Articles
SaaS Product Analyst Interview Case: Explain Rising NRR and Falling Customer Retention
Work a SaaS product analyst case where NRR rises as logo retention falls, using a reconciled customer and MRR cohort bridge, GRR, and testable hypotheses.
“Net revenue retention is up, but customer retention is down.” A weak interview response calls the dashboard contradictory. A stronger response notices that the two measures count different things: one follows dollars from an existing customer cohort, while the other follows the number of customers in that cohort. The numbers can move in opposite directions without an arithmetic error. The analyst’s job is to prove which customer movements produced the divergence, then decide whether the expansion is durable enough to offset a leaky base.
Source-backed definition: ChartMogul’s NRR documentation counts expansion and losses from existing subscribers; its customer-retention definition counts retained paying customers. Candidate reports: this is an original practice case, not a report of a named employer’s interview. Inference: the diagnostic order below is an editorial method for answering the case; no vendor source claims it is the only acceptable interview answer.
Start with the metric contracts, not a story
Ask for the exact observation window, the starting paying-customer cohort, and how revenue is normalized. “NRR rose” could mean monthly, quarterly, or trailing-twelve-month NRR. “Customer retention fell” could use a different start date, count accounts rather than subscriptions, or include trials. A plausible causal explanation is premature until those choices align. Compare the same product, currency policy, region, and cohort duration.
For a fixed starting group, customer or logo retention is retained paying customers divided by starting paying customers. New customers acquired later are not in the numerator. A company can add more new logos than it loses and grow its total customer count while its existing-cohort retention still falls. Likewise, one legal entity with many workspaces can look like many customers if account identifiers are not consolidated consistently.
A useful NRR contract is: (starting recurring revenue + expansion from starting customers − contraction − churn) / starting recurring revenue. Some subscription systems include reactivation from the starting cohort in their NRR movement bridge; ChartMogul’s report explicitly does. In the exercise below, reactivation is zero, so the simpler bridge is exact. New-customer revenue must be excluded. State whether one-time services, credits, usage true-ups, refunds, and currency movements enter recurring revenue. This is where two internally valid dashboards can disagree.
Gross revenue retention, or GRR, removes expansion from the bridge. It therefore answers a different question: how much starting revenue survived churn and contraction before upsell covered losses. ChartMogul’s GRR description provides that contrast. You should calculate logo retention, GRR, and NRR together; a high NRR alongside falling GRR reveals a different quality of growth from high NRR with stable GRR.
Work through an internally consistent case
Original practice data: compare two independent twelve-month starting cohorts, A and B. Each starts with 100 paying customers and $10 million in normalized monthly recurring revenue. Cohort A retains 90 customers, loses $0.8 million to churn and contraction, and gains $1.3 million from expansion among survivors. Cohort B retains only 80 customers, loses the same $0.8 million of recurring revenue, and gains $2.5 million from expansion. Assume no reactivation, foreign-exchange movement, or new business in either bridge.
| Twelve-month cohort | Starting customers | Retained customers | Starting MRR | Churn + contraction | Expansion | Logo retention | GRR | NRR |
|---|---|---|---|---|---|---|---|---|
| A | 100 | 90 | $10.0m | $0.8m | $1.3m | 90% | 92% | 105% |
| B | 100 | 80 | $10.0m | $0.8m | $2.5m | 80% | 92% | 117% |
For B, the dollar bridge is $10.0m − $0.8m + $2.5m = $11.7m; $11.7m / $10.0m = 117%. The customer bridge is 80/100 = 80%. For A, $10.0m − $0.8m + $1.3m = $10.5m, or 105% NRR, while 90/100 = 90% customer retention. The figures prove that the two directions are mathematically compatible. They do not prove why ten more customers left B, or whether the expansion is healthy.
The unchanged $0.8m loss with twice as many B customer exits suggests—only as a hypothesis—that B’s additional churn came disproportionately from smaller accounts. There could be other mixes: a large customer contracted less, the churners’ average starting MRR differed, or expansion happened in a handful of enterprises. Do not announce “small customers churned” as a fact until an account-level bridge shows it. Show the distribution of starting MRR among churners and survivors, not just an average across all logos.
A second issue is concentration. If the extra $1.2m of expansion comes from one customer, B’s NRR may be exposed to that account’s renewal, usage volatility, or one-time contract normalization. If it comes from forty customers renewing after reaching a product milestone, the pattern has a different risk profile. Ask for the number of expanding accounts, expansion by decile of starting MRR, the top-one/top-five share of expansion, and the portion tied to contracted versus metered usage. These requests convert a narrative into testable alternatives.
Build two bridges from the same frozen cohort
The cleanest diagnostic asset is a paired bridge. Freeze the customer IDs active at the first timestamp. For each ID, store starting MRR, ending MRR, customer status, segment, plan, first-paid date, and relevant price/usage changes. Then classify each account’s movement: churned to zero, contracted, unchanged, or expanded. A reactivated account needs a written convention. The total of those movements should reconcile to the starting cohort’s ending MRR. Separately count how many starting IDs remain paying at the end.
