Seasonality
Seasonality is expected, so a landscaping supplier going quiet in January is not a warning.
For the skeptic
Check every warning before you trust it.
Open the calculation, the source invoice, or the email quote. ChurnRisk shows what it read, how it reached the result, and where the evidence is incomplete.
Scope
What it reads
From QuickBooks: customers, invoices, credits, payments, and estimates. The connection is read only.
From email, it reads only the mailboxes you choose and respects your exclusions. It stores structured signals and short quotes, then links back to the source thread.
No personal mailboxes. No model training on your data.
The connector pages list this scope permission by permission: QuickBooks and Gmail and Outlook.
Baselines
Your customer is the baseline
A customer who orders every three weeks is late after five. A customer who orders twice a year is not.
ChurnRisk learns each customer's order cadence, value, quote conversion, payment timing, and reply patterns from their own history.
Warnings come from a break in their normal, not an industry average.
Exceptions
A slow January might be normal
A changed number can have an ordinary explanation. ChurnRisk checks the context before it raises the warning.
A large one-time project
A large one-time project can inflate a baseline. ChurnRisk treats that project as an exception when it measures recurring work.
Raw material prices
A raw-material price change moves invoice values even when the customer relationship is unchanged.
A cause on your side
A stockout on your side calls for an internal fix. ChurnRisk keeps that context with the warning so the customer does not get the wrong message.
Suppressed warnings stay visible, so an exception is never a hidden decision.
Four readings
One account, four questions
How likely is the loss? How much is at stake? How good is the evidence? What should happen next?
Risk
The probability something is wrong.
Revenue at stake
What it costs if it is.
Confidence
How much of the picture we can see.
What to do
Whether acting now could help, followed by the next step.
Low confidence calls for better coverage. High-confidence risk with major revenue at stake calls for action.
Where AI sits
AI does not score your customers
It reads text, summarizes the source, explains the result, and drafts a reply for you to review.
Models never assign a risk score or decide whether something is a risk.
Scoring is deterministic and reproducible. The same inputs produce the same result.
High-severity email signals require a matching source quote. Without one, they are downgraded automatically.
Coverage
Missing data is marked unknown
Phone orders, EDI, portals, unconnected mailboxes, and stale connections all leave gaps.
In every one of those cases, confidence drops and the interface says so. A customer we cannot see is marked unknown, never healthy.
Limits
What stays outside ChurnRisk
ChurnRisk cannot see orders outside QuickBooks, conversations outside connected mailboxes, or offline relationship changes. It cannot promise that an intervention will save an account.
It can show what QuickBooks and selected inboxes already know, with the biggest accounts first and the source attached.
Start with a customer you already know.
Open a workspace now, or ask us to walk through the evidence on a sample account.
Nothing connects until you approve it.
