Anshuman Kumar

Experienced data scientist skilled in AI/ML and causal ML models across diverse industries, including retail, technology, and life sciences. Utilizes advanced analytical techniques to drive data-informed decision making and optimize business processes. Collaborates with cross-functional teams to design innovative solutions that yield measurable results.

Self-trained Chat-bot using Rag and Langchain

• Problem Statement

Since many projects work in silos across different teams and LOBs, harmonizing insights from to form a perfect market strategy becomes an issue.

• Objective

Create an ecosystem to help aggregate insights and results from all the teams across the firm and helps develop a strategy that learnings from all these insights.

• Approach

Use a central storage where all the results can be uploaded and create a vector DB to store these encoded reports.

Use a Langchain framework and a RAG pipeline to summarize all the learning from these reports to help form a more informed market strategy.

• Outcome

1. Brand architects and CEM leads were now able to leverage all the learnings and insights from all the teams before strategizing.

2. Helped improve engagement and conversion rate by 6% for the Q3 and Q4 of 2023.

Speech Classifier

• Problem Statement

Due to Covid there was a high customer fallout rate, which was causing significant revenue loss

• Objective

To predict the TNPS score of the current client base and newly onboarded clients, to get a realistic idea of their brand loyalty and maintain revenue.

• Approach

Using Cognitive learning (RNNS) and speech recognition on the Avaya hardware at the service delivery units to get realtime speech data and making use of rescoring algorithm to classify the conversation into a potential promoter or a detractor.

• Outcome

1. Helped identify the pain points of the existing customer base.

2. Helped us with a more proactive approach.

3. Helped improve overall NPS

Causal Inference Modelling

• Problem Statement

Correctly identify the initiatives that significantly contribute to revenue growth

• Objective

Prove Causality between customer engagement and sales lift for different brands and campaigns.

• Approach

Used various techniques, like Bayesian inference graphs, Double ML and counter factual top prove estimate individual treatment effect on a specific cohort of customers.

• Outcome

1.Successfully determined marketing initiatives that contributed to a significant uplift in sales.

2.Helped make business decisions basis causality and not just correlation.

KPI creation and Benchmarking (META)

• Problem Statement

Since Oculus is relatively new, there was a need to assess and identify KPI’s that affect the Customer experience.

• Objective

Helping the business identify key support metrics that affect NPS and in turn revenue and help setting up benchmarks.

• Approach

Creating a framework to run automated frequent correlation analysis amongst support metrics and NPS metrics.

Also identified metric sweet spots to assess influence on NPS.

• Outcome

1.The automated framework helps run regular and periodic analysis of the key support metrics to assess their influence on NPS and assign them business priority accordingly.

2.Machine learning was also implemented, it helped identify the sweet spots or possible benchmarks for SLAs of the individual metrics

• HUX member certificate - Deloitte
• Python Programming for data science and AI development - IBM
• Data Visualization with Tableau Project - University of California, Davis

• Rugby
• Gym
• Cooking
• Reading