Do not build logo retention by subtracting all cancellation events. A customer may cancel one subscription while keeping another, or cancel and reactivate before the end date. Do not build NRR by adding all expansion events across the company: mid-period new customers are outside the starting cohort. These are common numerator errors. The paid subscriber retention report definition is a useful reminder that a subscriber who reactivates within the period may be handled differently from a raw cancellation event count.
Use a reconciliation table before interpreting. Starting cohort IDs plus their ending-state classification must equal starting cohort IDs; no customer can fall into two mutually exclusive end states. Starting MRR plus net movement must equal end MRR for the frozen IDs, allowing only explicitly listed adjustments. Compare the finance recurring-revenue ledger against the product analytics customer identity table. Differences caused by currency rates, backdated invoices, credits, and parent-child account changes should be disclosed rather than buried in “other.”
Which hypotheses deserve the first cuts?
Once the bridge is sound, segment by starting customer size, contract age, acquisition channel, country, product, sales-led versus self-serve, and renewal month. Keep the segment based on attributes known at cohort start when possible. Segmenting by ending plan can create survivorship bias: churned accounts have no ending plan, and expanded accounts migrated into high-price plans because of the outcome being studied. Show both start and end attributes when a migration itself is the hypothesis.
Several mechanisms can produce B’s pattern. A pricing increase can expand MRR among survivors while raising churn among price-sensitive accounts. Metered usage from a few large customers can lift NRR without any change in product satisfaction for smaller accounts. Sales may have acquired many low-fit small accounts in the previous year; their scheduled churn appears now, while enterprise expansion masks it. An accounting change might reclassify usage as recurring or shift the measurement window. A product rollout may genuinely deepen adoption in one segment and worsen onboarding in another. These are hypotheses, not conclusions from two percentages.
Choose cuts that would distinguish them. For pricing, compare affected and unaffected plans around notice and renewal dates, then inspect discounts and concession credits. For usage expansion, decompose price versus quantity and check whether contract minimums or temporary workloads drove the increase. For low-fit acquisition, compare cohorts by signup month and source with equal exposure time. For product quality, look at activation, feature use, support severity, and renewal outcomes. This is not a license to call every correlation causal; it is a way to prioritize the next measurement.
A strong interview answer also asks whether the trend is statistically or operationally meaningful. If there are only ten customers, losing one changes logo retention by ten points; if there are ten thousand, the same percentage change carries a different precision. Show counts, not only rates. Check whether the 117% NRR figure is concentrated in a small denominator or distorted by timing. Compare several cohorts and disclose uncertainty before recommending a companywide change.
What decision would you recommend?
Do not say “optimize NRR” as if the metric were an action. In the B case, first protect the expansion that is real and repeatable, then investigate the extra ten logo losses. If they are low-value trial-like accounts acquired through a channel with poor fit, the remedy may be targeting or onboarding rather than indiscriminate discounting. If they are strategically important customers or early signs of churn moving upmarket, the risk is larger than the $0.8m loss suggests. Revenue-weighted and logo-weighted views both matter.
Propose an intervention tied to a mechanism. For example: if first-month activation is weak in a specific self-serve segment and predicts twelve-month churn, test a guided setup experience on new eligible accounts. Primary outcome: paid cohort retention at a fixed horizon. Guardrails: gross retention, support burden, expansion MRR, and acquisition mix. Do not use total company NRR as the sole outcome of a short onboarding test; existing enterprise expansion could overwhelm the signal.
For expansion concentration, ask customer success to validate whether the largest expanding accounts have durable usage and upcoming renewals. For pricing, compare value delivered, margin, and churn by plan before changing rates. A hypothetical “save every churner” offer can destroy contribution margin; customer lifetime value and acquisition cost matter. State which information would change your recommendation. That is a better case answer than declaring a growth rate good or bad in isolation.
An interviewer may ask whether 117% NRR is “healthy.” The answer is conditional. NRR above 100% means ending recurring revenue from the specified starting cohort exceeds its starting recurring revenue under the chosen definition. It does not prove broad customer satisfaction, profitable growth, low concentration, or a healthy acquisition funnel. Falling logo retention is a reason to inspect the distribution and mechanism even when the revenue bridge is positive.
A two-minute answer you can practice
“I would first confirm that both rates use the same starting paying-customer cohort, period, account identity, and recurring-revenue definition. Customer retention counts survivors; NRR weighs the same cohort’s ending revenue, including expansion and excluding new business. In a simple bridge, 100 customers and $10m can become 80 retained customers and $11.7m from $2.5m expansion minus $0.8m churn and contraction. That gives 80% logo retention and 117% NRR, so the numbers can coexist.”
Then continue: “I would reconcile account-level movements, calculate GRR, and cut the change by starting account size, channel, plan, and renewal cohort. My leading hypotheses are expansion concentrated in large survivors, extra churn among smaller accounts, or a pricing/definition change; I would test those rather than assume them. The action depends on whether expansion is durable and which customers are leaving.”
Practice that bridge with Compute churn and revenue churn in SQL. Your strongest answer will tell the interviewer exactly what the two metrics measure, show numbers that reconcile, and identify the next data that could prove your explanation wrong.