A customer tells their account manager they are not renewing. The account manager is surprised. The last quarterly review went fine, the meetings were pleasant, nobody raised a complaint.
Then someone goes back through the data. The main contact stopped opening email in early July. In late July, three people from that company read the same troubleshooting page four times in a week. In August, a new VP appeared on a meeting invite. In September, someone from their domain looked at the pricing page.
All of that was recorded. None of it was watched, because watching it was not anyone's job.
Why marketing sees it before the account team does
Account reviews happen on a schedule. Behaviour changes do not.
If your reviews are quarterly, a customer can go quiet in week two and nobody looks until week twelve. And customers manage their side of the relationship carefully. They stay polite. They keep the meetings. They answer questions. Very few people tell a vendor they are evaluating alternatives, because saying so starts a conversation they do not want to have yet.
Marketing systems do not record what people say in meetings. They record what people do. That is the whole advantage.
One thing to be clear about before the list. None of these signals predicts churn on its own. Each one means something changed inside that account and you do not know what. That is enough. The value here is timing, not certainty: a conversation in week six is a different conversation from the one you have after the renewal decision has been made and communicated.
Signal one: email engagement collapse
This is the most reliable one, and the most commonly misread.
The number that matters is engagement at a named contact, not average open rate across your list. If your whole list drops, that is deliverability or content, and it is a technical problem. If one decision-maker who opened every message for a year has opened nothing in six weeks, that is an account problem.
In HubSpot you can build this as a list: contacts with an active deal or a customer lifecycle stage, whose last email open is more than 45 days ago, and whose previous 12-month engagement was above average. Salesforce and Pipedrive can do the same thing with slightly more work. The specific tool matters less than the shape of the query, which is this person used to engage and has stopped.
Two cautions. Apple Mail Privacy Protection and similar features inflate opens and make them unreliable in isolation, so pair opens with clicks and page visits before you trust the signal. And check holidays and parental leave before you escalate, because a person who has gone quiet may simply be away.
Signal two: the support content spike
A run of visits to your help documentation, troubleshooting pages, or setup guides from one account, with no support ticket attached, means someone is struggling and has decided not to tell you.
That decision is the part that matters. A customer who feels good about the relationship raises the issue. A customer who has started to disengage goes looking for the answer alone, and when they do not find it, the frustration stays inside their company and never reaches yours.
Watch two versions of this. A sudden spike where there was nothing before usually means a new problem or a new user who was never trained. A slow constant hum of documentation visits over months means the product is harder to use than they expected, and that one shows up at renewal as a general complaint about value.
If your knowledge base sits on a different domain or a third-party tool, this data is probably not connected to your CRM at all. Connecting it is a small piece of work with a large payoff.
The related version is a new user who was never trained. Someone joined the customer's team, got handed your product, and started teaching themselves from your documentation. That person forms an opinion about how hard your product is to use in their first two weeks, with nobody from your side in the conversation. A short call at that moment is cheap and changes the opinion.
Signal three: stakeholder change
In B2B, the relationship sits with a person. When that person leaves, the decision they made becomes a decision their replacement inherited, and a new manager reviewing inherited vendor contracts is normal behaviour, not a sign of trouble with your product.
Three versions to watch for.
- Your champion leaves. The first clue is usually an email bounce at a company address. Set your CRM to flag hard bounces on customer contacts and route them to the account owner within a day, not at the next review.
- A new senior title appears. Somebody you have never met joins a call, or a new name shows up on an invite. New leadership frequently runs a vendor review in the first 90 days.
- Your contact stops replying directly. Answers start coming from a more junior person, or from a shared inbox. Delegation of a relationship is a demotion of that relationship.
For a small book of accounts, checking LinkedIn manually once a month is enough. For a larger one, a data provider that flags job changes across your customer list does it automatically, and it pays for itself on the first save. If you are already keeping contact records clean and current, most of the plumbing for this already exists.
Signal four: competitor and alternative research
An existing customer looking at your pricing page or your comparison pages is usually building an internal case. Sometimes that case is for renewing, and sometimes it is for switching. Either way it means the purchase is being discussed inside their company without you in the room.
This one is easy to lose by accident. Many teams exclude existing customers from analytics segments and retargeting audiences to keep acquisition numbers clean, which is reasonable for media buying and destroys exactly the signal you want here. Exclude customers from the ads. Keep them in the behaviour data.
The related signal is external. If prospects start asking you about a competitor you had not heard of, your customers have heard of them too. Ask your last five closed-lost prospects who else they looked at, then check whether those names show up in your existing customers' questions.
There is one more version of this that most teams miss entirely. A customer who starts asking your support team about exporting their data, bulk downloads, or API access to their own records is preparing to move something somewhere. Sometimes it is a reporting project. Sometimes it is a migration. It is worth a call either way.
Building the early warning list
Do not build a model. Build a list.
Start with two signals, not four. Engagement collapse at a named contact and stakeholder change are the two with the least noise, so begin there. Produce one short weekly list: account name, which signal fired, and the date it changed. Nothing else.
Deliver it where the account team already works. A dashboard in a tool customer success opens twice a year is worth nothing. A task in their CRM, or a message in the channel they read every morning, gets acted on.
Then validate it before you expand it. Take your last ten churned accounts, pull their engagement history, and find the week their behaviour changed. Two things come out of that exercise. You learn how much warning your signals actually give you in your business, which is a real number instead of a borrowed one. And you learn which signals fired for accounts that did not churn, which tells you where the noise is.
One rule about what happens next. When a signal fires, someone calls. Not an automated re-engagement sequence, not a satisfaction survey, not a discount. A customer who is already frustrated and receives a marketing email at that moment gets confirmation that nobody is paying attention to them specifically. Pick up the phone.
The same watching discipline runs in the other direction. An account showing growth, heavy usage, or a newly promoted champion is not a churn risk, it is a signal that the account is ready to buy more, and it deserves the same weekly list and the same named owner.
The takeaway
Your marketing systems have been recording churn signals the whole time. Email engagement at named contacts. Support page runs with no ticket. New titles on the account. Customers reading your pricing page.
Pick two signals. Send one short list every week to the people who own the accounts. Call, do not email. Most churn is decided weeks before anyone says it out loud, and the decision is easier to reverse before it gets announced.