A useful alert is not one more notification in Slack. It says which client store is affected, which sales step is concerned, whether revenue is probably blocked and what to check now. It keeps proof that can be shared with the client, and it avoids guesswork.
For a WooCommerce agency, this framing changes everything. A client does not want a stream of technical signals: they want to know whether their store can still sell, whether the cart holds, whether checkout still displays payment methods, whether the failure comes from cache, theme, a plugin or a gateway, and above all who acted and when.
This guide describes a simple method. It assumes an agency runs several stores, each with different constraints, and that alerts must be separated into those that wake a team, those that feed a report and those that should trigger an AI intervention on a paid plan. It keeps customer data out of the operational noise, and it respects one central idea: an alert is worthless if it does not produce a decision.
1. First separator: availability or conversion
A site can be available and no longer sell. An HTTP ping can return 200, the homepage can load, the catalogue can stay visible, and yet the "add to cart" button no longer acts, the cart empties, checkout throws a JavaScript error, a card payment disappears, a wallet stays active in WooCommerce but absent on the public checkout, a required field blocks the order, a 3DS redirect fails, or a confirmation never appears.
The agency must therefore separate two families of alerts. The first covers general availability: it says whether a page responds, whether the server is slow, whether a route returns an error. The second covers conversion: it says whether a user can move toward the purchase, whether the critical steps hold together, whether revenue is at risk.
CashFlowCanary sits in that second family. The client cockpit must answer one short question: can this store sell right now?
2. Do not alert every page the same way
An important product page deserves a fast signal; a secondary category page can wait. The cart deserves a higher priority, the checkout the highest. The confirmation page needs a different reading: it proves the end of the journey holds. The account and contact pages do not carry the same direct impact or urgency.
The agency must map pages by impact: low, medium, high, critical. The word "critical" must stay reserved; used everywhere, it means nothing.
A good alert starts with a classification: which monitor observed the break, which path is affected, which client plan covers that path, which frequency is promised, which channel must fire, which report will be produced.
3. Choose channels by the type of decision
Email is useful for tracing, archiving and forwarding to the client; it is poor at waking a team in real time. Slack coordinates a team: it gives a thread and lets you tag an owner, but it becomes noisy if everything lands in one place. Telegram is direct and effective on mobile for fast alerts, as long as it stays reserved for signals that deserve attention. WhatsApp is useful when the client or team already lives in that channel; it must be configured properly and never used for outreach without opt-in.
The right choice is not a single one. An agency can use email for evidence, Slack for the technical team, Telegram for on-call and WhatsApp for some premium clients. The same incident can therefore generate several notifications, but each must have a reason.
4. Define a policy per plan
The free plan validates a signal. It must not promise a full intervention, all channels or AI intervention. The paid plan changes the level of commitment: it can open multi-channel alerts, longer history, advanced PDF reports and AI intervention.
This distinction must appear in the product, the site, the plugin, the reports and the alerts. A channel that is out of plan, unconfigured, suspended or delivered by the server but not chosen must never be shown as active.
Words matter. "Active" means an alert will be sent. "Not configured" means a destination is missing. "Out of plan" means the offer does not allow it. "Delivery not ready" means the destination exists but the server cannot deliver yet. This precision avoids pointless arguments.
5. Write a readable alert card
An alert should fit in a few fields: store name, affected path, monitor type, observed status, error code, opening time, probable impact, available proof, recommended action, next check.
The message must not be a raw dump. It must not carry unnecessary personal data, nor expose a cart, a customer email or a token. It must orient the team.
A good Slack message fits in six lines: "Store A — checkout blocked, payment method missing, critical priority, PDF proof available, recheck in three minutes, AI intervention launched if the plan allows it." That message is enough to act; the technical detail stays in the report.
6. Connect the alert to the report
An alert is instant, a report is durable. One triggers action, the other keeps the record. The agency must connect the two: each critical alert points to proof, each proof recalls the context, each report lists the monitors involved and explains the incidents precisely, without copy-pasted descriptions. Each AI intervention details what was attempted, specifically per monitor.
A report that repeats the same sentence everywhere is not a report, it is noise in PDF form. The report must help a developer, a project manager and the client understand. It must not hide failing monitors, show a perfect health score while monitors break, hide unconfigured channels or replace proof with a slogan. It must use the correct CashFlowCanary logo, or the configured agency logo when white-label branding is available.
7. Prioritize without dramatizing
Not every incident is a crisis. A one-off timeout can be observed, a slow page can be tracked. But a checkout that no longer loads, a cart that empties, a missing gateway or a blocked payment button are emergencies. A missing confirmation is a strong alert, but it is read after the context.
The agency must therefore use levels: information, watch, degraded, critical, resolved. Each level has a consequence. "Information" wakes no one. "Watch" feeds the report. "Degraded" goes to the team channel. "Critical" fires the real-time channels. "Resolved" closes the loop. Without a consequence, a level is useless.
8. Avoid false positives visible to the client
A false positive costs trust, and it costs more when it reaches the end client. So filter before escalation. An incident is confirmed by crossing several signals: HTTP code, page snippet, absence of an expected element, abnormal load time, last run state, comparison with the previous run, presence of an already open incident.
The system must not open ten identical incidents: it groups what belongs to the same symptom, closes cleanly when the recheck succeeds, keeps a timeline and says what changed. The client must not receive a green alert that contradicts a red one; the cockpit must stay coherent.
9. Configure alerts per client site
An agency does not run a single global channel, it runs portfolios. A premium client may require WhatsApp, a standard client may stay on email, an internal team may work in Slack, an external contractor may follow Telegram. The channel must therefore live at the monitored-site level.
Each site must show email, Slack, Telegram, WhatsApp and escalation, with an "active" or "not active" status per line, and any destination holding sensitive data must be masked.
The test button must be clear: it sends a verification message, without creating an incident, restarting monitoring or modifying the store. A channel test is a delivery test, not a break test.
10. Operational conclusion
A useful alert does not shout louder than the others. It connects the right site, the right channel, the right impact and the right proof. It keeps the agency responsible for the decision rather than trapped by noise. It launches AI intervention only when the plan and the safeguards allow it, and it confirms resolution before reassuring the client, within the limits visible on the pricing plans.
To see the proof level expected on the client side, open the sample PDF report. This is the discipline an agency can frame with a free audit before industrializing its alerts.