From a management degree to Credit Analyst at Unibank Baku. My real path into data — SQL, banking, and grit.
Get new posts delivered to your inbox. No spam, unsubscribe anytime.
September 2019. I walked into Unibank's Baku office with a management degree from Mingachevir State University, zero SQL knowledge, and a borrowed suit that was half a size too big. I was interviewing for a Credit Analyst position I barely understood.
The hiring manager asked me what tools I used for data analysis. I said Excel. She paused. Then she asked if I knew SQL. I said no, but I could learn. She paused again, longer this time. I thought that was it. I thought I'd take the marshrutka back to my apartment, open LinkedIn, and keep scrolling.
She called me back two days later with an offer. I still don't fully know why. Maybe they were desperate. Maybe she saw something. Maybe the other candidates were worse. Whatever the reason, that phone call changed the entire direction of my life.
This is the story of how a kid from Mingachevir with a management degree ended up building fraud detection models and running a tech startup. None of it was planned. Most of it was terrifying. All of it was worth it.
Let me be clear about something. I'm not from Baku. I'm from Mingachevir — a small industrial city about 300 kilometers west of the capital. If you know Azerbaijan, you know what that means. If you don't, here's the short version: 71.8% of permanent jobs created in Azerbaijan between 2019 and 2023 were in Baku. The rest of the country shares the other 28.2%.
Growing up in Mingachevir, my exposure to technology was limited. We had computers in school, but nobody was teaching programming. The career advice was simple: become a doctor, a lawyer, or go into business. I went with business. Sort of.
I enrolled at Mingachevir State University and graduated in 2016 with a Bachelor's in Management of Industry. A perfectly respectable degree that prepared me for... managing things. Factories, maybe. Supply chains. I learned organizational theory, accounting basics, some economics. Not a single line of code.
The gap between my degree and the tech industry felt enormous. Because it was.
I graduated in 2016. My first analyst job started in September 2019. That's three years. Three years that I mostly want to skip over, but I won't, because I think the gap is the most honest part of this story.
After graduation, I did what most graduates from small cities in Azerbaijan do. I looked for work in Mingachevir first. There wasn't much. I moved to Baku, like everyone else. The first year was odd jobs, confusion, and a slow realization that my management degree wasn't opening the doors I expected.
I worked temporary roles. Nothing related to data. Nothing related to analysis. I was surviving, not building a career. And I was watching friends with engineering degrees land tech jobs while I was stuck trying to figure out what I actually wanted to do.
Somewhere around 2018, I stumbled onto a YouTube tutorial about Excel formulas. Then pivot tables. Then VLOOKUP. I know that sounds small, but for someone who'd never thought of data as a career, it was a spark. I started watching more. Excel led to "what is SQL?" which led to online courses which led to late nights writing SELECT * FROM statements on my laptop.
I wasn't strategic about it. I didn't have a roadmap. I just followed what felt interesting and hoped it would lead somewhere.
Here's something nobody tells you if you're trying to break into data in Azerbaijan: banking is your best bet. Not tech companies. Not startups. Banks.
Azerbaijan's banking sector is small but it's one of the few industries that consistently hires analysts. Unibank, Kapital Bank, Bank of Baku, PASHA Bank, ABB — these are the companies with actual data teams. They need people to assess credit risk, detect fraud, build reports, and analyze customer behavior.
The barrier to entry is lower than you'd think. Banks care about whether you can think analytically and learn on the job. A management degree isn't ideal, but it's not disqualifying either. They're not looking for PhDs. They're looking for people who can figure things out.
When I applied to Unibank's Credit Analyst position in 2019, I had no SQL certification, no Python skills, and no GitHub profile. What I had was a basic understanding of Excel, genuine curiosity, and three years of wanting something better badly enough to chase it.
The interview at Unibank was two rounds. The first was HR — standard stuff. Tell us about yourself, why banking, what are your strengths. I got through that fine. Everyone can talk about themselves.
The second round was with the credit risk team. They put a laptop in front of me and asked me to look at a spreadsheet of loan applications. About 200 rows. They asked me to find patterns — which applicants were likely to default based on the data.
I didn't use any fancy tools. I sorted columns. I filtered by income brackets. I highlighted repeat addresses (a common fraud signal, I'd later learn). I grouped applicants by employment type and age range. It was basic, but I was thorough.
Then the manager — the same one who'd asked about SQL — showed me a simple query on the whiteboard:
SELECT
customer_id,
loan_amount,
monthly_income,
ROUND(loan_amount / (monthly_income * 12), 2) AS debt_ratio
FROM loan_applications
WHERE status = 'pending'
AND monthly_income > 0
ORDER BY debt_ratio DESC
LIMIT 50;
She asked me what this did. I didn't know SQL syntax, but I could read it well enough to understand: it was finding the 50 pending applications with the worst debt-to-income ratio. I told her that. She nodded.
