Company churn check
With one click on a company page, this agent tells a sales rep how likely the company is to stop ordering and what to do about it.
To have Propeller watch all companies by itself, use Churn risk alerts. It uses the same score.
| In short | |
|---|---|
| Where you use it | On the page of a company in the Sales Hub, behind the Workflows button |
| Starts | With a click on the agent |
| Changes in Propeller | Nothing. It only reads orders |
| Uses AI | No |
| Texts | English |
| Runs in n8n | One per click |
| Difficulty | Easy |
What you get
The agent is in the list of agents for a company. One click gives the answer in the same dialog. A company without enough order history gets no score:

An answer with a score reads like this:
🟡 Example B.V.: medium churn risk (47/100)
Ordering less often: 1 order in the last 30 days against 8 orders in the 60 days before.
Spending less per order: on average €1,900 now against €2,850 before.
Overdue: the last order was 14 days ago; they normally order every 10 days.
Suggested action: send a check-in message and ask about open issues or changing needs.
What you need
- An n8n account with the credential
Propeller API key. See Set up n8n. - An API key with Viewer access on Order Data.
- Backoffice access and the Configuration role with Owner access, to register the agent.
How it works
The workflow compares the orders of the company in the last 30 days with the 60 days before. Three signals make up a score from 0 to 100: the company orders less often, spends less per order or is quiet for longer than usual.
For the usual time between orders, the agent counts the days with an order. Several orders on one day count as one, so a company that places a few orders at once does not look like it orders every day. When the orders it looks at were all placed on one day, there is no usual rhythm, and the agent gives no signal for being quiet.
| Score | Risk | The agent suggests |
|---|---|---|
| 60 or higher | High | A call or a visit |
| 30 to 59 | Medium | A message to check in |
| Below 30 | Low | Nothing |
A company with fewer than 2 orders in the 60 days before has too little history. The agent then says so and gives no score.
The workflows and agents in the Agent Hub are free to use and adapt. They are examples, not supported product features. Test them in a test environment and check the results before you use them on real customers.
Install it
Install a workflow from the library explains each step.
- Import the file into n8n.
- Check that the node Fetch Order Stats uses the credential
Propeller API key. - Follow Register the agent in Propeller with the values below.
- Open a company in the Sales Hub, click Workflows and click the agent.
| Field | Value |
|---|---|
| Trigger | Button |
| Interaction mode | Single turn |
| Placements | Sales Hub · Company, the page of a company in the Sales Hub |
The answer shows in the dialog. An error instead of an answer means the address, the login or the key is not right. It can also mean that the workflow is not published.
Make it yours
| Setting | Default | What it does |
|---|---|---|
RECENT_DAYS | 30 | The period that is judged |
BASELINE_DAYS | 60 | The period before it, which counts as normal for the company |
MIN_BASELINE_ORDERS | 2 | With fewer orders in the period before, the agent gives no score |
WEIGHTS | 0.4, 0.3, 0.3 | The share of each signal: ordering less often, spending less and being quiet |
ORDER_STATUSES | NEW, CONFIRMED, VALIDATED, ARCHIVED | The statuses that count as an order. See Fill in order status codes |
TIMEZONE | 'Europe/Amsterdam' | The time zone in which the days with an order are counted |
The scores of the three risk levels and the suggested actions are in the node Score & Format.
Limits
- The average order value and the usual time between orders are taken from the latest 100 orders of each period. The answer says so for the order value when a company has more.
- The agent only works on a company page.
Related pages
- Churn risk alerts - the same score for all companies, twice a week
- Register an AI agent
- Adjust a workflow with an AI assistant - change what the example does