DriverData Analyst (Campaign Planning & Audience Segmentation)WaterAid UK

Yaser
Selvam

I run audience segmentation and donor targeting for multi-channel fundraising campaigns at one of the UK's biggest charities, and I'm bringing AI into the selections workflow, with a human in the loop at every step.

See the workGet in touchLondon, UK
01

About

I'm a data analyst at WaterAid UK, specialising in audience segmentation and donor targeting across multi-channel fundraising campaigns. I work in T-SQL across enterprise data warehouses, turning campaign briefs into accurate, auditable selections across Direct Mail, Email and Telephone.

I'm currently leading work to bring AI into our campaign selections, and building segment-level performance insights for campaign planners. I hold an MSc in Data Science and Analytics from the University of Leeds and came up through data engineering at Cognizant.

Outside the day job I founded Prometrix, an AI-powered CV optimisation tool built on a strict human-in-the-loop principle. I treat AI the same way everywhere: a tool to make people faster and work safer, always under human judgement and never trusted until it is checked.

T-SQL
daily driver at WaterAid
MSc
Data Science & Analytics, Leeds
AI
augmented + human-verified workflow
GDPR
compliant handling of supporter data
02

Experience

Data Analyst (Campaign Planning & Audience Segmentation)

WaterAid UK

Sep 2025 to Present
  • Design the audience segmentation and campaign selections behind around 330 campaign selections a year, across email, mail, phone, SMS and online activity.
  • Built an AI-augmented, human-verified SQL workflow on eight prompt templates, now in controlled trial across a three-analyst team, with an analyst checking every generated query before it runs.
  • Separated the fields that are genuinely personal data from those that only resemble it; the resulting restrictions document is being built into organisation-wide AI tool policy.
  • Authored the discovery report on the end-to-end selections workflow: scored 10/10, adopted as the basis for further work, and praised in writing by the department head.
  • Corrected two figures in the team's business case, one cutting the claimed saving by a third after separating elapsed time from effort, and sourced every rate to a documented method.
  • Owned the stewardship selection builds and data delivery for Walk for Water 2026 across four sends.
T-SQLSQL ServerClaudeDotDigital

AI Contributor

Outlier AI

Feb 2025 to Jun 2025
  • Completed ~150 AI training and evaluation tasks across four project queues, spanning response ranking, prompt creation, annotation and rewriting.
  • Evaluated 100+ pairs of LLM responses against 5-7 rubric criteria covering factual accuracy, reasoning quality, relevance and instruction adherence, producing 100+ written rationales supporting human-feedback training data.
LLMsPrompt EngineeringEvaluation

Programmer Analyst

Cognizant Technology Solutions

Nov 2022 to Jun 2023
  • Developed SQL and supported PySpark transformations across roughly six pipelines at a Fortune 500 IT firm, consolidating four retail sources into 20 to 25 curated tables that became the reporting base for 15 to 20 analysts.
  • Refactored a daily sales and inventory pipeline with a senior engineer, replacing full-history scans with incremental processing and cutting runtime from 90 minutes to 60.
  • Triaged around 12 failed production runs using Azure Data Factory and Databricks logs, resolving about eight without senior escalation and restoring loads before the reporting window.
SQLPythonPySparkAzure Data FactoryAzure Databricks

Junior Data Analyst

CRYSTALVIEWHD (formerly NLTVC Sdn Bhd)

Jan 2022 to May 2022
  • Built the reporting base for 150 customer sites, consolidating five SQL source tables and around 10,000 records into a single validated dataset in place of scattered files.
  • Analysed around 1,300 support cases to identify seven recurring fault categories and the sites failing repeatedly, ranking where engineering effort would pay back most.
  • Supported delivery of a three-page Power BI dashboard tracking nine KPIs that replaced manual spreadsheet comparison for eight stakeholders, and automated the monthly refresh to cut reporting preparation from four hours to 90 minutes.
SQLPower BIData Modelling
03

Selected Work

AI-Augmented Customer Segmentation (1M+ Transactions)

Flagship

End-to-end RFM segmentation and audience selection on 1M+ public retail transactions: nine customer segments with per-segment targeting recommendations, built with an AI-augmented SQL workflow I direct, verifying every query against audit checks before it is used.

SQLDuckDBLLMRFM

Read the case study →

Prometrix

Founder

AI-powered CV tool that gives candidates the same AI advantage employers use to screen them. Founded and in active build, human-in-the-loop by design: it strengthens a real CV, never fabricates one, and a person approves before anything is sent.

PythonLLMsPrompt EngineeringPipelines

Read the case study →

Product A/B Test: From Hypothesis to a Ship Decision

Experimentation

A worked A/B experiment end to end: hypothesis, power, validity checks, significance testing and guardrails, turned into a clear ship / don't-ship decision. On the public Cookie Cats dataset, moving a game's progression gate lowered 7-day retention, so the call was: don't ship.

A/B TestingPythonstatsmodelsExperimentation

Read the case study →

GA4 Product Analytics: dbt on BigQuery & Snowflake

New

End-to-end product analytics on the public GA4 e-commerce dataset: raw events modelled with dbt on both BigQuery and Snowflake (staging to marts, tested and documented), producing identical results across warehouses, analysed for funnel conversion, retention cohorts, and RFM segmentation, ending in a clear recommendation.

dbtBigQuerySnowflakeFivetranSQLPower BI

Read the case study →

Data Observability & Governance Layer

New

The operational layer top UK fintech data teams run in production, built on the GA4 dbt warehouse: a flag-never-drop data-quality model, volume and distribution monitors, an ownership-tagged alerter that routes failures to the owning team (Monzo's #data-monitoring pattern), and a Monzo-style CI gate that fails any pull request where a model lacks an owner, docs, or a test.

dbtData QualityObservabilityCI/CDPython

Healthcare Flu Shots 2022 Dashboard

Vaccination rate analytics and demographic insights for healthcare providers, from raw SQL to executive Tableau views.

SQLPostgreSQLTableau

Multiple-Outcome Treatment Effects (MSc Dissertation)

How hypothesis testing changes when a treatment is judged on several correlated outcomes: co-primary endpoints and joint rejection regions, the Type I / Type II error trade-off, and how outcome correlation shifts statistical power and required sample size. Reproducible R simulations (MASS, ggplot2) from my MSc dissertation.

Rggplot2Hypothesis TestingStatistical Power
04

Skills

Analytics & Targeting

Audience SegmentationDonor / Customer TargetingSuppression & Dedup LogicSQL / T-SQLData QAGDPR Compliance

AI & Automation

LLM-augmented SQLHuman-in-the-loop designResponsible AIPrompt EngineeringLLM Evaluation

BI & Visualisation

Power BITableauMicrosoft Fabric

Data Engineering

PythonPySparkDatabricksSnowflakeAzureAWSGCP

Web & Product

ReactNext.jsTypeScriptTailwind CSSNode.js
05

Education

MSc Data Science and Analytics, Merit

2023 to 2024

University of Leeds

Business Analytics · Statistical Theory · Machine Learning · Deep Learning

BE Computer Science and Engineering, 1st

2018 to 2022

Meenakshi Sundararajan Engineering College

Cloud Computing · AI · Database Management · Data Structures · Operating Systems

06 · CONTACT

Let's build something
data-driven.

Open to conversations about data, AI and analytics roles, or anything worth building.

yaseruk259@gmail.comLinkedInGitHub