The account was green for eleven months. Logins were steady, the NPS response was a 9, the CSM's sentiment field said "positive." Then the renewal notice came back unsigned, and the post-mortem found what the score had missed: the executive sponsor left in March, the admin who replaced her filed three tickets about the same integration, and seat usage on the second team fell by half. Every one of those events was recorded somewhere. None of them moved the number. (An illustrative scenario, but most CS leaders will recognise it.)
That is usually when a CS leader starts comparing churn prediction software. Buy a platform and you commit to months of implementation and CSM adoption. Rebuild the score and you might fix it in a month, or rebuild the same blind spot with new weights. This guide helps you tell which situation you are in.
SaaS Capital's 2025 survey also put median net revenue retention at 101%, so the median company barely replaces what it loses. The benchmark excludes companies under $1M ARR and breaks retention down by contract value: companies with ACVs above $250,000 show 95% median gross retention, those below show about 91%. The Bain figure is old and spans industries, so treat it as direction, not a forecast.
How we evaluated churn prediction software
VANDFORT publishes this guide, and VANDFORT builds the Churn Signal Watchtower, a custom churn-risk system, so we are one of the options here and have a stake in the outcome. To keep the comparison fair, we fixed five criteria before reviewing any vendor and applied them to every option, including our own.
| Criterion | What we look for |
|---|---|
| Signals it can read | CRM and product usage only, or also emails, tickets, calls and stakeholder changes. |
| How risk is scored | Weighted average of inputs, or individual signals that can raise risk on their own. |
| Explainability | Whether a CSM can see which signal moved an account, and when, without asking an admin. |
| What happens after the alert | Playbooks, tasks and owner assignment, or only a colour on a dashboard. |
| Fit and pricing model | Team size, implementation effort, and published price or pricing model, as of October 2026. |
We weight the second and third criteria most heavily: a score that averages ten inputs can hide the one that matters, and a score nobody can explain gets ignored. Vendors are described only from their own product, pricing and documentation pages or dated independent reporting, as of October 2026. Naming a vendor is not an endorsement, and we don't rank anyone first.
The churn prediction tools compared
| Option | Best for | Strengths | Limits | Pricing model, as of October 2026 |
|---|---|---|---|---|
| Gainsight CS + Staircase AI | CS orgs that want a full platform plus AI on communications | Scorecards; Staircase AI reads emails, tickets, chats and meetings for risk | Two products to scope; heavier implementation | Pricing not published; Essentials and Enterprise packages |
| ChurnZero | Mid-market CS teams that want scores tied to automated plays | ChurnScores by segment from usage, support, NPS and journey data | Score quality depends on the metrics you choose | Pricing not published |
| Vitally | Teams that want a no-code score builder across touch models | Configurable weights, real-time alerts, playbooks | Weighted scores can still average away a single serious signal | Pricing not published; three plans |
| Planhat | Teams that want to test several score designs | Health Laboratory, cohort scores, alerts to Slack and Teams | No AI scoring described on its health page | Pricing not published |
| Totango (now Odie) | Enterprise CS teams already on Totango or Catalyst | Enterprise-hierarchy health scoring; new AI platform | Rebranding and platform change under way | Pricing not published |
| HubSpot health scores | Teams already on Service Hub Professional or Enterprise | Already licensed; event and property groups with time decay | Rules you configure; no AI score described | Included in Service Hub Professional and Enterprise (seats required for some features) |
| Health score rebuild | Teams with good data and a score that fails for known reasons | Fast, cheap, uses data you already own | Needs someone to own and re-test it | Internal time |
| VANDFORT Churn Signal Watchtower | $3M to $30M ARR teams that need risk traced to a signal | Scores individual signals; tested on about 20 past accounts | An engagement, not self-serve; no CSM playbook suite | Fixed-fee engagement, scoped by a GTM Audit before anything is built |
Gainsight Customer Success and Staircase AI
Gainsight sells its Customer Success platform in two packages, as of October 2026. Essentials includes 10 full users and AI Insights and Automations; Enterprise includes 20 full users and adds Renewal and Expansion Forecasting and Organizational Mapping and Sponsor Tracking. Gainsight does not publish prices. Its Staircase AI product works differently from a classic scorecard: it automatically logs emails, support tickets, chats and meetings, surfaces churn risk, expansion momentum and stakeholder disengagement, and turns a risk signal into team action with assigned ownership and defined next steps. Gainsight's own framing is that "most churn risk and expansion intent never show up in CRM fields or dashboards until it's too late."
