3,000 accounts, no early warning system

HackerRank's Customer Success org kept getting surprised by cancellations.

With more than 3,000 business customers, CSMs had no reliable health score and no usage dashboard that spanned the book. CS leader Alaina Loori put it plainly: they did not have an effective way to track and evaluate the health of all those accounts.

Clients who looked happy on the last NPS or the last meeting still churned. That is the heart of surprise churn. Scale without a radar turns every renewal into a coin flip you only notice when it lands tails.

Timeline

Reactive mode. CSMs logged into individual accounts to spot-check usage. The team waited for customers to raise issues. There were no automated risk alerts and no clear baselines for healthy usage, so gradual drop-offs never rang a bell. A positive last conversation could sit on top of weeks of fading product activity.

The cost. Lost ARR on accounts that might have been saved. Missed expansion from accounts that were already cooling. Slipping net retention. Confidence in CS took a hit every time a "fine" account cancelled. The board does not care that someone felt busy if the book still leaks.

The build. HackerRank invested in product usage analytics and automated health scoring. Risk alerts surfaced accounts that were fading. CSMs moved from technical firefighting toward earlier, preventative outreach. The job shifted from "answer the ticket" to "act on the queue."

After. Churn among at-risk accounts dropped by about 50%. The team engaged earlier in the renewal cycle instead of discovering dormancy at cancel time. Millions in ARR sat behind that shift because surprise cancellations stopped being the default discovery mechanism.

What they missed

Scale without a shared definition of healthy.

Manual spot-checks cannot cover thousands of accounts. A positive last meeting can hide silent churn. Without baselines, a slow fade looks like normal variance until the renewal fails. Support that waits for the customer to shout will always miss the accounts that go quiet on purpose or by neglect.

The team was not lazy. The system asked them to do impossible coverage by hand.

The lesson

Early warning is a product of data habits, not hero CSMs.

HackerRank's results tracked four shifts:

1. Define health from usage, not vibes. Scores need live product signals, not quarterly memory or a warm meeting note.
2. Alert before renewal season. At-risk accounts need attention while there is still time to save, not in week 51.
3. Trade spot-checks for a queue. One ranked list beats 3,000 random logins and a hope that someone picks the right account.
4. Make CS strategic earlier. Outreach months before renewal beats a rescue call after the customer has already decided.

A customer health score is only useful if it updates often enough to catch the fade. The save still needs a human and a playbook. See how to reduce SaaS churn when the alert fires.

Why it felt healthy

At 3,000 accounts, "we check when we can" sounds responsible and is mathematically empty. Spot-checks create random confidence. You inspected the accounts you had time to open. You missed the ones that needed you most. A warm meeting or a decent NPS score then freezes the story in place while usage keeps sliding.

HackerRank's CSMs were stuck in a support posture: wait for the customer to declare a problem. That works for broken features. It fails for quiet disengagement. Customers who have already decided to leave often stop complaining first. The absence of tickets becomes a fake green light.

The 50% drop in at-risk churn after the radar shipped is the proof that foresight compounds. When risk is visible 90 days out, CS can run a real save motion. When risk is visible at cancel week, CS can only apologize and discount. Same team. Different clock.

If you manage thousands of accounts without a shared early warning list, you are choosing surprise as a process. That choice compounds every quarter until the board asks why retention is the silent tax on an otherwise fine sales story.

Try this Monday

1. Pick one usage metric you already have (weekly active users, key feature events, or last login).
2. Set a simple baseline per account (median of the last 8 weeks).
3. List accounts more than 30% below their own baseline for two weeks running.
4. Assign owners and a next step for the top ten by ARR before Friday. Write the step down. Do not leave it as "monitor." If there is no next step, there is no save motion.

3,000 accounts without an early warning system is a cancel factory with a delay. HackerRank cut at-risk churn in half when the radar came online. Build yours before the next "we thought they were fine" postmortem. Start with one metric and one owner list. Expand from there.