The model is a commodity
Same model, radically different outcomes. What separates them is system design, data quality, retrieval architecture and the shape of the context you hand it — not the frontier release on the label.
About
Abhishek Prasar
Enterprise AI Transformation Leader, WNS Global Services (now Capgemini) · Gurugram, India
I have spent fifteen years building AI inside enterprises — first the models, then the platforms around them, and now the organisations that have to live with both. The work has tracked the field: regression and churn models on telecom telemetry, NLP and entity extraction for enterprise clients, then large-scale people analytics, then GenAI, and now agentic systems in production.
Today I lead AI transformation for the commercial function of a $3.3B organisation — owning architecture, governance, investment decisions and the adoption roadmap, reporting to global commercial leadership. The current work is a GTM intelligence layer: an account-signal platform that says who to pursue and when, and an agentic planning platform that says how. Before this I built a 20-person AI practice from zero at LTIMindtree and ran a ~$500K annual cost base against it.
What I keep finding is that the interesting problems are not modelling problems. They are data-boundary problems, retrieval problems, governance problems, and — most often — adoption problems. I write here about that half of the job, because it is the half that decides whether anything reaches production.
How I think
Three things I keep coming back to
Not a philosophy. Just the patterns that have held up across enough organisations, and enough generations of the technology, to trust.
Governance from day one
Data boundaries, DLP, PII handling and responsible-AI review are cheap to design in and expensive to retrofit. In regulated environments they are also the difference between a pilot and something that ships.
Adoption is the product
A platform nobody opens has delivered nothing. Most resistance is rational — people burned by earlier tools, missing access, misaligned incentives — and each of those needs a different intervention.
Selected work
Platforms I have shipped
Systems that reached production and stayed there, rather than proofs of concept. Client detail withheld where it should be.
- 01
GTM Intelligence Layer — Signal Platform
WNS / Capgemini
Account intelligence built on open data: 10-K and 8-K filings, earnings calls, Yahoo Finance and Google News, executive movement, hiring and funding signals — combined into a composite buyer-intent score for pipeline prioritisation across BFSI, Healthcare and other verticals. Replaced a $4,000/year commercial subscription.
- 02
GTM Intelligence Layer — Account Planning Agent
WNS / Capgemini
Production agentic platform turning CRM records, account signals, internal proof points and an enterprise knowledge base into a pursuit-ready plan in under ten minutes rather than two to three hours. It closes the loop: signal, priority, plan, pursuit.
- 03
Proposal & Pursuit Strategy Copilot
WNS / Capgemini
LLM-powered pursuit system for $20M–$100M RFP cycles — pre-RFP strategy, win themes, executive briefings, SME mobilisation, orals prep and post-meeting iteration. Source-grounded and human-gated throughout.
- 04
AI Talent Marketplace (SmartMatch)
LTIMindtree
Two-way agentic matching across 100,000 users at 90% accuracy, with 90,000 structured skill profiles created behind it. Replaced a ~$200K annual vendor licence and, unlike it, could explain every match it made.
- 05
Enterprise GenAI Suite
LTIMindtree
The organisation’s first production GenAI, in 2023: a Sales Proposal Bot (~40% less RFP preparation effort), an HR Policy Q&A Bot (~35% lower helpdesk load) and a Cybersecurity Risk & Control Bot. Standards co-developed with Microsoft product teams.
- 06
People Analytics & ML Platforms
LTIMindtree
Joining propensity, a three-layer explainable attrition predictor built to be usable by HR business partners, employee-project matching on Azure ML with NLP embeddings at 90% accuracy, and salary forecasting.
Background
Where I have worked
The short version. The longer one is on LinkedIn.
- Sept 2025 — present
Senior Consultant (AVP), Enterprise AI & Sales Transformation
WNS Global Services (now Capgemini) · Gurugram
AI strategy and portfolio for the commercial function of a $3.3B organisation — architecture, governance and investment decisions, reporting to global commercial leadership. Built the GTM intelligence layer: a proprietary account-signal platform and a production account-planning agent, now serving 300+ users across a $570M+ deal portfolio.
- Nov 2019 — Sept 2025
Director, AI/ML Solutions, GenAI Products & Enterprise Intelligence
LTIMindtree · Bengaluru
Built and led a 20-person AI/ML practice from zero inside the CIO organisation, owning a ~$500K annual cost base. Shipped the organisation’s first production GenAI solutions in 2023, and a 100,000-user AI talent marketplace that replaced a ~$200K black-box vendor licence with an explainable in-house platform.
- Mar 2019 — Oct 2019
Advisor, Data Science & Analytics
Dell Technologies · Bengaluru
Predictive analytics across TB+ consumer datasets, end to end from data sourcing through production integration — a 20% operational efficiency improvement.
- Feb 2018 — Feb 2019
Lead Data Scientist
Tata Consultancy Services · Bengaluru
Led a three-person team on NLP text classification and entity extraction for enterprise clients across the US, UK and Asia — 30% improvement in reporting accuracy.
- Apr 2015 — Jan 2018
Machine Learning Engineer
Ericsson · Gurugram
Regression, classification and churn models for telecom network performance analytics, and the Python pipelines behind them, for clients in the USA and South Africa.
- 2011 — 2015
Data Scientist / ML Engineer
Saburi TLC · Axon Network Solutions · ASMA Financial Solutions
Churn prediction, customer segmentation, demand forecasting and revenue analytics across telecom, networking and financial services — where the Python, SQL and statistical foundations were laid.
Credentials
Education and the stack I build on
Tools change every eighteen months. These are the ones currently in hand.
- Executive MBA Indian Institute of Management (IIM) Indore
- 2025 — 2027 (in progress)
- B.Tech, Computer Science Dr. MGR Educational and Research Institute, Chennai
- 2011
Technical stack
- Python
- SQL
- Azure ML
- Azure OpenAI
- Azure AI Search
- Azure APIM
- Azure Monitor
- Azure DevOps
- Copilot Studio
- Power Platform
- LangChain
- LangGraph
- RAG pipelines
- Vector databases
- Prompt engineering
- LLMs
- LLMOps
- Machine learning
- Deep learning
- NLP
- scikit-learn
- PySpark
- Databricks
- Snowflake
- AWS
- GCP
- Power BI
- FastAPI
- Docker
- MLOps
- Streamlit
Working on the same problems?
I am glad to hear from AI and engineering leaders, founders, and anyone who disagrees with something I have written here.