About the founder
Arvind Mozumdar
I trained as a statistician at the Indian Statistical Institute and spent twenty years delivering data and AI outcomes for organizations with effectively unlimited budgets. Since April 2025 I have run CDO On-Demand, which brings the same caliber of leadership to businesses that could never justify the salary it normally commands.
Why I do this work
Over two decades I delivered the kind of results that large budgets make possible: twelve million dollars of deferred capital in a semiconductor fab, six million a year for a luxury retailer, four million from a single generative AI system. I know precisely what that caliber of leadership produces, because I produced it.
What became difficult to ignore was that the companies anchoring real communities — the banks, the advisory practices, the clinics, the carriers, the distributors — were locked out of it entirely, and AI was widening that gap rather than closing it. The obstacle was never talent or technology. It was a delivery model that only made sense at enterprise scale, and that is a fixable problem.
There is a second reason, and it is more personal. In 2025 I left the security of large consulting to build a small business of my own, so I now live what my clients live: every dollar examined, no bench to draw on, nobody to delegate to. I automated my own operations with AI well before I proposed any of it to a client. When I talk about doing more with less, I am describing how this business actually runs.
2025–2026
What I have been building lately.
Hands-on generative AI engineering across an AI product startup, a global commercial insurer, and mid-market operations teams. I build daily in Claude Code and Google Antigravity, which keeps the advice grounded in what these tools actually do rather than what the marketing claims.
Global commercial insurer
Senior AI engineering strategist. Benchmarked several open-weight models across short-form accuracy, long-form accuracy, coherence, and response quality on an isolated, data-resident deployment built to keep client information inside the enterprise boundary. Designed a lightweight graph-retrieval prototype with document classification, and defined the roadmap for the internal platform covering first notice of loss extraction, policy submission ingestion, and a transformation agent that pulls target fields out of non-standard spreadsheets.
AI product startup
AI engineering lead on a five-person team building a video-avatar training and simulation platform. Designed the prompt architecture that holds character fidelity and realism steady across long unscripted conversations, which the team adopted as its standard, and built a statistical test-analysis suite that turned prompt work into a measurable engineering discipline rather than a matter of taste.
Mid-market operations
Shipped automation with outcomes the client could measure. Structured extraction from multi-format vendor emails reduced a process that consumed more than five hours per person each week down to thirty minutes, producing a competitive pricing database along the way. Automated sales outreach moved weekly customer reach from 60% to 150% of the base at full inventory-match accuracy.
Client names withheld under confidentiality. Detail available on request.
Delivered impact
Across twenty years.
$12M
Capital expenditure deferred through predictive quality models at Seagate
$6M
Increased annual spend from a luxury retail customer insights suite
$4M
Annual manual effort removed by generative AI product attribution
$25M+
Cumulative revenue sold while building analytics practices
The longer arc
Where the pattern recognition comes from.
Fractional leadership works because somebody has already encountered your problem several times elsewhere. The engagements below are where I accumulated that, and each one involved building a capability rather than simply running one.
2022–2024
Logic Information Systems, acquired by Accenture
Client Delivery Leader and Data Science Practice Lead
Led global analytics delivery and grew the practice. A customer insights suite for a luxury retailer built across Snowflake, BigQuery, Vertex AI, and Pub/Sub produced $6 million in increased annual spend. A generative AI system assigning more than forty product attributes from imagery and metadata removed $4 million a year in manual effort and production error.
2019–2022
Daugherty Business Solutions, acquired by CGI
Enterprise Data Solutions Lead
Grew the regional data science practice from nobody to six people in the first year. Price optimization for an electronics retailer on BigQuery, Kubeflow, and Vertex AI carried $12 million in expected annual impact, alongside a cloud transformation program valued at $23 million over three years.
2015–2019
Seagate Technology
Data Science Principal, Wafer Manufacturing
Established the wafer AI practice and deployed the first machine learning models ever used in the fab, delivering more than $3 million in annual impact. Quality prediction from machine sensor and image data using deep learning deferred $12 million of capital expenditure across three years, and I designed the governance framework that the model suite ran under.
2006–2014
Mu Sigma
Director, Data Science and Analytics
Grew strategic accounts and sold more than $25 million in cumulative revenue while managing portfolios above $5 million and global teams of up to seventy analysts across twenty-five clients in financial services, insurance, healthcare, and retail. This is where I learned to build analytics teams and delivery practices from a standing start.
Education and credentials
- M.S. and B.S. in Statistics, Indian Statistical Institute, Kolkata
- AWS Certified Cloud Practitioner, recertified 2024
- SAFe 5 Product Owner / Product Manager, and Scrum Product Owner
- Five years architecting on Google Cloud: BigQuery, Vertex AI, Looker, Pub/Sub
- Working depth in Snowflake, Databricks, Azure AI Foundry, and AWS SageMaker
Based in Eden Prairie, Minnesota, and working with clients across the United States.
Happy to talk through your situation.
If any of the above resembles the problem in front of you, half an hour on a call will establish quickly whether I am the right person to help.