SaaS Subscription Churn Analysis
Category
Data Analysis
/
/
This project asks a simple question: why is this subscription business losing customers, and where should it act first? I found that $2.86M in revenue is at risk from churn, and traced it down to specific, fixable causes, not just a vague "churn is high."

The Problem
Leadership knew churn was a problem but not who was leaving or why. I set out to answer three things:
Which customers are churning, and how much revenue does that cost?
Is the company spending marketing budget on the wrong customers?
What should we fix first?
The Data
7,043 customers from the Telco Customer Churn dataset (Kaggle), reframed as a SaaS business, since the billing structure is the same.
No marketing cost data existed, so I built it myself: each customer got a modeled acquisition channel and cost. This is disclosed here plainly: it is modeled, not observed, data. The comparisons across channels are reliable; the exact dollar amounts are illustrative.
One cleanup step worth mentioning: 11 customers had a blank instead of a number for total revenue. Turned out they were brand-new, not yet billed, so I left them blank rather than counting them as zero.
Every number below was calculated in SQL, then checked again independently in Excel and in Power BI. All three agreed.

What I Found
Overall, 26.5% of customers churn, worth $2.86M in lost revenue. Here's where that's concentrated.
Contract | Churn rate | Avg monthly charge |
|---|---|---|
Month-to-month | 42.71% | $66.40 |
One year | 11.27% | $65.05 |
Two year | 2.83% | $60.77 |
1. Contract type is the biggest driver
Contract | Churn rate |
|---|---|
Month-to-month | 42.7% |
One year | 11.3% |
Two year | 2.8% |
Month-to-month customers are just over half the customer base, but cause 89% of all churn, $1.93M of the $2.86M at risk. It's not about price: monthly charges are almost identical across all three contract types. It's about commitment.
2. Marketing is spending on the wrong customers
The channels the company spends the most on (Paid Search, Paid Social) bring in customers who churn the most. Referral customers return 193x their acquisition cost over time; Paid Search customers return only 13x, a 14x gap. The single biggest group of customers, 1,140 of them, were both acquired through the most expensive channel and belong to the highest-churn segment.
Internet service | Churn rate | Revenue at risk | Avg. revenue at risk per churned customer |
|---|---|---|---|
Fiber optic | 41.89% | $2,483,257 | $1,915 |
DSL | 18.96% | $360,016 | $784 |
No internet service | 7.40% | $19,653 | $174 |
3. Fiber internet is the worst segment, no exceptions
Internet service | Churn rate | Revenue at risk |
|---|---|---|
Fiber optic | 41.9% | $2.48M |
DSL | 19.0% | $360K |
None | 7.4% | $20K |
Fiber customers churn more, cost more when they leave, and make up the largest share of revenue at risk. Every measure points the same way, worth investigating directly.
4. Payment method matters on its own
Electronic check customers churn at 45%, the highest of any group, even when I compare only customers on the same contract type. So it's not just standing in for contract type, something about that payment method itself is a real, separate risk factor.
The single biggest risk group
Month-to-month customers paying by electronic check: $1.21M at risk in this one group alone, where the two worst factors overlap.
The Dashboard
A two-page Power BI report, connected live to the data.
Page 1 shows churn by contract type and payment method, plus a ranked table of the riskiest combinations.
Page 2 shows marketing spend efficiency by channel, and churn by internet service type.
Each page opens with one sentence stating the headline number and what to do about it.
What I'd Recommend
Push month-to-month customers toward annual contracts, this fixes 68% of the problem on its own.
Shift marketing budget from Paid Search/Social toward Referral and Organic, they return 4-5x more per dollar.
Look into fiber optic service directly, price or quality, since it's the worst segment across the board.
Investigate why electronic check payers churn more, it's not fully explained by anything else.
Start with month-to-month + electronic check customers, that's the single highest-risk group.

Honest Notes
The acquisition cost numbers are modeled, not real, since the dataset didn't include them. The relative comparisons still hold.
With more time, I'd dig into why fiber optic specifically churns so much, price, service, or competition, SQL alone can't tell me that.
Throughout this, I tried to disprove my own first guess before trusting it. A couple of times it held up. Once it only partly did, and I reported that honestly instead of smoothing it over.