"You can learn SQL in two weeks," she said. "Analytical thinking takes years. You already have it."
I got the offer the next day.
My first six months at Unibank were brutal. I was learning SQL on the job, which means I was writing broken queries, getting wrong results, and bothering senior analysts with questions every twenty minutes.
But banking is a forgiving environment for learners, in a weird way. The data is structured. Tables have clear relationships. Loan applications have customer IDs that link to payment histories that link to credit scores. If you can understand JOINs, you can understand banking data.
Here's the kind of query I was writing within my first month:
SELECT
c.customer_id,
c.full_name,
l.loan_amount,
l.interest_rate,
p.total_paid,
l.loan_amount - p.total_paid AS remaining_balance,
CASE
WHEN p.last_payment_date < CURRENT_DATE - INTERVAL '90 days'
THEN 'HIGH RISK'
WHEN p.last_payment_date < CURRENT_DATE - INTERVAL '30 days'
THEN 'MEDIUM RISK'
ELSE 'LOW RISK'
END AS risk_category
FROM customers c
JOIN loans l ON c.customer_id = l.customer_id
JOIN (
SELECT
loan_id,
SUM(amount) AS total_paid,
MAX(payment_date) AS last_payment_date
FROM payments
GROUP BY loan_id
) p ON l.loan_id = p.loan_id
WHERE l.status = 'active'
ORDER BY remaining_balance DESC;
Nothing fancy. But writing queries like this every day for two years built a foundation that everything else sits on. SQL is the language of data. Not Python. Not R. Not Tableau. SQL.
I worked on risk assessment models for consumer loans. That sounds impressive, but in practice it meant: pull the data, clean it in Excel, look for patterns, write a report, present it to the team. The "model" was often just a set of rules — if debt ratio exceeds 0.4, flag for review. If the applicant has more than 3 active loans, escalate.
The most valuable thing I learned wasn't a tool. It was pattern recognition. After looking at thousands of loan applications, you start to see things. Inconsistent addresses. Income that doesn't match the employer. Multiple applications from the same household within a week. I didn't know it at the time, but I was training my brain for fraud detection.
In December 2021, I was promoted to Business Analyst. Same company, bigger scope. Instead of just looking at individual loan applications, I was analyzing business processes, building reports for management, and automating things that had been done manually for years.
This is where Python entered my life. Not because anyone taught me, but because I got tired of copy-pasting data between spreadsheets. I found a tutorial on pandas, wrote a script that merged three Excel files into one, and felt like a wizard.
Within six months, I was writing ETL scripts that pulled data from the bank's databases, transformed it, and loaded it into reporting tools. The manual reporting process that took the team 4 hours every morning? I automated it down to a 15-minute script. Reduced manual effort by 60%. That number went on my resume and it wasn't exaggerated.
I also got serious about SQL during this period. Not just writing queries, but understanding stored procedures, window functions, CTEs, and query optimization. In May 2022, I passed the Oracle Database SQL Certified Associate exam (Credential ID: 290631207OCASQL12C). It wasn't the hardest certification in the world, but it formalized knowledge I'd been building for three years. And in Azerbaijan's job market, certifications matter more than you'd think.
Here's the kind of thing I was building:
WITH monthly_revenue AS (
SELECT
branch_id,
DATE_TRUNC('month', transaction_date) AS month,
SUM(amount) AS revenue,
COUNT(DISTINCT customer_id) AS unique_customers
FROM transactions
WHERE transaction_type = 'credit'
GROUP BY branch_id, DATE_TRUNC('month', transaction_date)
),
ranked AS (
SELECT
branch_id,
month,
revenue,
unique_customers,
LAG(revenue) OVER (PARTITION BY branch_id ORDER BY month) AS prev_revenue,
ROUND(
(revenue - LAG(revenue) OVER (PARTITION BY branch_id ORDER BY month))
/ NULLIF(LAG(revenue) OVER (PARTITION BY branch_id ORDER BY month), 0) * 100,
2
) AS growth_pct
FROM monthly_revenue
)
SELECT *
FROM ranked
WHERE growth_pct IS NOT NULL
ORDER BY growth_pct DESC;
Window functions changed how I thought about data. Not just "what happened" but "how does this compare to last month, last quarter, last year?" That shift — from descriptive to comparative analysis — is what separates a junior analyst from a mid-level one.
