VB
Data • Product • AI & Analytics

Vyom
Bhushan.

Builder-Scaler in product and growth analytics. I design and scale data systems that turn ambiguity into measurable growth.

A bit about me.

Vyom Bhushan
I'm Vyom. I've been the sole data hire at a startup that scaled to 1M users, caught fraud that was quietly bleeding $100K/month, shipped a SaaS product to $2M ARR at a global consulting firm, and built scoring models that changed how an entire sales org prioritises outreach.

I work across the data full stack: experimentation, propensity models, analytics engineering, AI workflows. I'm most useful when things are ambiguous, the data is messy, and someone needs to figure out what to measure and why it matters.
6+
Years in Data & Analytics
30%
Conversion Lift (Best)
$125M
Monthly GTV Managed
1M+
Users Tracked & Analysed

Where Curiosity Led Me

A peek into the places I've explored, learned, and contributed

Allica Bank

Conversion & AI Ops

Sr. Analytics Engineer, Growth
Jan 2024 - Present • London (Remote)
Built XGBoost propensity models increasing conversions by 30%. Created customer scoring framework boosting RM outreach success by 45%. Shipped an internal RAG-based AI assistant as single source of truth.
CheQ

Founding Data Team

Lead Data Analyst, Product
Jul 2022 - Dec 2023 • Bangalore
10th employee. Sole data person for year 1. Architected entire analytics stack. Designed A/B test on fees → $800k/mo in platform revenue. Built fraud model saving $100k/mo. Scaled to 1M+ users.
ZS Associates

0-to-1 SaaS Product (ZAIDYN)

Senior Product Analyst
Jan 2021 - Sep 2022 • San Francisco (Remote)
Built and launched Budget Planner within ZS's ZAIDYN platform, helping pharma companies optimize marketing spend. Product Owner for team of 30. Reached $2M ARR in Year 1 with Fortune 500 clients.

Work I'm Proud Of

From the trenches

CheQ

Event Tracking & Onboarding Diagnostics

The Problem
With 1M+ users onboarding, there was no structured event tracking - zero visibility into where users were dropping off or why.
What I Built
Deployed a Segment-style event tracking architecture using mParticle, funneling behavioural data into BigQuery. This became the foundation for funnel analysis, cohort retention, and feature adoption tracking across the org.
The Incident
The funnel revealed a 50% drop-off at the onboarding page itself - far above expected. I built a step-by-step customer funnel to diagnose. Turned out half the users weren't receiving their OTPs for sign-up. The issue was on the SMS provider's end. Got it escalated and fixed - onboarding completion recovered immediately.
Event tracking became the backbone of all product decisions. The OTP fix alone recovered ~50% of lost sign-ups overnight.
mParticle BigQuery Metabase Figma
CheQ

Fraud Analysis - Rewards Engine Exploit

The Problem
Reward burn was running at 7–8x forecast. Unit economics weren't making sense at the cohort level. There was no fraud team, no monitoring, no alerts.
What I Found
~7% of users were driving 55% of total reward burn through two mechanisms - duplicate accounts exploiting referral bonuses, and transaction reversals made after rewards had already been earned. This was quietly bleeding the business model.
The Hard Part
Overcorrecting risked adding friction for legitimate users and slowing growth. Undercorrecting meant letting an incentive design flaw compound. That tension shaped every stakeholder conversation.
What I Built
Analysis in BigQuery using mParticle event data cross-referenced in Looker. Detection: device fingerprinting, behavioural clustering for duplicates, reversal-to-reward timestamp mapping. Designed a rules-based flagging system (device ID matching, reversal frequency thresholds, redemption velocity triggers). Redesigned onboarding to block duplicates. Set up automated Slack alerts and a Metabase dashboard for ops.
~$100K/month saved in reward burn. No noticeable retention impact. Gave leadership confidence in unit economics ahead of Series A.
BigQuery mParticle Looker Metabase Slack
ZS Associates / ZAIDYN

0-to-1 SaaS Product: Budget Planner

The Problem
Pharma commercial teams were planning marketing budgets in spreadsheets. No way to model channel-level ROI, no scenario simulation, no shared view across brands. ZS had deep consulting expertise in marketing mix but no self-serve product clients could use independently.
What I Did
I was the cross-functional glue between product, engineering and design on a team of 30. Defined the success metrics, designed the analytical models underpinning budget optimization, and owned the product roadmap. Spent most of my time translating what Fortune 500 pharma stakeholders needed into specs engineers could build against.
The Tradeoff
Building for pharma means navigating compliance, slow procurement cycles and stakeholders who think in PowerPoint, not product. The constant tension was between shipping fast to prove SaaS viability for ZS and building robust enough to survive enterprise scrutiny. We chose to ship narrow and deep rather than broad and shallow.
$2M ARR in Year 1. Budget Planner is now a live product in the ZAIDYN Marketplace, used across 65+ countries.
AWS Marketing Mix SaaS Product Strategy
Allica Bank

Customer Scoring Framework

The Problem
Growth and Sales teams were prioritising RM outreach manually with no consistent ranking. A lot of time was being spent trying to convert low-value customers. No one agreed on what a valuable customer looked like or when to reach out.
What I Built
Designed a composite scoring model combining opportunity signals (balance growth, wallet share, transaction volume, card activity, savings) with timing and engagement signals (call connect rates, email engagement, recency, primary account behaviour). Each component was individually scored and exposed, not black-boxed.
Data Quality
Built end-to-end quality checks, reconciliation logic across sources (HubSpot, CASS, balances), and guardrails to prevent silent row drops. Ensured every dashboard output was defensible to senior stakeholders.
The Tradeoff
A simpler score would have shipped faster but wouldn't have earned trust. RMs needed to understand why a customer ranked high, not just that they did. Exposing each component individually meant more complexity to build, but it gave stakeholders a score they could interrogate and believe in.
Gave RMs a single trusted ranking. Replaced broad manual targeting with data-driven prioritisation. Outreach success improved by 45%.
BigQuery HubSpot Metabase SQL

The Build Log

What I'm learning, building and shipping

Jan 31, 2026

The Reality of Text-to-SQL at Scale: Cost, Latency, and What Actually Works

A deep dive into what happens when you try to ship NL-to-SQL in production - the tradeoffs between cost, latency, accuracy, and the architecture decisions that actually matter.

Text-to-SQL LLMs Production AI

Pieces Of Me

A glimpse into the shades that define who I am

The Data Detective

Fraud nobody was looking for, OTP failures hiding in plain sight, pipelines silently dropping rows. I find what's broken before it becomes a crisis.

The Experimenter

From A/B tests to life decisions, I believe in letting data guide the way. Bayesian at heart, curious by nature.

The Builder

Whether it's a 0-to-1 product or an analytics stack from scratch, I thrive on creating something from nothing.

The Storyteller

Numbers don't move people, narratives do. I turn data into the kind of story that gets executives to act, whether it's sizing fraud in dollars or making a score stakeholders trust.

Let's Connect.

Always up for a conversation about data, AI, or the next big thing.

Email Me LinkedIn GitHub
More is lost by indecision than a wrong decision Action leads to motivation, not the other way round If something is 99% done, it's closer to 0% than 100% Cost of inaction is an exponentially increasing function Decline is slow, but the collapse is sudden Impatience with actions, patience with results More is lost by indecision than a wrong decision Action leads to motivation, not the other way round If something is 99% done, it's closer to 0% than 100% Cost of inaction is an exponentially increasing function Decline is slow, but the collapse is sudden Impatience with actions, patience with results