Radhakrishnan Annamalai

We can follow the below-structured approach to design a Proof of Concept (PoC) to test the feasibility and implementation of an AI-driven chatbot to improve customer service Set a clear objective and metrics Assess the technical feasibility of integrating AI bots with the existing customer systems Identify the areas to improve customer service, like decreased response time and increased resolution rates. Cost Savings and Return on Investment (ROI) Scope and Technology Selection Identify the customer service functions that chatbots can support like FAQs, appointment scheduling, etc. Select a chatbot platform, preferably a cloud platform to design, build, and test Integrate with customer support systems like Salesforce, Zendesk, etc. Methodology Design a basic conversation flow, intents, and responses. Build a chatbot using the selected platform with the existing customer system Test the chatbots with a small group of customers Evaluate the performance and accuracy of chatbots Refine the chatbot conversion flow and intents based on the evaluation results Primary Risks and Mitigation Strategies Poor Chatbot Performance Train the chatbot continuously with accurate data and real-world customer queries Implement fallback mechanisms if required Customer Resistance to Adopt New Technology Notify customers that they are interacting with AI Provide an alternate option like “Talk to Human. Encourage customers to use the new technology by offering decreased waiting time and faster resolution Customer Data Security Risk Ensure data encryption and compliance with security and privacy laws Frequent audits on Chatbot logs Integration Issues Choose and integrate the flexible, API-driven chatbot platform Conduct Integration testing before deployment Resistance from the Customer Service Team Educate that AI Chatbot is a tool to assist, not replace, humans Train employees how to work concurrently with AI so that repetitive tasks will be reduced, and agents can focus on complex queries PoC Deliverables: Chatbot Performance Report (Accuracy, Resolution Rate) Customer Feedback (CSAT) & Adoption Insights ROI & Cost-Benefit Analysis Scalability Assessment Risk Assessment Recommendation for Full Deployment or Modification This structured PoC approach will ensure that the AI-driven chatbot aligns with business excellence goals while mitigating risks for a smooth transition.

Break Even Analysis (BEA) is a financial measure that calculates the minimum return or sales volumes required to cover the product or service costs. A Lean Six Sigma project that focuses on a support process will not have direct sales or revenue generation. In such cases, BEA can be applied using alternative metrics that reflect the service value or process output. Here is an example of how to apply BEA in a Lean Six Sigma project for a support process: Break-Even Analysis for a QA Support Process in the Medical Coding Industry. The break-even analysis for a medical coding QA process helps to find the point at which the support activities bring in the same amount of money as they cost. This means that the company doesn't have to pay the client a penalty for a claim error. Key Components of Break-Even Analysis: Fixed Costs: Annual Salary of QA Staff ($50000) Variable Costs: Software Licenses, Membership Renewals, and Other Expenses ($1/chart audited) Revenue: Savings from error reduction or avoiding paying a penalty for poor quality from client audits ($3/client error findings) Output Metric: Avoid paying a penalty for claim errors from the client audits. Break Even Formula: Fixed Costs/Revenue per Unit - Variable Cost per Unit 50000/(3-1) = 50000/2 = 25000 Conclusion: The company needs to avoid paying penalties to clients for at least 25,000 claims per year to cover QA support process costs.

ISO 31000: It is an industry-standard risk management principle, framework, and process that provides guidance for implementing risk management practices in any organization. FMEA: Failure Mode and Effect Analysis (FMEA) is a popular tool used to identify and prioritize potential failures in processes, products, or services. Both ISO 31000 and FMEA can be integrated so that we can use the ISO 31000 comprehensive risk identification process to identify potential failure modes within the product or process and then leverage it utilizing the FMEA framework to analyze the severity (S), occurrence (O), and detection (D) for each potential failure mode to calculate the risk priority number (RPN) in order to develop prioritized risk mitigation strategies in the early stages of development so that it will be cost-effective. Integration of ISO 31000 and FMEA provides various synergies, including: Detailed Risk Identification: ISO 31000 identifies all relevant risks, whereas FMEA’s structured approach helps to identify potential risks with prioritization. Detailed Risk Analysis: ISO 31000 evaluates the impact of each risk, whereas FMEA’s SOD ratings provide a detailed understanding of each risk. Improvement on Risk Prioritization: ISO 31000 prioritizes risk based on their potential impact, whereas FMEA’s RPN calculation provides a quantitative measure of each risk’s priority. Effective Risk Management: The ISO 31000 risk management framework enables effective risk management by developing and implementing controls, while FMEA focuses on failure mode and effects to address root causes of possible failures. Example: Company ABC wants to assess the risk of combining various physiotherapy devices (IFT, ultrasound, TENS, etc.) into a single product. Utilizing the FMEA tool, they can identify the potential failure modes with SOD ratings. Failure Mode Effects S O D RPN Device Electrical Malfunction Harm to Patient 9 3 2 54 Incorrect Therapy Dosage Ineffective Treatment 8 2 2 32 Device Software Issue Unable to Operate 8 3 1 24 Device Component Malfunction Harm to Patient 6 3 1 18 Using ISO 31000, Company ABC can assess and prioritize these risks based on their impact. In the preceding scenario, they will decide that the chance of device electrical malfunction is larger, necessitating greater care. If we integrate both ISO 31000 and FMEA, we can leverage the strengths of both frameworks to improve the risk prioritization and mitigation, which will enhance the overall risk management capabilities.

