# Who Maintains Your AI Automations After Launch? Published 2026-10-16 by Sid Jashnani. Why automations break after launch, what a runbook should contain, and how to avoid a workflow that only one freelancer can fix. ![Who Maintains Your AI Automations After Launch?](/blog-images/who-maintains-ai-automations-after-launch/cover.png) Automations break when an app changes, a login expires, or a model starts answering differently. If nobody watches and nobody has a runbook, the failure stays quiet until a customer notices. Name an owner. A full-time [AI Implementation Specialist](https://www.rekruuto.net/ai-implementation-specialist) is one way to do that. A written runbook is the other half, whoever you hire. The checks below are the operating habit that keeps a workflow alive after the person who built it moves on. ## Why do Zapier, Make, and n8n automations break? They break because they depend on other people's software, on credentials, and on data that your team keeps editing. A field gets renamed in the CRM. A form adds a required question. An API version is retired. A token expires. A filter that used to match "Closed Won" now matches nothing because someone changed the spelling. None of that is exotic. It is a Tuesday. AI steps add another failure. The prompt still runs, the text still looks polished, and the fact is wrong. A rules-only zap fails loudly more often. A model step can fail politely. You need a check that does not trust the tone. Volume makes small error rates expensive. A workflow that is wrong once a week is a nuisance. The same workflow on every order is an incident. That is why the first automations should be the reversible ones in [which processes to automate first](https://blog.rekruuto.net/which-business-processes-to-automate-with-ai-first). Workflows break when an app, a login, a field, or a model changes. Decide who hears about it before you switch the workflow on. [Upwork's Zapier developer page](https://www.upwork.com/hire/zapier-developers/) cites a 2024 Zapier study that puts the average time saved at 13 hours a week per automated workflow. That saving only holds if someone still watches the workflow after launch. ## What happens when the freelancer who built them leaves? The workflows keep the status they had, and the knowledge leaves, unless you moved both the access and the notes before the last day. On Zapier Team and Company plans, owners and super admins can review a member's Zaps, transfer ownership, and export a CSV. Zapier documents that in [Safeguard business-critical Zap workflows when a team member transitions](https://help.zapier.com/hc/en-us/articles/16534140057485-Safeguard-business-critical-Zap-workflows-when-a-team-member-transitions), updated 29 May 2026 when checked. The follow-on article, [Remove members from your Team or Enterprise account](https://help.zapier.com/hc/en-us/articles/8496281082253-Remove-members-from-your-Team-or-Enterprise-account), updated 27 August 2026, is the one to read before anyone is removed. A few details are easy to miss: - Private app connections can be set to stay active for 30 days, or until the app expires them. - Webhook URLs that use Catch Hook or Catch Raw Hook change when ownership changes, because the URL is tied to the owner. You must update the other apps that call that URL. - If an admin removes a user, Zapier transfers the Zaps to the account owner. The admin cannot pick a different new owner. - The CSV link in the confirmation email lasts 7 days. - On a SCIM setup you remove the user in your identity provider, and private connections stay active for up to 30 days by default. Make and n8n have their own ownership models. Do not assume the Zapier article covers them. Before a freelancer's last day, export what you can, move connections into a company user, and change any webhook the outside world calls. If the zap was built inside the freelancer's personal account, you may not have a transfer button at all. That is the expensive version. Build in your account from the first task. The interview question for this is in [21 interview questions](https://blog.rekruuto.net/ai-automation-specialist-interview-questions). ## What should an automation runbook include? One page per workflow. If it does not fit on a page, the workflow is probably two workflows. Copy this and fill it. Leave a line blank rather than inventing an owner. | Line | Write this | | --- | --- | | Name | What a colleague would call it | | Purpose | The business outcome, in one sentence | | Trigger | The event that starts it | | Tools | Each app, and which company account | | Steps | The short path, including any model prompt location | | Rules versus AI | What is a formula, what is a draft | | Human checkpoint | Who approves, and what they look for | | Owner and backup | Two names. One is not enough | | Alert | Where a failure shows up, and who is pinged | | How to pause | The exact click, so a non-builder can stop it | | Common fixes | The three failures you have already seen | | Monthly cost | Tool seats plus model usage, from the invoices | | Last reviewed | The date someone ran a test record | Store the page next to your other procedures. The format in [virtual assistant SOPs](https://blog.rekruuto.net/virtual-assistant-sop) is the right home. A runbook that lives