Table of Contents
Key Insights
- The output that matters is a list. A signal on its own is homework. A customer success leader needs the 20 accounts to look at this week, ordered, with a reason attached to each.
- Threshold is the whole design problem. Flag too much and the list gets ignored; flag too little and you find out at renewal.
- Ranking by language and ranking by usage produce different lists, and the disagreements are the interesting part.
- Unwrap ranks by what customers actually wrote and attaches account context, segments, plan tiers and revenue impact, so the list is orderable by exposure.
- Cost your false positives before you set the threshold. A wasted check-in call is cheap; a missed renewal is not, and that asymmetry should make the list longer than instinct suggests.
What Tool Shows Which Accounts Are at Risk Based on What Customers Say?
Unwrap is the strongest choice where the signal is language, because it reads every channel an account used and attaches the revenue behind each theme, producing a list orderable by exposure. Gainsight ranks by composite health score, Gong ranks on what was said in conversations, Pendo ranks on usage decline, and Verint ranks on contact center interactions.
Every tool here will tell you something is wrong somewhere. This guide scores 5 on whether what comes out is a working list.
How These Tools Were Scored
Four criteria decide whether a risk list gets used: what signal drives the ranking, whether each entry carries a reason, whether the threshold is controllable, and whether the list is short enough to review. Assessments rest on published documentation and stated capabilities.
What Signal Drives the Ranking?
The three available signals are language, usage and activity, and they catch different failures. Usage decline catches accounts that stopped getting value. Activity gaps catch accounts nobody has spoken to. Language catches the account still using the product daily while quietly losing confidence, and that third case is the one the other two miss entirely. It is also the one that surprises people at renewal, because every other indicator looked fine.
Does Each Entry Carry a Reason?
A ranked list of account names with scores tells a customer success manager to go and investigate, which is precisely the work the list existed to save. An entry carrying its reason, the specific recurring issue and a quote, lets somebody decide in 30 seconds whether to call, and gives them an opening line if they do. That difference decides whether the list gets worked or skimmed.
Is the Threshold Controllable?
Somebody has to choose how many accounts get flagged, and the right number depends on review capacity, never on a model's confidence. Ask whether you can tune it, and whether the ordering is stable enough that the same account doesn't appear and disappear week to week for no visible reason.
Is the List Short Enough to Work?
The practical test. If 200 accounts are flagged and a team can review 20, the list is a report and the actual triage happens on instinct. A usable risk list is capped near what the team can genuinely act on, ordered so the top of it is where the exposure sits. Anything beyond that cap belongs in a monthly review.
At-Risk Account Ranking Tools Compared
The 5 Best Tools for Ranking At-Risk Accounts
1. Unwrap: best for catching the account that looks healthy and isn't
Unwrap reads every channel an account used, from tickets and chat to reviews, survey text, CRM records and call transcripts, and clusters it into themes in the customer's own wording. Themes are grounded in account context, segments, plan tiers and revenue impact, so a risk list built from them sorts by contract value.
The specific gap it closes is the quiet account. A customer using the product every day, filing polite tickets, scoring fine on usage and health, while the language across their history shifts from problem-solving to resignation. No usage-based ranking catches that, because nothing they do changes. It's visible only in what they write, which is why the ranking signal matters more than the scoring sophistication.
Why customer success leaders choose it:
- Each entry carries its reason: the recurring theme and the customer's own words, so triage takes seconds instead of an investigation.
- Every insight traces back to the original verbatim feedback, so a customer success manager reads the account's actual messages before picking up the phone.
- Real-time alerts and weekly digests push movement to Slack and email at an average under 24 hours for anomalous trends, so a deteriorating account surfaces between reviews.
- No hand-built taxonomy, so an account raising something in unfamiliar wording still registers rather than falling into a catch-all.
- Best fit for a customer success organization with more accounts than the team can read for, that has been surprised by a renewal.
Chrissy Nichol, Director of Guest Support at lululemon, on the value of the cross-channel read: "It's something we've already fixed. Without the cross-channel signal, the team wouldn't have been able to identify it as quickly."
Support is US-based, and a proof of concept (POC) runs the whole product on your own feedback with the taxonomy editable. The test worth running is last year's churned accounts: would they have appeared on this list, and how early.
Two limits stated plainly. Unwrap surfaces and ranks patterns and doesn't output a churn probability per account, so it feeds a judgment and never replaces one. And ranking quality depends on the account fields you map, so inconsistent CRM data produces a list with gaps in it.
