What 'fully managed' actually means, and what it does not

Fully managed does not mean you pay a fee and walk away.

It means someone else runs the machine. You still decide where the machine goes.

Business owners hear the phrase and imagine a quiet office. No one chasing the same spreadsheet. No one copying data between three tabs. No one typing the same email for the fourth time today. That part is real. The part that gets blurred is what happens before the quiet sets in.

The six things the fee should cover

A fully managed AI automation service is a bundle of ongoing work. When you write the cheque, you are buying six distinct activities. If any one of them is missing, the word "managed" is being stretched.

Infrastructure. Someone has to own the servers, the containers, the environment where the automation lives. This includes handling credentials, environment variables, rate limits and uptime. If the automation breaks at 2 a.m., the infrastructure owner wakes up. Not you.

Model choice and optimisation. An automation that sends customer replies is different from one that reads invoices. The model matters. A fully managed service selects the model for each task, tests it against your actual data, and switches when a better option appears. You do not need to know the difference between GPT-4o, Claude 3.5 Sonnet and a fine-tuned open-source model. You need to know the output is accurate. Someone else makes that true.

Monitoring. Automations decay. A PDF format changes. An API endpoint moves. A prompt that worked in January drifts by March. Monitoring is the practice of watching outputs, catching failures before the client notices, and logging every error. If the service cannot show you a dashboard or a weekly health report, they are not monitoring. They are hoping.

Support. When your operations lead notices something odd, she sends a message. A human reads it, diagnoses the issue, and fixes it. That human is part of the fee. Support is not a chatbot. Support is a person who knows your automations by name.

Maintenance. This is the unglamorous sibling of building. Building an automation is fun. Keeping it running across six months of small upstream changes is the actual job. Maintenance means patching libraries, updating model endpoints, adjusting prompts when the underlying model behaviour shifts, and refactoring flows that grew brittle. A fully managed fee includes this as standard. A build-only fee does not.

Handover documentation. One day the relationship may end. When it does, you should receive a folder that lets another team pick up the work. Handover documentation is not a nice extra. It is a signal. A provider that refuses to document what they built is a provider that intends to keep you dependent. Fully managed means fully documented.

What stays on your desk

Now the harder part. Even a fully managed service does not remove every decision.

You still decide which process gets automated first. A good provider will help you map your workflows and identify high-return candidates. But the ranking is yours. Automating the wrong process is worse than automating nothing, because it trains the team to distrust the tool. No external partner can know that a particular report feels trivial but keeps a key client calm. You know that. You choose.

You still decide what "correct" means. An automation that drafts customer responses needs a standard. Is a reply acceptable if it is factually right but sounds cold? Is speed more important than tone? These are business judgments. A provider can tune the model to match your answer. They cannot supply the answer.

You still decide when to intervene. A fully managed automation runs without you. But it should also have a handbrake. For high-stakes outputs (a contract clause, a pricing quote, a termination letter), the flow should pause and wait for a human. You define those pause points. The provider builds them. The responsibility for the final output does not transfer to the code.

You still own the relationships. An automation can draft a follow-up email to a client who went quiet. It cannot read the history of that relationship. It does not know that the client's founder just left, or that they prefer calls to emails, or that they once saved your business with a late payment. Those facts live in your team's memory. The automation is a tool. The relationship is yours.

The quiet test

There is a simple way to tell whether a service is genuinely managed.

Ask this: "If I do nothing for two weeks, will everything still work?"

A build-only provider will say no. They will point out that models drift, that someone needs to check outputs, that you should assign an internal owner. All of that is honest. But it is not fully managed.

A fully managed provider will say yes. With one caveat. "Yes, everything we built will keep running. We will monitor it, fix it, and report to you. If a business decision is needed (a new product line changes your categories, a regulation shifts), we will flag it for you. The decision itself will wait for your answer."

That distinction is the whole definition. Fully managed means the machine is not your problem. The direction of the machine still is.

The price signal

Monthly retainers for this work cluster around a predictable band. The low end is a freelancer who builds and hopes. The high end is a large consultancy that charges for logo prestige. The middle is where the work actually happens.

A fixed monthly fee, somewhere in the low four figures, that covers build plus all six ongoing activities, is a market signal that the provider has done this enough times to know what breaks. It is not cheap. It is also not the cost of one full-time hire, who would still need tools, training and management.

When you see a fee far below that band, ask which of the six activities are excluded. You will usually find that monitoring and maintenance are absent. Those are the two that matter most after month one.

This week

Pick one process in your business that repeats at least five times a week. Something small. Invoice data entry. Lead qualification from a web form. A weekly report that pulls from three sources.

Write down who currently does it, how long it takes them, and what they would do with that time if the process vanished.

Then write down one decision inside that process that you would never hand to software. "Approve the final invoice total." "Check the lead's industry before routing." "Review the report summary before sending."

That list is your boundary. Everything below the boundary is what a fully managed automation can absorb. The one decision above it stays with you.

You now have a definition you can use. Not a slogan. A checklist. Infrastructure, model choice, monitoring, support, maintenance, handover. Six things you pay for. One set of decisions you keep. When a provider says "fully managed," hand them the list and watch their face.

Some will nod and walk you through each item. Those are the ones who mean it., -

At Nexibeo, we automate 1 to 2 larger processes or 2 to 3 smaller ones each month for a fixed fee of $1,900 a month (or $19,000 a year). Infrastructure, model choice, monitoring, support, maintenance and handover documentation are all included. The decisions stay with you. If you want to see what that looks like for your specific workflows, we offer a free process review.