Managed AI services vs an AI consultant: which one your firm needs

If you own a service firm in Canada and you've started shopping for help with AI, you've likely met two kinds of offer. One is an AI consultant: a project with a scope, a timeline and a handoff. The other is a managed AI service: someone who builds the systems and then runs them for you, month after month, the way a managed IT provider runs your laptops and email.

Both are legitimate. They solve different problems, and they leave you holding different things when the work is done. One reliable way to waste money on AI is to buy one when the firm needed the other.

What an AI consultant actually delivers

AI consulting is usually a defined engagement. A typical one looks like this:

• An assessment. Interviews, a look at your systems, a list of opportunities.
• A roadmap or recommendation. Which tools, which workflows, in what order.
• Sometimes a build. A proof of concept, an automation or a configured tool.
• Training and handoff. Documentation, a session with your team, and the keys.

Then the engagement ends. That is the point of it: you pay for expertise for a period, and you keep the result.

A consultant is a good fit when the problem is well defined, when you need a strategy question answered (should we do this at all, and where first), or when you have someone in-house who will own the system after the handoff. That last condition matters more than any other.

What a managed AI service actually delivers

Managed AI services include the build, then keep going. A reasonable one covers:

• Running what was built. Watching it, fixing it when a supplier changes a form or someone renames a folder, handling the cases it can't.
• Keeping it current. Model and vendor changes, updated price books and templates, new staff added to the right accounts.
• New builds over time. Once the first workflow is steady, the next one on the list gets built.
• Reporting. A regular, plain account of what it did and how many hours it returned.

Managed is a good fit when nobody on your team has the time or background to look after automations, when the repeat work is spread across several workflows, and when you want results measured rather than described. The risks are an ongoing fee and dependence on the provider. Both can be managed, and the questions below cover how.

How the cost structures differ

We won't quote prices here, and the market range is wide. The shapes of the costs are what matter when you compare.

Two costs hide in both models. The first is software and usage: model subscriptions, API usage and any new tools. Ask whether they are in the fee or billed to you, and on whose account. The second is your own team's time. A consultant's project can look cheaper until you count the hours your office spends maintaining it, or the second engagement to fix what drifted. When you compare, compare two years, not the first invoice.

The question that decides it: who runs this in month seven?

AI systems aren't a one-time install. Inputs change, models get retired, your price book changes, and a system that worked well in spring is quietly worse by fall. Someone has to notice and fix it.

Plenty of owner-led firms with 10 to 100 people don't have that someone. There is an outside IT provider for the network and a bookkeeper for the books, and nobody whose job is to keep an automated quoting workflow honest. If that describes you, a consultant's handoff lands on the busiest person in the office, and the system decays. If you do have an operations or systems person with time to own it, a well-run consulting project can be the better buy.

We covered this and six other buying questions in Seven questions to ask anyone selling you AI.

A short decision guide

Lean towards an AI consultant if:

• You have one clear problem, such as choosing an estimating tool or setting a usage policy.
• Someone in-house will own the result and has the hours to do it.
• You want a second opinion on strategy before committing to anything ongoing.

Lean towards a managed AI service if:

• Repeat office work is spread across intake, quoting, reports and invoicing.
• Nobody inside has the time to watch and fix automations.
• You want a baseline and a monthly number, not a slide deck.

A hybrid is reasonable too: a short consulting engagement to set direction, then either a managed service or an internal owner to run it. What doesn't work is a project with no owner afterwards.

Questions to ask either one

1. What is my baseline? How many hours does the work take today, and will you measure it before you build anything?
2. What is in the fee, and what is extra? Get software, usage, new builds and support hours in writing.
3. What is the term, and how do I leave? Know the minimum and what happens on exit.
4. Who owns the accounts, automations and documentation? The answer should be you, under your admin.
5. What waits for a person before it goes out? Quotes, invoices and client reports should.
6. Where does our data go, and is any of it used for training? Many AI models aren't hosted in Canada, so ask where processing happens rather than assume. Ask how they handle personal information under PIPEDA or the provincial laws in Alberta, BC and Quebec that apply instead of it.
7. How will I know it's working? Ask what the report looks like and how often you get it.
8. Who do I call when it's wrong? A name and a response time, not a ticket queue.

Red flags in either model

• No measurement before the build. Without a starting number, any result is a story.
• A platform you can't leave. If the automations only exist inside the provider's account, you're renting your own processes.
• A roadmap with no owner. A consulting report that names twelve opportunities and nobody to run them is shelf-ware.
• A monthly fee with nothing new in it. A managed service should show what it did this month: fixes, new builds, hours returned. If the report is the same every month, ask what you're paying for.
• Everything sent automatically. Speed is not the goal if a wrong figure reaches a client.

Where we sit

LaunchBird is a managed AI service. We start with a week on us: a time study of your team's repeat work, a baseline you sign, and a first build live in your own accounts. After that we run it monthly, add builds from the list the study produced, and report the hours in one page. Nothing reaches a client without a person approving it, and you own everything we build.

That isn't the right shape for every firm. If you have one defined problem and someone ready to own the answer, hire a good consultant. If you're not sure which you need, see what the work looks like for your industry, or book a call and we'll tell you honestly.