Case study · 01 - Enterprise SaaS

Renewal Operations Transformation

$140M ARR enterprise SaaS · 1,900 customers · 34-person post-sales org

A post-sales org flying blind between QBRs got a nightly, evidence-backed read on all 1,900 renewals - and started catching churn while there was still time to save it.

Company profile

Revenue: $140M ARR
Employees: 640
Customers: 1,900 enterprise & mid-market
Post-sales org: 22 CSMs · 8 AMs · 4 RevOps
Motion: High-touch CS, quarterly QBRs
Retention: 88.7% GRR · 103% NRR

Stack:

Salesforce

Gainsight

Gong

Zendesk

Amplitude

Slack

Snowflake

12 sources connected in total, including billing, product telemetry, and the support escalation queue.

The challenge

The average enterprise customer was worth $74k ARR on an annual renewal. Gross retention had slipped under 89%, and nobody could say with confidence which accounts were actually at risk - the health score said one thing, the CSMs' gut said another, and the churn post-mortems kept siding with the gut.

61%

of churned ARR in the prior fiscal year was marked green in the health score 90 days before the termination notice arrived.

54 days

median staleness of health scores, which were hand-updated by CSMs on a quarterly cadence - usually right before the QBR they were meant to inform.

40+

accounts per CSM, with roughly 9 hours a week going to assembling QBR decks by hand instead of working the risk the decks described.

6 systems

held the real risk signals - usage in Amplitude, sentiment in Gong, escalations in Zendesk, commercials in Salesforce, chatter in Slack, invoices in billing - and no two of them were ever joined.

Accounts didn’t churn because nobody cared. They churned because the first credible risk signal arrived attached to the termination notice.

Residency, week one

A Revllama forward-deployed engineer embedded with the post-sales org: shadowed four CSMs and the VP of CS through a full QBR prep cycle, read eight quarters of churn post-mortems, and mapped every renewal workflow from signal to save-plan.

1
The health score was a lagging vanity metric. It correlated with churn only after the save window had already closed - it described losses, it never predicted them.
2
The signals CSMs actually trusted - champion silence, ticket-severity mix, seat-utilization slope, invoice disputes - all existed somewhere in the stack. Eleven distinct signals in daily use by the team; zero of them present as fields in the CRM.
3
“At risk” had five competing definitions across CS, AM, finance, and the board deck. Every escalation conversation started with an argument about the denominator.
4
A draft rule set built from the team’s own tribal knowledge, backtested against the prior eight quarters, caught 8 of 11 enterprise churns with 90+ days of lead time.

What we deployed

The engagement codified 62 renewal rules - what at-risk actually means, per segment, with thresholds the VP of CS signed off on - and put four agents on top of them.

Renewal Risk Agent

Scores all 1,900 renewals nightly against the codified rule set. The moment an account’s evidence turns - usage slope, champion contact, ticket mix - it opens a save-plan with the receipts attached and routes it to the owning CSM.

Deal Risk Agent

Watches open expansion opportunities for the same failure patterns the sales org sees - champion silence, stalled stages, competitor mentions - so upsells stop dying quietly in parallel with the renewal.

Whitespace Analysis Agent

Benchmarks every healthy account against its peers - products owned, seats deployed, usage depth - and hands AMs a ranked expansion play instead of a hunch.

QBR Brief Agent

Custom-built for this team: assembles the QBR deck from live data - usage story, support history, roadmap asks, commercial position - so CSMs walk in prepared instead of spending the week preparing.

sources connected

renewal rules codified

1,900

accounts scored nightly

Everything runs behind the org’s MCP hub: CSMs and leadership pull any account’s full story from Claude or Slack - same permissions, same audit trail as the agents themselves.

The rollout

Week 1

Connect & map

12 sources wired into the Revllama Context Engine. Operating map of the post-sales org delivered: every renewal workflow, where it breaks, what’s worth automating.

Week 2

Codify & backtest

62 rules codified and run in shadow mode against eight quarters of historical churn. Backtest caught 8 of 11 churns with 90+ days of lead time - reviewed line by line with the VP of CS before anything went live.

Weeks 3-4

Renewal Risk Agent live

Nightly scoring in production, alerts in Slack. Save-plans drafted for 14 accounts in the first two weeks; 84% accepted by CSMs with light edits.

Weeks 5-6

Expansion & QBR motion

Whitespace and QBR Brief agents live. Deck-assembly time dropped from ~9 hours to under 2 per account, with the saved hours redirected to working flagged accounts.

Results

$1.2M

at-risk ARR surfaced

in the first 60 days, before the QBR cycle would have caught any of it

94 days

median risk lead time

up from ~21 days under the hand-updated health score

+3.1 pts

gross revenue retention

sustained across two consecutive quarters

7 hrs/wk

returned per CSM

QBR prep and account research now assembled by agents

Value delivered

$1.9M-$2.6M year-one value

$1.1M-$1.6M

Saved renewals

Matched-cohort comparison: flagged-and-worked accounts vs comparable unflagged accounts from the prior year. Only closed-won saves with a documented save-plan counted, discounted to 70% for year-one attribution.

~7,700 hrs/yr

CSM capacity returned

7 hours x 22 CSMs x 50 weeks, valued at loaded cost - not at opportunity value, which would be higher.

$1.4M

Expansion pipeline sourced

Whitespace plays accepted and worked by AMs, credited at a 25% expected close rate against the team’s historical expansion win rate.

All estimates discounted for year-one realization. No credit taken for second-order effects.

Beyond the revenue org

The Revllama Context Engine now feeds teams that were never in scope:

Finance consumes the risk feed directly for revenue forecasting, eliminating a two-week lag in churn-allowance updates.
Product receives a weekly digest of feature gaps cited in at-risk accounts, ranked by ARR exposure.
Neither effect is included in the value figures above.

Company details anonymized. Figures are representative of the engagement pattern shown, pending customer-cleared publication.