01Why the break is quiet, not loud
A loud failure is the easy kind. A form throws an error, a person sees it, someone fixes it. The dangerous kind is quiet. The automation keeps running and does the wrong thing, or it receives nothing and so has nothing to fail at. This is our reasoning from how connected tools work, not a count we hold. A login or key lapses between two tools. A connected tool changes a field or a format the automation relied on. Someone edits the form or the calendar the automation reads, and the old rule no longer matches. None of these sends an alert to a person on its own. The first signal is usually a client saying nobody answered their booking. We publish no figure for how often this happens, because we do not measure other people's systems.
- An expired login or key stops the flow without telling anyone
- A connected tool changes a field or a format: the rule no longer matches
- Someone edits the form or the calendar the automation reads
- The trigger receives nothing, so there is nothing to raise an error
02What a health check has to catch
An error alert covers the loud case. The quiet case needs the opposite: an alert when something that should have happened did not. If a booking flow normally confirms appointments on working days, a day with none is worth a message to a person, because it may be a quiet day or a dead trigger, and only someone who knows the business can tell. PROOF48's automation page says every automation has a health check, that failures alert you and not your clients, and that we write the failure paths, such as an API being down, before the first production message goes through. The page does not describe the method, so when anyone says monitoring, ask what is checked, how often and who gets the message. If you want to see how that would look for your own flow, the automation page is where to start.
- Alert on errors and on silence: a flow that should have produced something and did not
- Ask what is checked, how often and who receives the message
- Send a test through the full path after any change to a connected tool
- PROOF48's page says failures alert you, not your clients
03Who owns the fix
Three parties can be responsible, and the terms decide which. If you built it yourself in a tool you pay for, the fix is yours and the vendor answers only for its own platform. If a freelancer built it and moved on, nobody may own it. If a studio maintains it, the support agreement says how fast and within what limits. PROOF48's automation page says every automation includes monitoring and a defined alert path, and that maintenance is included in all growth plans and available as a standalone arrangement for custom builds. Our plans page lists website maintenance at 27, 57 and 97 EUR a month, with a 2-business-day, a 1-day and a same-day response. Those terms are written for the website. The plans page states no response time for an automation failure and does not name which maintenance level the growth plans carry, so we quote none here. Ask for it in writing, from us or anyone.
- Built by you: the fix is yours, the vendor covers only its platform
- Built by someone who moved on: nobody owns it until you name an owner
- Built and maintained by a studio: the written terms decide speed and limits
- PROOF48 publishes no response time for an automation failure
04Five questions to ask before you buy
Put these to every vendor, including us. They cost nothing, and the answers separate a maintained system from a hope. Ask what happens to the messages, orders or bookings that arrive while it is broken, because a queue that retries is different from a gap you refill by hand. Ask who gets the alert, on which channel, and whether it reaches a person at the weekend. Ask what is documented: our automation page says the system is written up so anyone on your team can understand what runs and why, not just the person who built it. Ask who holds the logins and accounts. Our post on what to ask before you buy an AI chatbot covers the chatbot side of the same conversation. If you would rather ask us, send the flow you have in mind from the automation page.
- What happens to the work that arrives while it is broken: queue, retry or manual refill?
- Who gets the alert, on which channel, and does it reach a person at the weekend?
- What counts as broken, and how is silence detected?
- Who holds the logins and the accounts it runs on?
- What is written down, so someone other than the builder can fix it?
05Where PROOF48 is the wrong choice
If your automation is one simple rule inside a tool you already pay for, and you are happy to check it yourself every week, a managed plan is more than you need. AUTOMATE costs 2,997 EUR a month and includes an AI website chatbot, AI DM and email auto-responders, booking automation with a CRM, automated review management and two custom automations a month. It does not cover WhatsApp or phone. We publish no response time for automation failures and no count of the failures we have fixed. If you need a response time in minutes written into a contract, ask us what we can put in writing, because the plans page does not state one. Our post on when not to automate a process covers the cases where waiting is the better call.
- One simple rule in a tool you already pay for: check it yourself weekly
- AUTOMATE: 2,997 EUR a month, no WhatsApp, no phone
- No published response time for an automation failure
- No published count of fixed failures
Two failure cases designed in from the start
ATP Spedition's case study lists an offline-first driver flow, so a job survives losing signal mid-route, and Romanian e-Factura and e-Transport handled as first-class requirements. The first is a failure case designed in from the start: the signal disappears. The second is a place where a system outside the business can say no. The entry records what shipped, not what has gone wrong since, so we claim no uptime figure, no incident count and no response time from it.
See the build: ATP SpeditionQuestions, answered
How do I know if an automation has stopped working?
Look for the result it should have produced, not for an error. Send a test through the full path on a schedule and after any change to a connected tool, and ask the builder what its health check covers. PROOF48's automation page says every automation has one.
Who fixes it when a connected tool changes something?
Whoever the support terms name. If you built it, you. If a studio maintains it, the written agreement decides. PROOF48's automation page says maintenance is included in all growth plans and available as a standalone arrangement for custom builds.
How fast will a broken automation be fixed?
We publish no response time for automation failures. The plans page lists 2-business-day, 1-day and same-day response for website maintenance only. Ask for the automation terms in writing before you sign.
What does AUTOMATE cost, and does it cover WhatsApp?
It costs 2,997 EUR a month. It covers an AI website chatbot, AI DM and email auto-responders, booking automation with a CRM, automated review management and two custom automations a month. It does not cover WhatsApp or phone.
Where PROOF48 fits
A broken automation often fails quietly, so ask for an alert on silence as well as on errors, name who fixes what before you buy, and get the response terms in writing. PROOF48's automation page says every automation has a health check and a defined alert path, and the plans page publishes no response time for an automation failure. AUTOMATE costs 2,997 EUR a month and does not cover WhatsApp or phone.
Cosmin Tiu · Founder, PROOF48
Cosmin has worked in online marketing and software development since 2018, and has built with AI every day since January 2023. He still writes code from scratch daily, in Python, React, Rust and C#. PROOF48 ships business websites in 48 hours and invoices afterwards, so every opinion on this blog was paid for in real builds, not in theory.