and deep learning (such as Deep Neural Networks (DNNs), natural language processing, prediction models, recommendations, and personalization to assist people or automate tasks.

To prevent your VA from humiliation, adopt a robust VA implementation strategy encompassing the top 10 considerations that can help service providers to ensure their customers engage in the VA interaction, increasing overall customer satisfaction. This strategy provides key recommendations on the most important focus areas that are imperative for a successful rollout. Some of these include:

  • Choosing the right use case: Group the inbound calls into different categories like customer service enquiries, technical troubleshooting, sales etc. Based on these categories, different use cases can be invoked. For instance, kickoff the least complex rollout with self-service flows.
  • Analyzing the complexity of intents: Analyze the length of conversation and time taken by the agent to complete the conversation. Further, build a hierarchy of intents and sub-intents to identify high-volume intents and complex intents.
  • Considering variations in intent: Analyze the scope, lifecycle, and precursor of intents to improve engagement by increasing precision or recall.
Product Engineering

Use AI to Bolster your Network Capacity Planning decisions

The Content Delivery Network (CDN) market is poised to explode as content consumption gains more momentum. This calls for an efficiency-focused approach towards CDN capacity planning.

As per a Cisco report, the annual global IP traffic has already crossed the zettabyte (ZB) threshold. To cope with the increased content consumption by users, more supply chains should be established along with a reliable and scalable infrastructure. This puts a lot more pressure on the Content Delivery Networks (CDNs), which forms a well-established global backbone for content delivery.

For service providers, it becomes vital to take an efficiency-focused approach towards CDN capacity planning. This means satisfying the future capacity requirements without increasing the total cost of ownership.

The legacy manual way of capacity planning uses basic statistical tools to collect data and set a static threshold on capacity requirements. Such manual planning typically does not analyze the network in a holistic manner and produces a final proposal with a “one rule fits all” approach. However, this approach is inefficient in today’s scenario where consumer behavior changes very dynamically. Manual planning is also prone to human error, so the outcome might deviate from time to time, wasting a substantial number of resources and time. The service providers often run out of capacity due to increased data consumption and changes in the consumption patterns, which are not identified correctly during capacity planning.

To satisfy the customer demands in a timely fashion, it is necessary to have a modern capacity planning strategy.

Network planners need to confront these challenges before it impacts the customer experience. Leveraging Artificial Intelligence (AI) can significantly improve network capacity planning, thereby improving the end-user experience and reducing the total cost of ownership.

Software Intensive Networks

Predicting and preventing network problems leveraging AI

Implement a network event prediction model to improve service assurance

Today, service providers’ customers expect access to the products and services and enhanced customer experience anytime, anywhere. Hence, service providers should focus on service assurance and look for ways to address common problems such as accumulated faults, traffic congestion, and reactive event handling of networks. Further, reacting to a network event after it has occurred is not acceptable.

With the overwhelming volume and complexity of data from the service assurance domain, AI/ML techniques bring much value. By leveraging AI/ML in service assurance, service providers can analyze tons of data from various sources, derive insights, and take real-time preventive actions. Hence service providers should implement a network event prediction model to predict the network event failures and outages even before they occur.

By leveraging AI/ML in service assurance, service providers can analyze tons of data from various sources, derive insights, and take real-time preventive actions.