What started as a relatively simple geospatial analytics project has evolved into something much bigger: a decision-support system built around real business and financial transaction data.

​

The original question was straightforward:

​

Where are our customers actually coming from?

For a location-based business, answering that question can fundamentally change how you make decisions.

​

The project began with customer transaction data collected through Stripe, our payment processing system. Rather than treating those transactions simply as financial records, I wanted to explore what additional business intelligence could be extracted from them.

​

Using Python, Mapbox and pgeocode, I developed a pipeline that could connect customer and payment information with geographic data.

​

This allowed us to visualise where our customers were concentrated and start asking more useful operational questions:

• Where are our strongest customer clusters?

• Are our service locations positioned in the right places?

• Which geographic areas are producing customers despite receiving relatively little attention from us?

• Where might the data support opening or relocating a service location?

​

Over time, I improved the underlying pipeline, replacing fragile dependencies and developing more stable methods that didn't rely on poorly maintained online services.

​

Then AI changed how I developed the system.

I began using ChatGPT as a development tool — not to replace the analytical thinking, but to accelerate debugging, challenge existing approaches and identify alternative technical solutions.

​

This helped move the project towards a more stable and maintainable stack:

​

Python | SQLite3 | GeoPy | Pandas | Plotly | Matplotlib | Seaborn | Scikit-learn

​

Instead of depending heavily on external online services that could change, disappear or stop being maintained, more of the processing and analytics could be handled within our own pipeline.

​

After completing a BrainStation Data Science course, I expanded the project beyond geospatial analysis.

​

The next question became:

​

Location tells us where customers are — but what can their transaction history tell us about their behaviour?

Because the underlying Stripe data contained customer payment histories, I could derive variables such as customer lifetime, total spend and churn behaviour.

​

I introduced customer lifetime and churn analysis, using machine-learning techniques including logistic regression, alongside statistical data visualisation.

​

Histograms and density distributions revealed something particularly interesting:

​

Our customers weren't one homogeneous population.

​

Distinct groups emerged.

​

Some customers stayed substantially longer and generated significantly greater lifetime value.

​

Others clustered around early churn, shorter customer lifetimes and lower total spend.

​

That changes the meaning of the geographical analysis.

​

A postcode containing 20 customers isn't necessarily more valuable than an area containing 10.

​

The more interesting question becomes:

​

What kinds of customers does each area produce?

Now geographic areas can potentially be evaluated using a combination of customer concentration, retention, churn risk and lifetime value — creating a much richer picture for business decision-making.

​

More recently, I've taken the project another step forward by using Base44 to build the application and user-interface layer around the analytics.

​

This reflects something I increasingly believe about modern software development:

​

Humans shouldn't be doing work that machines can reliably do for them.

​

My primary interest is in the analytics, architecture and decision-making system. AI-assisted development allowed me to move the project towards a production-level application without spending months becoming a specialist UI developer or immediately recruiting someone else to build that layer.

​

The project has therefore evolved into an end-to-end system:

​

Stripe transaction data → Data processing → Geospatial analytics → Customer segmentation → Churn/LTV modelling → Data visualisation → AI-assisted application development → Business decision support

​

And the principle behind it remains simple:

​

Automate the stages where human involvement adds little value. Keep humans firmly in the loop where judgement and critical decisions matter.

​

For me, that's one of the most exciting applications of AI.

​

Not replacing human decision-making.

​

Building systems that give humans better information on which to base those decisions.

​

#MachineLearning #DataScience #DataEngineering #GeospatialAnalytics #Python #ArtificialIntelligence #Analytics #Automation #Stripe #BusinessIntelligence #AIEngineering #ProductDevelopment