What does ongoing AI workflow maintenance actually cost, and what are you paying for?
Ongoing AI workflow maintenance typically costs $249 to $499 per month for a small business running one to three automated workflows. That fee covers watching the workflow, alerts when a run fails, fixes when a connected app changes, the AI usage costs, and adjustments as your business changes.
The build is the cheap part. Workflows that connect to software you don't control need someone watching them, because those connections keep changing and the failures are quiet rather than obvious.
Why does an AI workflow need maintenance at all?
Because almost none of the pieces are yours to control.
An AI workflow is mostly connections to other people's systems: your CRM, the AI provider, an email service, a data source. Each of those changes on its own schedule, without telling you.
In practice that means:
- Your CRM starts requiring a field it didn't need before
- The AI provider retires the model version you were using
- An email provider tightens sending limits
- A supplier changes the format of the file they send you
- You hit a usage limit for the first time, because volume grew
Each change is small on its own. Together, they mean the workflow you built in March is running against very different systems by September.
The workflow doesn't stop. That's the problem. It keeps running, and the results quietly get worse.
What does a quiet failure actually look like?
Here is a concrete example, because the general description makes it sound smaller than it is.
A workflow adds company details to your CRM records. In month four, the data provider changes how it reports company size, sending text where there used to be a number. The workflow doesn't crash. It just leaves the field empty.
Every record enriched from that point has a blank where a value should be.
Nobody notices, because nobody checks a field that has always been right. Three months later someone runs a report, the numbers are wrong, and now there's a data cleanup project plus every decision made on bad data in between.
Caught in week one, it's a small settings change. Caught in month six, it's weeks of cleaning up bad data. The Humanoid.Guide publishing workflow is a live example of this kind of system: it runs every day against outside sources that change without notice.
What is actually included in the monthly fee?
| Included | What it means in practice |
|---|---|
| Alerts when a run fails | A failed run sends an alert straight away. You hear about it from the system, not from a customer. |
| Monitoring | The person who built the workflow watches it, alongside the approval steps and guardrails agreed at launch. You deal directly with the person who built it: no ticket queue, no handoff to someone who has never seen it. |
| Fixes when a connected app changes | When one of the apps in the chain changes, fixing it is our job. Ideally you never find out it happened. |
| AI usage costs | Included up to your plan's fair-use limit. On higher tiers you can use your own AI provider account, so the spend sits with you and you can see it directly. |
| Changes | Businesses don't stand still. A workflow built for how you worked in spring needs adjusting by autumn. |
What does it cost?
| Plan | Onboarding | Monthly | Roughly per year |
|---|---|---|---|
| Pilot | $399 | $249 | ~$3,400 |
| Team | $899 | $499 | ~$6,900 |
What each plan includes, and the custom tier for larger jobs, is on the pricing page.
When is it cheaper to do this in-house?
Sometimes it genuinely is, and it's worth saying so.
The alternative to a maintenance fee is someone on your own team owning the workflow. Not full-time, but they need to understand how it works, notice when it starts going wrong, and have the time to fix it when a connected app changes.
In-house makes sense when:
- Someone on the team already understands the process and the tools
- That person has genuine spare capacity, not theoretical spare capacity
- You want the knowledge held internally
- You have more than a handful of workflows, so one internal owner covers a lot
Outsourcing makes sense when:
- The person who'd own it is already your most stretched person
- You have one or two workflows, not twenty
- Nobody on the team has built AI workflows before
- If everyone assumed someone else was watching, nobody would notice for months
The trouble is the in-between case: nobody really has the time, but the workflow gets assigned to someone anyway, and it becomes the thing nobody is actually watching.
Frequently asked questions
Is AI workflow maintenance a support contract?
No. A support contract responds when you report a problem. Maintenance is meant to catch the problem before you'd notice. Most failures here are silent, so waiting for a report means waiting months.
What happens if nothing breaks for six months?
Then the fee bought you six months of nothing going wrong, which is the point. Workflows that depend on other companies' software don't hold still for six months, so a quiet half-year is the result of the work, not a sign there was nothing to do.
Are model costs included?
Yes, up to a fair-use limit on each plan. Higher tiers let you use your own AI provider account, so the spend sits with you and you can see exactly what is used.
Can I cancel and keep the workflow running?
You keep your data and the settings; what you keep if you leave covers this in detail. Running it yourself means taking on the monitoring and maintenance described above, which is a real job rather than a formality.
How do you find out something broke?
Failed runs send an alert. The harder case is when the workflow keeps running but produces slightly wrong results. That is what regular review is for.