AI Automation for Small Business

5 Questions to Ask Your IT Provider

Your AI automation request has been “in scoping” since June. You mentioned it on a monthly call. Your IT provider said they’d look into it. Someone made a note. A few weeks passed. Then a few months. Now the project exists in a strange corporate limbo: discussed often, owned by nobody, and somehow always one meeting away from movement. This is a common scenario of AI automation for small business right now. You made a reasonable call. Your managed service provider already has the keys to the kingdom. They run your network, know your users, handle security, pick up the phone, and send invoices that are at least mildly familiar. Of course you asked them first. Before you hand over the project—or another batch of billable “discovery” hours—ask five questions.

Kaseya’s 2026 State of the MSP Report found that 48% of MSPs rank AI and automation as the top client need for the year. Just 13% identify AI and automation as a meaningful revenue source. Translation: demand obvious, but delivery is rare.

AI and IT Are Different Jobs

Keeping 90 endpoints patched, backed up, secured, and out of the local news ransomware segment is hard work. Good MSPs earn their keep doing it. But that work is built on repeatability:

• The same endpoint agent across clients.
• The same backup stack.
• The same patch policies.
• The same ticket workflows.
• The same playbook when someone clicks a cursed attachment labeled Invoice_FINAL_reallyfinal.zip.

Standardization is the point. It’s how an MSP supports dozens of businesses without hiring an army every quarter.

Building an AI workflow for your business is a different trade.

Say you want an automation that reads quote PDFs, extracts project details, verifies the data, and pushes it into your CRM. That’s not “turning on AI.” It is custom process design. Someone needs to watch how your estimator works, uncover the spreadsheet workaround everyone forgot about, connect systems through APIs, set up guardrails, test exceptions, and keep the workflow alive when one vendor changes a database column name six months later. Some MSPs can absolutely do this work, some are building the capability now, and others are reselling licenses with a fresh coat of AI paint.

These questions will help you tell the difference before your automation project becomes a long-running subscription to optimism.

5 Questions to Ask Your IT Vendor

1. What have you built like this…and can I talk to that client?

A strong answer includes a real company, a named process, a rough project size, an outcome, and ideally a customer reference.

You are looking for something like: “We built an intake workflow for a 75-person construction company. It reads subcontractor documents, flags missing insurance details, and updates their project system. It took six weeks. Here’s what changed, and here’s a client contact who agreed to speak with prospects.” That is evidence.

Be wary of an answer made entirely of vendor badges, platform names, or certifications. Microsoft Copilot, ChatGPT Enterprise, Power Automate, and other platforms can be useful. But deploying a license or enabling a feature is not the same as building the workflow you asked for.

There is a big gap between:

“We can roll out Copilot.”

and:

“We can automate the handoff between your sales inbox, PDF quotes, CRM, and accounting system without creating a new species of spreadsheet chaos.”

If you would be their first AI project, that is not a dealbreaker. It just means the price, scope, and risk-sharing should reflect reality. Don’t pay premium rates to be somebody’s training montage.

2. Who writes the code, and what is their name?

Ask this question plainly. Then wait. A good answer names a person or a specific team. It also explains if any part of the work is subcontracted.

Outsourcing is not inherently bad. Plenty of excellent projects involve a specialist developer or automation partner. The problem is the foggy version: your MSP sells the work, another firm builds it, a third party hosts it, and when it breaks, everyone suddenly develops a keen interest in forwarding email threads.

You want a clear chain of responsibility:

• Who designs the workflow?
• Who builds the integration?
• Who tests it?
• Who supports it after launch?
• Who is accountable when it fails on a Tuesday morning?

“Our team will handle it” is not a valid answer. Ask who, specifically. You are hiring people to maintain a business process.

3. What happens when an API changes?

Most buyers skip this question. It is also the one that decides whether your AI automation is useful in 18 months or sitting quietly in a digital ditch.

Every workflow depends on someone else’s software. Your CRM changes its API. Your ERP vendor updates permissions. A model provider retires a feature. A field gets renamed from Customer ID to Client Reference because apparently software companies enjoy scavenger hunts. Then the automation breaks.

A good provider will tell you:

• What monitoring exists.
• How quickly they respond to failures.
• Whether maintenance is included or separately billed.
• Who pays for changes caused by third-party platforms.
• Where the documentation lives.
• Whether another qualified firm could take over if the relationship ends.

That last bullet point matters more than vendors like to admit. Your business should own its process documentation, integration credentials where practical, workflow logic, and access to the systems it pays for.

If the answer is “that should be covered by your existing retainer,” get it written down. Custom workflows are not the same thing as resetting passwords and patching laptops. Do not wait until the process is down to discover that “support” means “we can schedule a call next Thursday.”

4. What is the fixed price, and exactly what does it buy?

A fixed price is not just a budgeting preference. It is a test of whether the provider understands the project well enough to define it. You want a written scope that states:

• The one process being automated.
• The systems being connected.
• The inputs and outputs.
• The human approval points.
• The success metric.
• What is explicitly out of scope.
• The implementation price.
• Any ongoing support cost.

Be careful when a provider suggests running the project against your existing block hours or general managed-services retainer. That can work for a tiny, tightly bounded task. For a real workflow project, it can turn your IT budget into an R&D tab with no ceiling and no finish line.

A four-week automation should not become a recurring monthly expense that nobody can explain after the original champion leaves. Fixed scope creates an ending. Endings are underrated.

5. What date will it go live (in production)?

Not the date there will be a demo, the date the project enters “pilot,” or the date someone says, “We’re making excellent progress.” Ask for the date your staff will use the workflow to do real work. A good answer gives you a specific go-live date, names the business process, and identifies an owner on your side. That owner is crucial. Automation projects fail when they are treated as an IT project with no business owner who can answer questions, approve changes, and decide what “good” looks like.

For a first AI automation project, keep the scope small enough to reach production in 30-60 days. If the project needs six months of discovery, it is probably too big or vague. Cut the scope until it fits. Automate one painful, repetitive process. Measure one outcome. