What is a managed AI service for a small business?

A managed AI service is an arrangement in which the supplier builds an AI worker for a defined job in your business and then runs it for you: monitoring it, tuning it, updating it, handling what it escalates, reviewing it with you each month and fixing it when something breaks. You pay a one-off build fee, then a monthly fee for all of that. You do not buy software to operate yourself, and you do not need anyone on your staff who knows how the AI works. In IT the same phrase can also mean hosting and running AI models for a company's own engineers; this page is about the other meaning, where the supplier runs a piece of the business's work.

Why does "managed" matter to a small business?

IT support is bought this way. Few owners run their own email servers. They pay an IT support company a monthly fee, and when something fails it is a ticket and not the owner's weekend. The owner buys the result and somebody else owns the upkeep.

An AI subscription often works the other way round. You take out a subscription, and from then on the work is yours: writing the instructions, connecting the systems, noticing when the answers go wrong, repairing it when a supplier changes a screen.

A managed service puts the upkeep with the supplier, and the upkeep is what decides whether the worker is still doing the job a year later.

What does the monthly fee cover?

• Monitoring. Somebody other than you watches that the worker is running and catches it when something goes wrong.

• Tuning. When the worker gets something wrong, or your team finds a better way to word a reminder, its instructions are changed.

• Updates. The systems a worker uses change, and so do the AI models. Keeping the worker working through those changes is the supplier's job.

• Escalation handling. When the worker meets something it cannot settle, it goes to a person, and the reason is looked at so that it happens less.

• A monthly review. The worker's KPIs, as written in its role specification, are reviewed with you each month.

• Support. If something is wrong, you raise it and it is fixed.

• The cost of running the AI model. It is included in the fee. You do not hold a separate AI subscription for the worker.

What does it look like day to day?

On most days there is nothing to notice. The worker operates inside the tools your team already uses. An insurance broker's account handler finds the renewal reminders drafted and waiting for approval. A practice manager at an architecture firm gets the list of timesheets still missing.

The management shows on the bad days. A supplier changes its portal and the worker can no longer read it. With a subscription, finding and repairing that falls to you; with a managed service, the supplier finds it and fixes it.

What stays with you?

A managed service is not a way of handing over responsibility for your business. Three things remain yours. A named person approves the actions that carry consequences, such as a message to a customer or a change to a record, and every decision a regulator would care about stays with a person. Your team tells us when the job changes: a new product, a new rule, a new form. And you judge whether the worker is worth having, from the KPIs you signed.

How is it different from software, a consultancy or outsourcing?

• A software subscription. A subscription gives you a tool and leaves the work of using it with you. A managed service gives you the job done and leaves the tool with the supplier.

• A consultancy project. A consultant advises, or builds something and hands it over, and the engagement ends. A managed service has no handover: the supplier is still accountable for the worker in month ten. Our comparison of an AI consultancy and a managed service goes further.

• Outsourcing. An outsourcer does your process with its own people, usually in its own systems. A managed AI worker does the job in your systems, under your rules, with your people approving what matters.

• Managed IT. The buying pattern is the same, but an IT provider keeps your equipment and accounts working. A managed AI service does a piece of the business's own admin. If you run an IT support business yourself, our page for MSPs and IT support shows where the two meet.

• A temp or a virtual assistant. Both are people, engaged by the hour. A managed AI worker's job is described in finished tasks, not in time. See what counts as a task.

How long does it run?

At The AR Dept your first AI worker starts on a 90-day probation contract, measured against the KPIs in the role specification. At day 90 you can walk away with no ongoing commitment, whatever the figures show. If you stay, it rolls into a standard 12-month managed service. The management described on this page starts on the first day of the probation, not after it. Our starting monthly fee is on the homepage.

What should you ask an AI provider before signing?

These questions show the difference between providers:

1. Have they set up this kind of job in a business like yours? The work and its rules differ by trade, and a regulated trade has its regulator's requirements on top. Ask for examples in yours.

2. How do they deliver? A good provider can tell you what happens in the first week, who owns what, how success is measured and what happens if the work is not up to standard.

3. How do they define success, and when? The measures should be agreed before the work starts: what the task takes today, what the target is and when it will be reviewed. An answer such as "efficiency improvements" or "time savings" is a warning sign.

4. Who is responsible if something stops working three months in? The answer should name the provider.

5. What will they not take on? A specific answer is the trustworthy one. A provider who says it can automate any process should be treated with caution.

Be wary of talk of transformation before anyone has looked at your work, of anything that needs significant technical skill of your own to run or maintain, and of a contract that commits you before the work has been tested in your business.

Before the first conversation, answer three questions about your own business. Which admin task, if it were taken off your team or cut back sharply, would make the biggest difference to your capacity or margin? Does everyone do it the same way today? And who in the business would own the relationship with the provider?

"Managed" is the half of that idea people skip. Our explainer on what AI staff are covers the rest.

A managed service would give your business one defined job done in your own systems, with the upkeep and the fixes on our side and the approvals and the verdict on its KPIs on yours. If you took out an AI subscription and ended up maintaining it yourself, talk to our agent about handing the upkeep over.