To address the collaboration and coordination challenges due to unclear priorities, delayed feedback, and miscommunication that lead to missed milestones, we can use the below KPI to measure the collaboration and unclear priorities. We must first set the priority task, and we can measure the priority clarity using the Priority Clarity Index. It measures the percentage of correctly identified top 3 priorities by the team members. The target set is usually 80% or above. PCI = (Number of team members who correctly identify top 3 priorities/Total number of team members) * 100 We can measure the collaboration’s effectiveness using the Collaboration Effectiveness Index KPI. Collaboration Effectiveness Index (CEI) – (On-Time Milestone Achievement x Feedback Loop Efficiency)/Miscommunication Rate On-Time Milestone Achievement Rate: It measures the percentage of milestones achieved on or before the scheduled deadline. Feedback Loop Efficiency: It measures the average time to receive and incorporate feedback from the customers or stakeholders. Miscommunication Rate: it measures the frequency of miscommunications caused by inadequate communication. KPI Values: Below are the KPI values that we can set to improve the collaboration of the cross-functional team. On-Time Milestone Achievement Rate: 90% Feedback Look Efficiency: 80% of feedback to be incorporated within 1 business working day Miscommunication Rate: Less than 5% of total communications. Actionable Insights: Tracking the CEI can provide the following actionable insights: Identify bottlenecks in the communication process Identify areas of miscommunication to establish clear communication channels Assess current collaboration method and implement new processes or tools to improve communication and feedback loops Evaluate and prioritize milestones that require more attention or resources Conduct R&R to reward the team members who communicate and collaborate effectively to encourage other team members to do the same.

“Thinking, Fast, and Slow” was introduced by psychologist Danile Kahneman. It highlights System 1 and System 2 thinking. System 1 is fast, instinctive, and emotional; it is quick and automatic thinking, and the decision-making is influenced based on historical experiences and cognitive biases. Example: Playing a “blitz” chess game requires a fast and automatic way of thinking, which does not require conscious thought. System 2 is slow, deliberate, and logical; it is conscious and controlled thinking, and the decision making is influenced based on analytical and logical thinking to be more accurate. Example: Playing a “classical” chess game requires concentration and attention. Business leaders usually adopt System 1 thinking during high-pressure situations, whereas they adopt System 2 thinking for long-term strategic planning. To balance System 1 and System 2 thinking, the business leaders should adapt the following strategies: Recognize and act based on the type of situation: Adjust the thinking style based on the situation, whether it demands quick action or requires strategic planning. Brainstorming sessions: Encourage diverse viewpoints to avoid the influence of cognitive biases. Develop a decision-making framework: Develop a decision-making process to ensure that both System 1 and System 2 thinking are utilized appropriately based on the situations. Slow down for critical decision-making: Take time to engage system 2 thinking while making critical decision-making. Business leaders need to adjust their strategies based on various circumstances or situations by knowing the advantages and disadvantages of both System 1 and System 2.

Polanyi's paradox emphasizes the need for tacit knowledge, which is the human experience and intuitive side. There are numerous ways that this understanding can be leveraged. Emphasize human skills, including creativity, critical thinking, and sophisticated problem-solving. Combine artificial intelligence capabilities with human tacit knowledge to generate a hybrid intelligence system. Transfer tacit knowledge among humans through coaching and mentoring. Use AI to support human decision-making along with data-driven insights. Below are the areas where human skills will remain indispensable: Solving the complex problem Creativity and Innovation Emotional Intelligence and Compassion Decision-making that involves moral ambiguity Below are some of the strategies that individuals and organizations can adapt: Equip the employees by upskilling and re-skilling, focusing on human skills and emerging AI technologies. Continuous learning enables employees to adapt to evolving job demands based on advanced technologies. Create diverse and inclusive working environments that leverage the unique strengths and perspectives of team members. Communicate across the organization that AI augmentation will enhance job satisfaction, productivity, and quality rather than replace employees or cause job displacement. Individuals and organizations may prosper alongside AI by recognizing the importance of human-centric skills and tacit knowledge, utilizing both human and machine strengths to promote success, innovation, and growth.

