My Portfolio
Pipelines, attribution platforms, dashboards and models — each one with the number it moved.
Marketing Intelligence Platform
Sole architect of a multi-touch attribution platform unifying Meta Ads, GoHighLevel, GA4 and Google Search Console through native API integration — a Streamlit application that replaced siloed marketing and CRM data with one source of truth.
5,200+ leads and a USD 10K/month media budget tracked end to end. CPL USD 7.45, cost-per-appointment USD 11.58. Adopted by the CEO as the primary strategic decision tool.
The Half That Never Showed Up
Nearly a third of booked consultations never happened — and outcome data was too unreliable to see it. I fixed the measurement first, then found the leak and closed it.
Show rate climbed from 68% to 82% — roughly one in five no-shows recovered, from the same bookings and zero extra ad spend.
The Most Valuable Thing Was What I Threw Away
A lead conversion model that scored a suspicious 100%. I audited it, found the leakage, deleted the data that made it perfect, and rebuilt it honestly.
77% accuracy against a 62% baseline. Top-scoring 20% of leads converted at 84%; referral converted at 90% versus 26% for paid social.
GoHighLevel CRM Architecture
Full CRM build-out — workflows, automations, pipelines, calendars and platform integrations — with data governance protocols that made GoHighLevel the enterprise single source of truth.
Every lead lifecycle stage captured consistently, which is what made attribution and show-rate measurement possible in the first place.
Illustrative — shows the direction and scale of the change.
Proven Results
Numbers from real work, each linked to how it was done.
Show rate recovered
Fixed outcome tracking so the number could be trusted, then closed the drop-off it exposed. One in five no-shows recovered — same leads, zero extra ad spend.
1 hr → 14 sReporting cycle compressed
Automated ETL replaced a recurring manual reporting workflow. Transcription errors eliminated, 4+ team hours reclaimed every iteration.
100% → 77%A model I refused to ship
A perfect score is a red flag. I found the leaking features, deleted them, and rebuilt honestly against a 62% baseline.
90% vs 26%Where the money actually was
Referral leads converted at 90% against 26% for paid social — a 3.5× quality gap that cost-per-lead hides completely.
4 → 1Platforms unified
Meta Ads, GoHighLevel, GA4 and Search Console reconciled into one source of truth. Adopted by the CEO as the primary decision tool.
+20%Planning efficiency
Z-test model predicting fabric dispatch from reprocess history, with colour-coded on-time probability across the floor.
68% → 82%Show rate recovered
Fixed outcome tracking so the number could be trusted, then closed the drop-off it exposed. One in five no-shows recovered — same leads, zero extra ad spend.
1 hr → 14 sReporting cycle compressed
Automated ETL replaced a recurring manual reporting workflow. Transcription errors eliminated, 4+ team hours reclaimed every iteration.
100% → 77%A model I refused to ship
A perfect score is a red flag. I found the leaking features, deleted them, and rebuilt honestly against a 62% baseline.
90% vs 26%Where the money actually was
Referral leads converted at 90% against 26% for paid social — a 3.5× quality gap that cost-per-lead hides completely.
4 → 1Platforms unified
Meta Ads, GoHighLevel, GA4 and Search Console reconciled into one source of truth. Adopted by the CEO as the primary decision tool.
+20%Planning efficiency
Z-test model predicting fabric dispatch from reprocess history, with colour-coded on-time probability across the floor.
How I Work
Make it measurable, then move it. In that order — it is the part everyone skips.
- 01
Audit
Map every source and document exactly where the definitions conflict. Most businesses have never seen that list written down.
- 02
Model
One agreed definition per metric, then the pipeline that enforces it. This is the step everyone skips and everyone regrets.
- 03
Build
The dashboard people actually open — headline first, drill-down underneath, no scrolling to find out if the month is on track.
- 04
Validate
When a number is not reliable yet, it gets flagged, not shipped as fact. A confident wrong number is worse than no number.
- 05
Decide
The "so what", on schedule. What changed, what it means, and what I would do about it.
About Me
Hafsa Saleh
Marketing Intelligence · ETL & Attribution · Predictive Modelling
I build the unglamorous infrastructure that decisions rest on. At The Migration, I engineered the cross-platform integration layer unifying Meta Ads, GoHighLevel, GA4 and Search Console into one Streamlit source of truth — tracking 5,200+ leads against a USD 10K/month media budget, adopted by the CEO as the primary strategic decision tool.
I came to this from statistics rather than marketing. My degree is in Statistics from the University of the Punjab, my first analyst role was on a factory floor, and I've published peer-reviewed research in Acta Geophysica. That's where the rigour comes from — and why I care more about a number being true than impressive.
Tools I Use for the Hard Part
Hover a card to see how each one earns its place.
Pandas, NumPy and Requests — every pipeline I build starts here
Modelling, joins and reconciliation across multi-source datasets
The live intelligence platform the CEO opens every morning
Executive dashboards and DAX measures over star schemas
Campaign, ad-set and creative data pulled natively, not exported
Workflows, pipelines and the governance that makes data trustworthy
Joining on-site behaviour and organic demand to revenue
AI-assisted development — a good part of my integration work
Hypothesis testing and the regression work behind my paper
Programmatic OKR sync between analytics output and delivery
Order lifecycle monitoring in a live manufacturing environment
Pivot tables and Power Query, for when that is genuinely the right tool
Research & Writing
Peer-reviewed work, and notes on turning data into decisions.
Acta Geophysica (Springer) · 2025Multiple Linear Regression Model for Air Quality Prediction in Islamabad
Models PM2.5 concentration in Islamabad as a function of sulphur dioxide, nitrogen dioxide, temperature and humidity using multiple linear regression. The validated model provides a reusable framework for ongoing air-quality monitoring and gives policymakers a quantitative basis for emissions and climate decisions in the region.
Read the paper →Drowning in Data, Starving for a Decision
Why more dashboards make decisions slower — and how the practices that move fast replaced five conflicting reports with one number they'd bet on.
Read → 24 Aug 2026Clicks Don't Pay the Bills
Why cost per lead is a trap — and how connecting online ad data to offline revenue reveals which channel actually produces paying clients.
Read → 22 Aug 2026You're Filling a Leaky Bucket
The cheapest growth you have is keeping the clients you're quietly losing — and the signal that predicts who's about to lapse is already sitting in your data.
Read →What I Offer
Five fixed-scope services, priced up front. The dashboard starts with a free month.
Plug-and-Play Intelligence Dashboard
The live dashboard from my portfolio, connected to your own platforms. Ads, CRM, analytics and payments in one view — so you can finally see cost per paying client by channel, not just cost per lead.
Free first month, no card Digital productClient Reporting Dashboard Template
Turn the monthly client report from a half-day rebuild into a five-minute job. Paste your numbers into one tab; a branded, client-ready dashboard updates itself.
$39 one-off, instant download Machine learningPredictive Model — Churn or Lead Scoring
A Random Forest model that ranks every client or lead by risk of lapsing, or likelihood of converting. Built honestly, audited for leakage, and delivered with the reasons behind the score.
From $750 one-off CRM architectureFull GoHighLevel Build
Workflows, pipelines, calendars and integrations built properly the first time — with the data governance that makes every downstream number trustworthy.
From $850 one-off Done-for-you systemAutomated Reporting System
One month of work and reporting stops being someone's job. A self-running system that pulls, reconciles and delivers your numbers on schedule — needing only occasional attention afterwards.
From $1,800 four-week buildLet's Build Something Measurable.
Open to analytics roles, freelance work on attribution and reporting infrastructure, and research collaboration.