The trade-off is scope: you are evaluating a full CS platform and an AI layer, so make the proposal say which one solves your problem. Gainsight is better suited to CS organizations with several teams and a CS operations function than to a team of four CSMs.
ChurnZero
ChurnZero's health scores, called ChurnScores, track account health using metrics you choose: account attributes, engagement, product usage, journey progress, support history and loyalty measures such as NPS, CSAT and CES. You can build separate scores by segment or lifecycle stage, and score changes can trigger automated campaigns and alerts. Its AI Knowledge Sources feature connects AI agents to playbooks and documentation in tools such as Zendesk or Confluence. ChurnZero does not publish pricing. It suits mid-market teams that want scores and plays in one place, but the score is only as predictive as the metrics and weights your team picks.
Vitally
Vitally describes a no-code health score engine that combines metrics, traits, activities and events with weights you can fine-tune, including by customer type, with real-time alerts when scores change and playbooks that trigger outreach. It sells three plans, Tech-Touch, Hybrid-Touch and High-Touch, and prices are available on request. Vitally fits teams that want admins to adjust scoring without engineering help. The limit is shared by every weighted model, including yours: one serious signal can be averaged out by several benign ones unless you design for it.
Planhat
Planhat's customer health page describes a Health Laboratory for building real-time scores by cohort or lifecycle stage, alerts through email, Slack or Teams, and health data pushed back to the CRM. Testing several score designs helps if you suspect your score is wrong but don't know why. The page does not describe AI scoring and does not publish pricing.
Totango, now Odie
Totango merged with Catalyst in early 2024, with the two brands kept separate at the time, according to CX Dive. As of October 2026, Totango's website says "Totango is now Odie," described as a "Customer Knowledge Engine" for post-sales teams, with a new platform. The homepage no longer describes the earlier Totango and Catalyst products. Pricing is not published. If you are a Totango or Catalyst customer, ask how health scoring carries over to the new platform before you renew.
HubSpot health scores
If you run on HubSpot, you may already own a health score. HubSpot's customer success workspace health scores are available in Service Hub Professional and Enterprise, with seats required for certain features, as documented in June 2026. Scores combine event groups, such as meetings and calls, with property groups, such as company properties or list membership, with point caps and time-based decay. Pre-defined groups include 1:1 Engagement, Support Tickets, CSM Sentiment and NPS. Multiple scores need Enterprise, up to 50 per account. The documentation does not describe an AI-generated score, so the prediction is only as good as the rules you write.
A health score rebuild
The option most guides leave out is fixing what you have. If your product usage, tickets and contacts already sit in your CRM or warehouse, rebuilding the score usually means three changes: let certain signals raise risk on their own instead of being averaged, add the events your post-mortems keep finding (sponsor departures, repeated tickets on one issue, seat reductions), and test the new score against accounts whose outcome you already know. TSIA notes that heavily weighting CSM sentiment is associated with declining retention, which is a good first thing to check. Our guide to backtesting a scoring model uses the same discipline on the sales side.
VANDFORT Churn Signal Watchtower
The Churn Signal Watchtower is a custom system we build inside your stack. It watches for the events that come before a loss, such as a champion who stops logging in, a second ticket on the same problem or a quiet seat reduction, and scores risk on individual signals, so one serious signal is not cancelled out by four benign ones. Each at-risk account comes with the signal that fired and when. Before it goes live, it is tested twice against about twenty past accounts your team labels as churned, saved or never at risk; it ships at 85 percent or it does not ship. It runs on read-only access and never emails a customer or changes an account record.
It reads cleaner data when the Pipeline Hygiene Sentinel runs upstream, and its alerts feed Renewal Radar, so a signal starts a renewal conversation. The limits are real: it is an engagement measured in weeks, and it is not a CS platform with portfolio views, success plans or customer-facing playbooks.
How to choose: platform, rebuild or system
Run a post-mortem on your last ten churned accounts. For each, note what the score said 90 days before renewal. Mostly green means a prediction problem. Red with no action means a workflow problem, which a better score will not fix.
List the signals the post-mortems found. Sort them by where they live: CRM, product analytics, support desk, email and calls. If most sit in systems you already connect, a rebuild is likely enough. If most sit in conversations nobody logs, you need a tool that reads communications.