The cobra effect is a phenomenon where our solution to a problem backfires and makes the issue worse than earlier. Some of the real-time examples are: In 1985, Coca-Cola announced a new formula to re-energize the brand to compete with Pepsi. The new formula was thought to have a positive impact among the customers, but it resulted in poor feedback from customers. Microsoft introduced Windows Vista as an advanced version of Windows XP, but its intensive design and compatibility issues made it one of the worst operating systems. To avoid the Cobra Effect during the Improve phase of a project, we need to follow the below steps: Define clear objectives to ensure that the solution addresses the actual root cause of the problems. Gather data and do brainstorming sessions with domain experts to analyze the potential unintended effects. Explore different alternatives to address the problems and evaluate their potential risks and benefits before selecting the final solution. Engage the stakeholders and end users who the new solutions will affect to make sure they live up to their expectations. Do “POC” and collect feedback to refine it further before the final implementation of the proposed ideas or solutions. If we follow the above guidance, we can minimize the risk of the Cobra Effect and be able to create better solutions to address the problems without any new or unintended consequences.

Ensemble methods combine many models with diverse architectures, and the predictions of individual models are aggregated using techniques such as averaging and voting, which reduces variance while improving prediction accuracy and robustness in data analysis. Advantages of Ensemble Methods: Improved Accuracy - Reducing defects by combining the strength of individual models Increased Robustness – Less sensitive to outliers Reduce variance Reduce overfitting Improved ability to adapt and handle uncertain data distribution over time Limitations of Ensemble Methods: Increased complexity requires strong expertise to handle it Difficult to interpret due to the complexity of the combined models Data overfitting if the data from individual models are incorrect Examples: Demand Forecasting based on season, market, and economic trends KTM 390 adventure model bike sold in 2025 Q1 was 1500 (Jan’25 – 500, Feb’25 – 450, Mar’25 – 550) and the expected sales in 2025 Q2 is around 1500 if the market condition remains the same.

Some team members or leaders in the organization do not report measures of dispersion for the following reasons: Reporting only the mean or median, presuming it is the best practice or easiest. Lack of knowledge about calculating the standard deviation and interquartile range (IQR). To exclusively portray positive outcomes of the central tendency to management or clients, rather than underlying negative trends such as variance, IQR, etc. Reporting only the measure of central tendency which will have an impact on decision making are: Incorrect conclusion since the dispersion measures like variability and IQR are excluded Uncovered potential risks since the outliers are ignored Incorrect forecasts since the underlying trends and patterns are not reported Example: For Client X, a team of four experienced and three new data abstractors are working. Each of them abstracted 200 images per month. 10% of their work is sampled the team's Median score is reported as 85%. Based on these results, the client approved onboarding three more new data abstractors to scale up the resources for their outsourced work. Since the measure of dispersion is not reported, it resulted in the incorrect decision of allocating a few more new data abstractors while the existing new data abstractors maintained low accuracy. However, on reviewing the measure of dispersion, it is noted the Inter Quartile Range (IQR) is 45%. Client X Sampled Correct Incorrect Accuracy Experienced Employee 1 20 20 0 100% Experienced Employee 2 20 19 1 95% Experienced Employee 3 20 18 2 90% Experienced Employee 4 20 17 3 85% New Employee 1 20 12 12 60% New Employee 2 20 10 10 50% New Employee 3 20 8 8 40% Overall 140 104 36 74% Q1 – Lower Quartile Part in the above-given data (Median Score of New Employee 1, 2, 3) – 50% Q2 – Median of the above-given data (Score of Experienced Employee 4) – 85% Q3– Upper Quartile Part in the above-given data (Median Score of Experienced Employee 1, 2, 3) – 95% IQR = Q3 – Q1 = 95% - 50% = 45%

Large Language Models (LLMs) are largely used in data analytics. One of the common issues that we could face while using ChatGPT after uploading the datasets in an Excel file is reading or accessing the Excel files. It is mainly due to the - Irregular formatting - Inconsistent data structure - Delimiters (like , .) present in the data itself - Formulas, macros, conditional formatting in the file itself - File size limitations