· 6 min read· Milo Works

5 Signs Your AI Automation Will Break in 90 Days

Most AI projects fail within 90 days. Not because the tools broke — because nobody was watching. Here are the five warning signs we see before every failure.

Systems ThinkingWhen NOT to Automate
Jump to article

Your Automation Is Probably Going to Break. Here Is How to Know.

Most AI automations built for small and mid-size businesses fail within 90 days. Not because the technology is bad, but because nobody planned for what happens after the build.

We have seen hundreds of broken automations across companies of every size. The failure patterns are remarkably consistent. Here are the five warning signs that your automation is on borrowed time.

1. Nobody Is Watching It

This is the most common and most destructive problem. Your automation was built, it worked for two weeks, and everyone moved on. Nobody checks whether it is still running. Nobody gets alerted when it fails.

We call this the "set it and forget it" trap, and it is responsible for more wasted money than any technical failure.

What it looks like in practice:

  • Your Zapier dashboard has 14 zaps. Three of them have been erroring for six weeks. Nobody noticed because nobody logs into Zapier.
  • Your lead routing automation silently stopped working when HubSpot updated their API. Leads have been sitting unrouted for 11 days. Your sales team thought it was a slow month.
  • Your invoice automation processed a duplicate payment because the deduplication logic broke and there was no alert.

The fix: Every automation needs monitoring. Not "we will check on it sometimes" monitoring. Actual alerts that notify a specific person when something fails. If you do not have this, your automation is a liability, not an asset.

2. There Is No Documentation

Ask yourself: if the person who built your automation disappeared tomorrow, could someone else understand how it works?

For most companies, the answer is no. The automation lives in one person's head, in a Make.com scenario with 47 nodes and no labels, or in a Notion page that was last updated eight months ago.

What it looks like in practice:

  • Your operations coordinator built a complex n8n workflow. They left the company. Now you have a flow that does... something... and nobody is sure which parts are critical.
  • A freelancer set up your Zapier automations. The zaps reference Google Sheets that reference other Google Sheets. The logic is spread across four platforms. There is no map.
  • Your CRM automation has 12 conditional branches. Each one was added over time to handle edge cases. Nobody remembers why branch 7 exists or what triggers it.

The fix: Documentation does not have to be elaborate. A simple diagram showing what triggers the automation, what it does, what systems it touches, and what the expected output is. If you cannot produce that document in 10 minutes, your automation is at risk.

3. The Consultant Disappeared

You paid someone $5,000-$15,000 to build your automations. They did good work. Then they delivered it, sent a final invoice, and vanished.

This is the build-and-abandon model, and it is the default in the automation consulting industry. The consultant's incentive is to build and move on. Your incentive is to have something that keeps working. These incentives do not align.

What it looks like in practice:

  • You email your automation consultant about a bug. They respond three days later with a quote for additional work.
  • The consultant built everything on their personal accounts. You do not have admin access to your own automations.
  • Your automation breaks because a third-party API changed. Nobody on your team knows how to fix it. The consultant is booked out for six weeks.

The fix: Automation is not a project. It is an ongoing operation. Either build internal capacity to maintain your automations, or work with a partner who includes monitoring and maintenance in their model.

4. It Was Built on a Shifting Process

You automated a process that was not stable yet. The team was still figuring out the workflow when someone decided to automate it. Now every time the process changes, the automation breaks.

What it looks like in practice:

  • Your client onboarding automation sends emails in an order that no longer matches your actual onboarding process because you changed the steps in month two
  • Your lead scoring automation uses criteria that your sales team abandoned three months ago
  • Your reporting automation pulls from a dashboard that has been reorganized twice since the automation was built

The fix: Automate stable processes. If a workflow has changed more than twice in the last quarter, it is not ready for automation. Stabilize it first with a manual checklist, run it consistently for 30-60 days, then automate.

5. Nobody Validated the ROI

The automation was built because someone said "we should automate this." Nobody measured what the process cost before automation. Nobody measured what it costs after. Nobody can say whether the automation actually saved money.

What it looks like in practice:

  • You are paying $300/month in platform costs for a Zapier automation that saves one person 20 minutes per week. That is $3,600/year to save $2,600 worth of time.
  • Your lead nurture automation sends emails, but nobody has checked whether those emails actually convert better than the manual approach.
  • You automated a report that three people used to spend time on. Two of those people still manually check the automated report because they do not trust it.

The fix: Measure before and after. Time spent, error rates, throughput, revenue impact. If you cannot quantify the benefit, you cannot justify the cost — and you definitely cannot optimize something you are not measuring.

The Graveyard Is Full

Every company we diagnose has what we call "the graveyard" — abandoned automations still running in the background, half-built workflows in Make.com, Notion databases that were supposed to be automated but never were, Zapier zaps that error silently.

The graveyard is not just wasted money. It is active risk. Broken automations send wrong data, miss deadlines, and create problems that take more time to fix than the automation was supposed to save.

What to Do About It

Audit what you have. For every automation currently running in your business, answer these questions:

  • Is someone monitoring it for failures?
  • Is it documented well enough that someone new could understand it?
  • Is the underlying process stable?
  • Can you quantify the ROI?

If any answer is no, that automation is a ticking clock.

Want us to do the audit for you? Book a diagnostic call. We will inventory your current automations, identify what is at risk, and give you a prioritized plan to fix or retire what is broken. No pitch, no pressure — just an honest assessment of where you stand.

Or start with the ROI calculator to see whether your existing automations are actually delivering value.

Next step

Want to see if automation makes sense for your business?