
Workflows that stop sending
A workflow can say Published and send nothing for a week. Watchdog learns each workflow’s normal volume and alerts when it goes quiet, spikes, or slips back to draft.
workflow.silentWorkflows that stop sending. Conversation AI switched off. Voice-AI calls that promised a booking nobody made. Opt-outs creeping toward a carrier flag. Watchdog checks every sub-account, all day, and tells you before your client does.
Brightwater Pools & Spas: “Pool season follow-up” has not sent in 3h 04m. Still published.
Open workflow in GHL →[Critical] Brightwater: workflow went quiet
Trigger fixed by Sam. First send in 2m. Watching.
Alert sent to chat and email, with a link straight to the workflow in GHL.
Illustrative replay. Fictional agency and businesses; time is fast-forwarded.
From the fictional demo agency, computed by the same checks the demo runs.
Broken workflows rarely throw errors. They just stop. The dashboard stays green because nothing ran, so nothing failed. We have lived this pattern: it took the better part of a week, and a phone call from the client, to find out.

Illustration of a real failure pattern; no client, dates or figures from it are shown.
Each runs on a schedule against every sub-account you connect. Thresholds are yours to set, per client.

A workflow can say Published and send nothing for a week. Watchdog learns each workflow’s normal volume and alerts when it goes quiet, spikes, or slips back to draft.
workflow.silent
You say which bots should be on. If one is switched off, by anyone, you hear about it with who changed it and when.
convo_ai.off
Calls where the AI promised a booking that never landed, callers who asked for a person and were not transferred, calls cut off in the greeting.
voice.booking
The 7-day opt-out rate against a ceiling you set, the send that caused the climb, and brand or campaign status changes before carriers act.
a2p.optout
Opportunities that sit in a stage longer than you allow, with the dollars attached, so follow-up happens while the lead still remembers you.
funnel.stuck
One number per client every morning, built from the checks above, plus a written brief of what changed overnight and what needs a person.
location.healthAn AI summary is easy to read and easy to get wrong. So Watchdog splits the job. Every number is computed by plain, tested code and locked before any AI sees it. The AI arranges those numbers into a readable morning brief; a validator rejects any brief whose numbers do not match.
Across sub-accounts, Watchdog verified messages actually left in the last 24 hours. things need a person this morning; the rest can wait for your normal check.
“Pool season follow-up” is still published, but its last message went out ago. It usually sends about a day. The sub-account’s total sends dropped from to . Open the workflow and check its trigger first; nothing errored, it simply stopped running.
“Estimator assistant” on SMS was switched off by “a user in the sub-account”. Since then it has answered messages. If nobody meant to do that, turning it back on is one click in GHL.
Locked numbers are counted by code before the AI sees them. The AI can place them; it cannot change them.
Install from the GHL Marketplace and approve read-only access, or book a white-glove screen-share and we do it with you.
Watchdog reads each workflow’s history to learn its normal volume, and asks which bots should be on.
Checks run all day across every sub-account. Health scores update; nothing in your account changes.
Problems go to email and your team chat, routed per client, each with a link straight into GHL.
Draft tiers for discussion. Prices are not set yet.
One agency, a handful of clients.
$ —/ month
Placeholder · price not setThe agency that runs on GHL.
$ —/ month
Placeholder · price not setLarge books of sub-accounts.
$ —/ month
Placeholder · price not setEvery plan: GHL Marketplace install or white-glove onboarding over screen-share, read-only by default, cancel any time. Tier limits and prices are a draft for discussion.
Not by default. It connects with read-only access and only reads workflows, conversations, call logs, opportunities and messaging health. Any fix happens in GHL, through the link in the alert.
As a GoHighLevel Marketplace app: you install it at agency level, pick the sub-accounts, and approve the access it asks for. If you would rather not do that alone, we do it with you on a screen-share.
From the workflow’s own history. If a workflow normally sends a few messages an hour and has sent nothing for longer than your threshold, that is an alert, whatever its status says.
No. The checks are ordinary code with thresholds you can see and change. AI writes the morning brief around numbers it is not allowed to alter.
Email, plus a chat channel of your choice. You can route each sub-account to a different person or channel and set quiet hours.
This site is a product preview. The demo dashboard runs on fictional data and makes no calls to GoHighLevel.
The demo walks the whole thing: connect, overview, a client drill-down, alert settings and the morning brief.
Open the live demo