Dipali Yadav

Honestly, advancement in AI is beyond imagination. And not because personalization is bad, but because it’s happening without real consent. Personalization is great when it’s obvious and useful. If I get better recommendations or save time, I’m all for it. But the moment AI starts predicting things about me that I never shared or quietly tracking behavior; that’s where it crosses the line. The industry loves to frame this as a trade-off, but I don’t fully buy that. It’s not “privacy vs personalization". You can build great personalized systems without hoarding every piece of user data. For me, the boundary is simple; If I didn’t knowingly give AI that data, or I can’t easily control/delete it, AI shouldn’t be using it. Until companies treat user data with that level of respect, personalization will keep feeling less like a feature and more like surveillance with better UX.

No, AI should not run the entire process without humans. No, artificial intelligence should not be allowed to operate an entire process without human involvement. Understanding AI Artificial Intelligence (AI) is designed to simulate human thinking and decision-making. However, while it can process data and identify patterns, it does not guarantee accurate or contextually appropriate outcomes in every situation. Rationale There are several reasons why AI should not function independently in critical scenarios, particularly in industries where human life is directly impacted, such as healthcare and the automotive sector. In healthcare, for instance, there is growing adoption of robotic-assisted surgeries. While these technologies enhance precision, they rely heavily on pre-programmed instructions and trained datasets. Consider a complex procedure like heart surgery—if the system encounters an unexpected situation or if the programmed inputs fail to align with real-time conditions, the consequences could be life-threatening. Limitations of AI Inability to understand or respond to human emotions Limited capacity to adapt to unforeseen or untrained scenarios Risk of significant financial loss in case of system failure Importance of Human Intervention Humans can take immediate control in critical or unpredictable situations They help minimize the risk of severe harm or casualties They possess the ability to interpret context and adapt decisions based on changing circumstances In conclusion, while AI is a powerful tool, it should complement human expertise rather than replace it—especially in high-stakes environments.

In the Quality and Transformation team, we manage large volumes of operational data under strict SLOs. Over time, preset validation rules were implemented to ensure data quality and prevent escalations. However, despite these controls, escalations continued to occur. Because the process had been in place for a long time, it was largely accepted as “working as designed,” and issues were addressed reactively through RCA rather than proactively questioning the effectiveness of the rules themselves. Where AI Revealed InefficiencyTo better understand the issue, we created a Power BI dashboard to structure and visualize the data flow. The analysis revealed: 70% of inputs were being archived before entering the system — a clear example of Lean waste. 20% of data was processed for final validation. 10% failed at first-level approvals, which was contributing to escalations. The RCA showed that validation rules were not functioning as effectively as assumed. The inefficiency was not visible earlier due to volume pressures and reliance on legacy validation logic. We are now developing an AI agent to proactively identify the 10% of failed datasets and systematically notify users. This shifts the process from reactive correction to proactive intervention. Response we received from stakeholders: Frankly speaking initially there was lot of resistance from stakeholders to take this initiatives knowing bandwidth and cost issues. But, it got approved using following tactics; With the help of AI checks we are able to identify usage% of current validation rules and not generic sentiment of users. Out of 69 validation rule AI was able to identify 40 rules which needed immediate attention for modification. We had finalized one function as a POC from six function to check the feasibility and success of AI. Established monthly cadence and RACI model to ensure governance is in place for this project. Working with SME experts to deliver on time results to improve performance of the validation rules. Improvements of the AI-Driven ApproachIncreased accuracy of validation rules this is the the focus we had Involvement of stakeholders in this initiatives helped reduction in escalations Timely and predictable deliverables Better workload alignment Stronger governance and monitoring mechanisms Reduced emotional resistance due to objective, data-backed insights In summary, when AI surfaces inefficiencies that challenge established practices, the organization must respond with openness, structured governance, and collaborative problem-solving. When positioned correctly, AI becomes an enabler of continuous transformation rather than a trigger for defensiveness.

I work in Investment Data Operations, where multiple teams collect 46,000 data points across \20,000 fund companies using diverse external sources. Despite reported quality levels of 98% across Accuracy, Timeliness, and Completeness, clients raised frequent escalations related to missing data, low coverage, and incorrect values, indicating a significant gap between reported quality and experienced quality. Define: With AI support, we reframed Define as problem validation, not problem articulation. Dashboards helped identify patterned complaint clusters by data point, fund type, and client segment. This revealed that the “98% quality” baseline was statistically invalid and masking localized failures. Human judgment was important in reframing the problem statement, aligning sponsors, and defining SMART goals that balanced client impact and operational feasibility. Measure: Measure was the most transformed phase. Manual sampling (30–100 records per fund) was replaced by automated audits connected directly to final production databases. True completeness and accuracy measurement at the data-point level Revised baselines were established using statistically valid methods instead of uniform targets. Analyze: lack of VSMs and shallow RCAs. AI-enabled lineage mapping traced each escalation to the exact source, transformation, and handoff. Usage analytics identified high-impact data points disproportionately affecting strategic clients. Validations are still manual for few to avoid false alarms created by AI Improve: Automated audit mechanisms Power BI monitoring dashboards with lineage-based impact views SOP knowledge hub using an LLM to validate process adherence and version accuracy AI helped simulate; Impact of focused improvements on high-usage data points Expected reduction in escalations by client segment Control: Established governance AI-driven monitoring flagged anomalies, drift, and emerging risk patterns Governance models, escalation thresholds, and ownership remained human-defined Control plans evolved dynamically instead of being “set and forget” Final solution selection remained a human decision, balancing cost, risk, and change adoption.

"Six Sigma Green Belt training is very helpful to understand and improve processes in a structured manner. Trainer perceives excellent knowledge of the subject. Great sessions with impactful learning. Keep up the great work...." -Dipali Yadav, Project Leader, Here Maps