Check how your current score combines signals. A weighted average is the most common design and the most common reason a single serious event disappears. Ask whether one signal, on its own, can turn an account red. If not, fix that before you buy anything.
Count what your CSMs need besides the score. If you also need success plans, onboarding journeys and tech-touch campaigns, a CS platform earns its cost. If the team works well in the CRM, a platform adds a second place to look.
Test on accounts with known outcomes. Ask each vendor, and your own team, to score twenty past accounts as they looked 90 days before renewal. Compare with what happened. Every option here should be able to do it, including ours.
Check that a CSM can explain the alert. In the demo, pick a red account and ask why it turned red this week. If the answer is a number without a reason, expect your CSMs to override it.
A rule of thumb, offered as a suggested starting point rather than a benchmark: if your post-mortems point to signals you already capture, rebuild the score and re-test it every quarter. If they point to conversations and stakeholder changes nobody records, look at tools that read communications. If your problem is that alerts go nowhere, fix the renewal workflow first; our NRR early-warning system guide covers that side.
Where VANDFORT fits, and where it doesn't
You are a $3M to $30M ARR B2B SaaS company and your post-mortems keep finding churn signals that were recorded but never moved the score. Your health score averages inputs and your CSMs have stopped trusting it. You want risk scored on the data you already hold in your CRM, product and support tools, with each alert traced to the signal behind it, and tested on your own past accounts before anyone relies on it. The Watchtower is one of the nine systems in our Nine Systems Framework, built to sit next to the CS tools you already use.
If you need a full customer success platform with success plans, onboarding journeys and tech-touch campaigns, buy one such as Gainsight, ChurnZero, Vitally or Planhat. If you are on Service Hub Professional or Enterprise and have never configured its health scores, try that first. If your data is clean and you have an analyst who can rebuild and re-test the score, do it in-house. And if you want a self-serve tool your admin can configure in a day, buy software.
Frequently asked questions
What is the best churn prediction software for SaaS?
It depends on what is failing. Gainsight, ChurnZero, Vitally, Planhat and Totango are the main customer success platforms with configurable health scores, and Gainsight's Staircase AI goes furthest in reading emails, tickets and meetings for risk. If your current score fails for known reasons, such as averaging signals or ignoring sponsor changes, rebuilding it on data you already hold is often enough.
How accurate are customer health scores at predicting churn?
Often less accurate than teams assume. TSIA, in an article updated in May 2026, says most organizations using non-AI health scoring models do not consider their scores highly accurate, and only 22% use AI for health scoring. The fastest way to know is to score 20 or more past accounts as they looked 90 days before renewal and compare the result with what actually happened.
Can HubSpot predict churn?
HubSpot can score account health, but it predicts churn only as well as the rules you write. Health scores in the customer success workspace are available in Service Hub Professional and Enterprise. They combine events, such as meetings and calls, with company properties and list membership, and support point caps and time decay. HubSpot's documentation does not describe an AI-generated churn score.
What signals predict churn in B2B SaaS?
The most useful signals are usually events rather than averages: an executive sponsor or champion leaving, a drop in active seats or usage on a key team, repeated support tickets on the same problem, and a change in who attends meetings. NPS and CSM sentiment help but are lagging and subjective. Test any signal against your own churned and retained accounts before weighting it.
Do we need a customer success platform to predict churn?
Not always. A platform earns its cost when your CSMs also need success plans, onboarding journeys, portfolio views and automated campaigns. If your team already works in the CRM and your usage and support data are connected, a rebuilt health score or a custom churn-risk system may be enough, without adding a second place for CSMs to look. Test either option on past accounts before you decide.
Sources: SaaS Capital, 2025 B2B SaaS Retention Benchmarks (14th annual survey, 1,000+ private B2B SaaS companies, September 2025). TSIA, Customer Health Scoring: The Shift Toward Predictive Models (Darlene Kelly, updated May 2026). Harvard Business Review, The Value of Keeping the Right Customers (Amy Gallo, October 2014, citing Frederick Reichheld of Bain & Company). CX Dive, Catalyst and Totango merger (March 2024). Gainsight, Staircase AI and pricing pages. ChurnZero, customer health scores page. Vitally, pricing page and health tracking software comparison (February 2026). Planhat, customer health page. Totango website. HubSpot Knowledge Base, create a health score in the customer success workspace (updated June 2026). Vendor pages accessed October 8, 2026.
Last reviewed: October 2026. Vendor features and pricing change; check each vendor's site before buying.