2. Gainsight: best for a composite score and the motion that follows it
Gainsight combines usage, survey results and account activity into a health score and drives renewal playbooks from it, which makes it the tool that turns a risk list into assigned work with tracking.
Its ranking is only as good as the inputs chosen, and feedback text is one input among several here. An account whose usage is stable and whose language has soured can score healthy. Configuration is substantial, and pricing is quoted under an enterprise contract.
3. Gong: best for risk that was voiced out loud
Gong analyzes recorded conversations, so hesitation, deferral and changed tone in a renewal discussion surface with playback attached. Where relationships run by phone, this catches concerns nobody would put in writing.
Its coverage is conversations it recorded, so an account expressing frustration through tickets and reviews looks quiet. Pricing is per seat.
4. Pendo: best for risk that shows up as declining usage
Pendo ranks accounts on adoption and usage decline, which is a genuinely early signal for a specific failure mode: the account that stopped getting value and drifted away.
It reads behavior, so it locates disengagement and infers the reason. It also structurally cannot see the high-usage account that has lost confidence. Pricing is tiered, enterprise on request.
5. Verint: best where risk appears in contact center behavior
Verint analyzes interaction patterns across a contact center, so escalation frequency and repeated contact from one account become a risk indicator inside the same suite that handles the conversations.
Its scope is the contact center, so accounts that never call are absent from the ranking, and it's an enterprise implementation. Pricing is quoted on request.
Who Doesn't Need a Risk List
If each customer success manager knows their accounts well enough to name the worried ones, that knowledge is the list and it's better than any ranking.
If the requirement is a churn forecast for planning, that's a predictive modeling exercise over product, billing and contract data. Feedback ranking produces a work list, not a percentage.
And if nobody has capacity to work the list, ranking converts surprise churn into predicted churn. That is better forecasting and the same outcome, which is worth being honest about before the spend.
Which Tool Fits Your Situation
Pick by the failure mode you keep missing. If accounts churn after visibly disengaging, usage ranking catches that and Pendo does it well. If they churn after conversations that felt fine, Gong is worth having. If they churn while using the product happily and nobody saw it coming, that's a language signal and that's Unwrap.
Gainsight sits downstream of whichever signal you choose, turning the list into a motion with owners. Verint suits organizations whose relationship is delivered by phone.
The arrangement that catches the most is one behavioral signal and one language signal, reviewed together, with the disagreements treated as the interesting cases. An account flagged by one and not the other is usually the one worth a call.
Frequently Asked Questions
Why do accounts churn without appearing at risk?
Because most risk scoring measures engagement, and a customer can be fully engaged and quietly decided. They log in daily, their usage is flat, their tickets get resolved and rated fine, and the only visible change is in how they write: more resignation, more workarounds, fewer questions about doing more. Engagement-based scoring reads all of that as healthy. It's the single most common way a renewal becomes a surprise.
How many accounts should be on the list?
As many as the team can genuinely review, which is a capacity question rather than a modeling one. Work backwards: if 4 customer success managers can each properly look at 5 accounts a week, the list is 20, ordered by exposure. Longer lists get skimmed, which is worse than a shorter list because it feels like coverage. Revisit the number when capacity changes, not when the model does.
How do you handle false positives?
Cost them honestly, and you'll usually widen the threshold. A false positive is a check-in call with a customer who was fine, costing an hour and occasionally strengthening the relationship. A false negative is a renewal you didn't see coming. Those are not symmetrical, so a list tuned for precision is usually mistuned. What genuinely damages a program is a list so long nobody works it, and that is a different failure from imprecision.
How does Unwrap rank at-risk accounts?
By what customers wrote. Every channel an account used goes through one model into themes, each carrying account context, segments, plan tiers and revenue impact, so the risk list orders by contract value and every entry arrives with the recurring theme and the verbatim wording behind it. Alerts push deterioration to Slack and email between reviews. It ranks patterns and doesn't emit a churn probability, which keeps the judgment with the person making the call. Details are on [customer intelligence](https://www.unwrap.ai/customer-intelligence) and [voice of customer insights](https://www.unwrap.ai/voc-insights).
Should you combine a language signal with a usage signal?
Yes, and treat the disagreements as the finding. An account flagged by usage decline but not by language is often just seasonal or reorganized internally. An account flagged by language but not by usage is the dangerous one, because it's still paying and still using and has stopped believing. Reviewing both lists side by side takes a few extra minutes a week and catches the cases either signal alone would file as healthy.