Here's the full progression, laid out honestly:
| Period | Role | Company | Key Skills Gained |
|---|---|---|---|
| Sep 2019 – Dec 2021 | Credit Analyst | Unibank | SQL fundamentals, credit risk assessment, Excel, pattern recognition |
| Dec 2021 – Feb 2023 | Business Analyst | Unibank | Python (pandas, automation), ETL pipelines, stored procedures, BI dashboards, Oracle SQL cert |
| Feb 2023 – Sep 2025 | Fraud Analyst | Kapital Bank | Machine learning (scikit-learn, TensorFlow), behavioral scoring, feature engineering, anomaly detection |
| Oct 2025 – Present | Data Analyst | Bank of Baku | Automated data pipelines, predictive modeling, query optimization, large-scale data processing |
Four roles. Three banks. Six-plus years. Each job built directly on the one before it. That's the thing about a career in data — it compounds. SQL knowledge makes Python easier. Python makes machine learning accessible. ML makes you dangerous.
Moving from Unibank to Kapital Bank in February 2023 was the biggest professional risk I'd taken. I was leaving a comfortable role where I knew everyone, where I'd been promoted, where I'd built systems that I maintained. But the Fraud Analyst position at Kapital Bank offered something Unibank couldn't: machine learning.
Kapital Bank had a growing fraud detection team. Real fraud. Credit card scams, identity theft, account takeovers, synthetic identities. The data was messier, the stakes were higher, and the tools were more advanced.
I spent my first three months just understanding the fraud landscape in Azerbaijani banking. The patterns are different from what you'd read in a US-focused textbook. The transaction volumes are smaller, the fraud types are more personal (less organized crime, more individual schemes), and the data infrastructure is less mature.
But the fundamentals are the same everywhere. You're looking for anomalies. Transactions that don't fit the pattern. Behavior that deviates from the baseline. And for that, you need models.
I learned scikit-learn first. Random forests, gradient boosting, logistic regression for classification. Then TensorFlow for more complex patterns — sequential models that could detect fraud based on transaction sequences rather than individual transactions.
Here's a simplified version of the kind of feature engineering I was doing:
import pandas as pd
import numpy as np
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report
def build_fraud_features(transactions_df):
"""Build behavioral features for fraud detection."""
features = transactions_df.groupby('customer_id').agg(
avg_transaction=('amount', 'mean'),
std_transaction=('amount', 'std'),
max_transaction=('amount', 'max'),
transaction_count=('amount', 'count'),
unique_merchants=('merchant_id', 'nunique'),
avg_time_between=('timestamp', lambda x: x.diff().mean().total_seconds() / 3600),
).reset_index()
features['std_transaction'] = features['std_transaction'].fillna(0)
features['coefficient_of_variation'] = (
features['std_transaction'] / features['avg_transaction'].replace(0, np.nan)
).fillna(0)
return features
# Train the model
features = build_fraud_features(transactions)
X = features.drop(['customer_id', 'is_fraud'], axis=1)
y = features['is_fraud']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
model = GradientBoostingClassifier(n_estimators=200, max_depth=5, random_state=42)
model.fit(X_train, y_train)
print(classification_report(y_test, model.predict(X_test)))
The coefficient of variation was one of the strongest fraud signals we found. Legitimate customers have relatively consistent spending patterns. Fraudsters don't — they either drain an account in a burst of large transactions or test it with tiny amounts before going big.
Building behavioral scoring models taught me something no online course could: domain expertise matters more than algorithmic sophistication. A simple logistic regression with the right features beats a complex neural network with bad features. Every time. I watched junior team members obsess over model architecture while ignoring feature engineering. The features are where the money is.
In April 2023, about two months into my Kapital Bank role, I started building birjob.com. It's a job aggregator for Azerbaijan — scraping listings from multiple sources, deduplicating them, and presenting them in one clean interface.
Why? Because job hunting in Azerbaijan is painful. Listings are scattered across a dozen websites, many are duplicates, half are outdated, and there's no good way to search across all of them. I'd experienced this firsthand during my own career transitions. I figured if I was frustrated, other people were too.
The data skills from banking translated directly. Scraping is just ETL with messier inputs. Deduplication is a classification problem. Building a search index is database optimization. The biggest difference was that at the bank, someone else owned the infrastructure. With birjob.com, I was the infrastructure.
I built it with Python, scraping data from job boards, cleaning and deduplicating with pandas, and storing everything in PostgreSQL. The same tools I used at work, applied to a different problem. That's the thing about data skills — they transfer. Once you can clean, transform, and analyze data in one domain, you can do it in any domain.
birjob.com taught me things no banking job ever would. Deployment. Server costs. User feedback. SEO. Marketing. The difference between building something that works and building something people actually use. It's still running, still growing, and it's become the project I'm most proud of — not because it's technically impressive, but because real people in Azerbaijan use it to find jobs.
I joined Bank of Baku in October 2025. Another bank, another data role. But the scope is different this time. I'm working on automated data pipelines, predictive models for market trends, and query optimization for large datasets.
