The Leaky Bucket: Diagnosing SaaS Churn & Growth by Maruffuzzman TanvirThe Leaky Bucket: Diagnosing SaaS Churn & Growth by Maruffuzzman Tanvir

The Leaky Bucket: Diagnosing SaaS Churn & Growth

Maruffuzzman Tanvir

Completed work

Business Analyst

Data Analyst

Data Visualizer

Microsoft Excel

Microsoft Power BI

Microsoft SQL Server

Computer Software

The Leaky Bucket: Diagnosing SaaS Churn & Growth

A SQL + Power BI deep dive into a SaaS subscription business's MRR movement, retention, cohort behavior, customer value, and churn risk over a 37-month, 9,581-account history.

Thesis

Growth numbers alone don't tell you if a SaaS business is healthy, a company can add customers every month and still be losing ground if churn is draining revenue faster than new business fills it back in. This project uses 12 SQL questions across 5 categories to find out exactly where this business's growth was real, where it was masking a leak, and where the risk sits right now, how fast is revenue actually growing, how much of it survives, which cohorts and segments are driving the loss, and which currently active accounts look like the ones that already left.

The Analysis

Section 1: MRR Movement & Waterfall

The starting point, how much recurring revenue exists, and exactly where it's coming from and going to each month.

Q1. Monthly MRR & ARR

What I did:

Reconstructed each account's current MRR as of every month from the raw event ledger, since each row is a movement event, not a snapshot

Rolled up to total MRR and ARR per month, plus month-over-month growth rate

Major finding:

MRR grew from $5.43M to $333.68M across 37 months, roughly 61x, ending near $4.0B ARR

Q2. MRR by Movement Type

What I did:

Grouped monthly MRR deltas by movement type: new, expansion, contraction, churn to build the waterfall breakdown

No reactivation events exist in this dataset, so the waterfall is four-part rather than the usual five

Major finding:

$425.4M in gross additions vs. $91.7M in gross losses and churn alone accounts for 73% of everything lost

Q3. Net New MRR & Inflection Points

What I did:

Calculated net new MRR per month from the Q2 breakdown

Ranked months by swing size to flag the strongest and weakest periods

Major finding:

No month in the entire dataset went negative every single month grew net-net, just at very different speeds

Section 2: Retention

Growth means nothing if it isn't kept. This section isolates how much revenue survives on its own, separate from new business.

Q4. Gross Revenue Retention (GRR)

What I did:

Isolated the cohort of accounts active at the start of each month

Calculated the % of their revenue kept, excluding new accounts and expansion entirely

Major finding:

GRR holds in a 90–98% band across the full period, averaging 94.2% stable, not declining

Q5. Net Revenue Retention (NRR)

What I did:

Repeated the GRR calculation, this time adding back expansion revenue from the same existing-account base

Major finding:

NRR beats 100% in 35 of 36 months, averaging ~108% expansion consistently outpaces what's lost to churn and downgrades

Section 3: Cohort Retention

Retention averages hide a lot. This section tracks individual signup cohorts over time to see whether that pattern holds, and where it breaks.

Q6. Logo Retention by Cohort

What I did:

Grouped accounts by signup month and tracked what % of the original cohort was still active at 1, 3, 6, and 12 months out

Major finding:

Month-1 retention sits at 92-100%, but the steepest drop happens between month-1 and month-3 churn risk is heavily front-loaded, not a slow long-term fade

Q7. Revenue Retention by Cohort

What I did:

Repeated the cohort tracking, this time following revenue instead of headcount, letting values exceed 100% where expansion outweighs churn

Major finding:

Every mature cohort's month-12 revenue sits above 100% of where it started, ranging 122-187% the accounts that stay expand enough to cover the third that leave

Q8. Best vs. Worst Cohort

What I did:

Ranked cohorts by month-12 revenue retention, then broke down the top and bottom cohort's account mix by plan, region, and industry

Major finding:

The best and worst cohorts had nearly identical logo retention the 65-point NRR gap between them came down to how concentrated their revenue was in a handful of large accounts, not who stayed

Section 4: LTV & Efficiency

Not every customer is worth the same. This section puts a dollar figure on retention and checks whether growth is still efficient as the business scales.

Q9. Average LTV by Plan

What I did:

Calculated lifetime value per plan tier using MRR divided by that tier's churn rate, since no margin data exists to build a cost-based LTV

Major finding:

Enterprise LTV ($14.33M) outweighs Starter ($3,369) by roughly 4,250x losing one Enterprise account costs as much lifetime value as losing hundreds of Starter accounts

Q10. Quick Ratio

What I did:

Calculated (new + expansion MRR) ÷ (contraction + churned MRR) per month to measure growth efficiency over time

Major finding:

The ratio has never dropped below 1.0, but declined from the 50s early on to a 2-5x range recently a sign of a maturing base, not a warning sign

Section 5: Segmentation & Risk

Where exactly is the leak, and who's at risk of adding to it right now?

Q11. Churn Rate by Segment

What I did:

Calculated churn rate across every plan, region, and industry combination, then rolled up to single dimensions to compare their explanatory power

Major finding:

Plan tier explains a 24-point churn spread; region and industry each explain under 3 points the leak is the Starter tier specifically, not any one region or industry

Q12. At-Risk Watchlist

What I did:

Defined a risk profile from historically high-churn segment combinations, then filtered currently active accounts matching that profile, ranked by MRR

Major finding:

1,948 active accounts about 20% of the base carry $3.1M in at-risk MRR, with the top 10 highest-value accounts all sharing the same combo: Business / North America / Fintech

Full Findings

The summary above only covers the headline finding for each question. For the complete breakdown, including secondary numbers and caveats not shown here, see insight.md.

SQL Queries

All 12 questions are written as standalone .sql files, grouped into 5 section files. Full logic and inline comments are documented directly in the files.

📁 sql_queries/