in the builder's laptop is not a runbook. Review the date monthly. A page nobody opens is how a renamed CRM field survives unnoticed. ## How much ongoing maintenance do AI workflows need? Plan a weekly check, then size the seat to the backlog. A handful of stable workflows can sit with an assistant who already has the runbooks. A changing backlog fits a full-time owner. Plan on four kinds of upkeep. - Prompt drift. The same prompt gets worse as the inputs change, or a model update shifts the style. Re-run a small set of old examples after any model change. - App changes. Read the failure log when a vendor ships an update. Fix the mapping. Do not wait for a customer. - Cost. Open the provider bill. A workflow that was cheap at ten runs a day is a different product at a thousand. Caps belong in the dashboard. - Documentation. Update the runbook in the same week as the change. Future you will not remember the filter. A project freelancer will do this only if you pay a retainer for it. A full-time specialist from $2,500 a month has it in the job. An AI-skilled assistant at the $1,800 starting price can watch a small set and escalate breaks, if the workflows are simple and the runbooks exist. Pick the seat that matches the count of live workflows, not the excitement of the first demo. The vendor choice is in [specialist vs consultant vs agency](https://blog.rekruuto.net/ai-implementation-specialist-vs-ai-consultant-vs-agency). ## Is a project build or a full-time owner safer? A project build is safer for one workflow with a real end. A full-time owner is safer when workflows stay live and keep changing. The unsafe option is a project build with no owner and no runbook. Use the orphaned-automation check twice a year, and whenever someone who builds workflows leaves. - List every live workflow. If you cannot list them, that is the finding. - For each one, name the company account that owns it. Personal accounts fail this check. - Name two humans. If you can only name the person who just resigned, it is orphaned. - Confirm a failure alert goes to a shared channel, not to one inbox. - Open the runbook. If the pause steps are missing, it is not finished. - Check the last monthly cost against the cap you set. - For Zapier, confirm you could transfer ownership using Zapier's own help article, because you are on a plan that allows it. - Change any webhook that would break if the owner changes. Do this before the person leaves, not after. - Remove access you no longer need. The habits in [data security for remote teams](https://blog.rekruuto.net/top-strategies-for-data-security-for-remote-teams-protect-your-business) apply here. - Decide: keep, pause, or rebuild. Do not leave a zombie zap "just in case". If more than a handful fail the two-human test, stop new builds until the list is clean. More automation will not fix an ownerless stack. ### A weekly maintenance half-hour Put a recurring half-hour on the owner's calendar. In that slot they open the failure log, re-run one old example through any workflow that uses a model, glance at the provider bill, and change the "last reviewed" date on any runbook they touched. If nothing failed, they write "nothing failed" in the same channel. Silence is how a dead zap hides. Skip the half-hour and the runbook becomes folklore within a month. The half-hour is also the moment to pause something that no longer matches the process, instead of leaving it on because it once saved a click. New ideas go on the backlog. They do not jump the queue during maintenance. ## Frequently asked questions ### Why do automations fail after they worked in the demo? Because apps, logins, fields, and model behavior change. A demo uses a happy path. Production meets the next Tuesday. ### What should I do before a freelancer leaves? Move the workflows into a company account, export what the tool allows, update webhooks, and make sure two people can pause each one. On Zapier, use the official transfer steps. Do not wait until the user is already removed. ### What goes in a runbook? Purpose, trigger, tools, steps, who approves, two owners, how you hear about failure, how to pause, common fixes, monthly cost, and the date you last tested it. ### How much time does maintenance take? Size it to the live count. A changing backlog fits a full-time owner better than a one-off project. A stable set with written runbooks can sit with the person who already runs the tools. ### Who should own maintenance after launch? A named person on your side, with a second person who can pause each workflow. A full-time specialist fits when the backlog keeps changing. A project fee fits when the build has a finish line. ## Get matched with your hire If the orphan check produced a list with one name on it, [get matched with your hire](https://www.rekruuto.net/va-strategy-call) and look at an [AI Implementation Specialist](https://www.rekruuto.net/ai-implementation-specialist) as the second name. The [1-week trial with replacement or refund](https://www.rekruuto.net/risk-free-va-trial) should include a runbook for one live workflow, not only a new demo. Pricing: Most roles $1,800/mo (£1,395 + VAT in the UK, €1,650 in Europe). Developers and AI Implementation Specialists from $2,500/mo (from £1,945 + VAT in the UK, from €2,295 in Europe).