The tools are familiar — Python, SQL, pandas — but the problems are bigger. More data, more stakeholders, more complexity. I'm also pursuing my MBA in Artificial Intelligence at Azerbaijan State University of Economics (UNEC), which started in 2024. Yes, I'm getting a master's degree while working full-time. No, I don't recommend it for your mental health.
The MBA is filling gaps. My undergrad was management from a small city university. The MBA is AI-focused, rigorous, and connected to Baku's tech ecosystem in ways Mingachevir never was. It's not necessary for the work I do — I was doing machine learning at Kapital Bank without it — but it's opening doors to conversations and networks I didn't have access to before.
Let me give you the honest picture of what this market looks like right now.
Data analyst salaries in Azerbaijan range from about 420 to 1,550 AZN per month at the 80th percentile. That's roughly $250 to $900 USD. Yes, it's significantly lower than global averages — entry-level data analyst roles globally sit around $68,000 to $81,000 per year. We're not competing on salary.
But here's the context that matters: cost of living in Baku is a fraction of US or European cities. And the growth trajectory is real. Banks are investing in data infrastructure. Fintech is growing. The government is pushing digitalization. The jobs are getting better, even if the salaries lag.
Globally, the outlook is strong. 11.5 million new data science and analytics jobs are projected by late 2026. The World Economic Forum flagged data roles as 5 of the top 15 fastest-growing job categories. But here's the catch: entry-level positions are the most impacted by increased competition. More people are entering the field, which means standing out matters more than ever.
In Azerbaijan specifically, banking remains the primary employer for data roles. If you want to be a data analyst here, learn banking. Understand credit risk, transaction data, regulatory reporting. These aren't sexy topics, but they're where the jobs are.
If I could go back to 2018 — the version of me watching Excel tutorials on YouTube in a cramped Baku apartment — here's what I'd say:
1. Learn SQL first. Not Python. Not R. SQL.
SQL and Python appear in the vast majority of data analyst job postings, but SQL is the one you'll use every single day. I still write more SQL than Python, six years in. Every database speaks SQL. Every company has data in tables. Start there.
2. Banking hires analysts without CS degrees. Go there.
I walked into Unibank with a management degree and they gave me a chance. Banks need people who understand business logic, not just people who can code. Your non-technical degree might actually be an advantage — you can translate between business and data in ways pure CS graduates sometimes can't.
3. Build projects. Put them on GitHub. That's your real resume.
3 to 5 polished portfolio projects is the sweet spot. Not 20 half-finished Jupyter notebooks. A few well-documented projects that show you can take raw data, clean it, analyze it, and present findings. birjob.com has done more for my career than any certification.
4. Get certified. The Oracle SQL cert opened doors.
In Azerbaijan's market, certifications carry weight. The Oracle Database SQL Certified Associate exam isn't the hardest thing in the world, but having that credential on my resume got me past HR filters that would have screened me out otherwise. It cost time, not money (relatively speaking), and it paid for itself immediately.
5. Don't wait to feel ready.
I wasn't ready when I applied to Unibank. I wasn't ready when I moved to Kapital Bank. I wasn't ready when I started building birjob.com. If I'd waited until I felt ready, I'd still be doing odd jobs in Baku with a management degree and no direction. Most career changers become job-ready in 7 to 12 months of focused learning. I took longer because I didn't have a plan. You can be smarter about it.
6. The market is getting more competitive. Move now.
Entry-level data roles are facing more competition than ever. Companies care more about whether you can actually do the work than about your credentials. That's both good and bad news. Good because a non-traditional background like mine is less of a barrier. Bad because you need to actually demonstrate ability, not just list skills.
Here's my position, stated plainly.
The path to a data career isn't linear. It never was, and pretending otherwise does a disservice to everyone trying to break in. A management degree from Mingachevir State University isn't supposed to lead to ML-powered fraud detection models, an Oracle certification, an MBA in AI, and a running tech startup. But it did.
Not because I'm special. I'm not. I'm persistent and I got lucky a few times. That hiring manager at Unibank who took a chance on a kid who didn't know SQL — that was luck. Everything after that was showing up and doing the work.
The only credential that actually mattered was the work I could show. Not the degree. Not the university name. Not the city I grew up in. The work. Can you write a query that answers a business question? Can you build a model that catches fraud? Can you take messy data and turn it into something useful? If yes, someone will hire you. If no, no amount of credentials will save you.
Azerbaijan is a small market. Baku is a small city by global standards. The salaries aren't competitive internationally. But the opportunities are real and they're growing. If you're sitting in Mingachevir or Ganja or Sumgait right now, reading this, thinking the path to a data career doesn't exist for someone like you — I'm telling you it does. Because I walked it. Badly, messily, with a wrong-sized suit and zero SQL knowledge.
It worked anyway.
You just have to start.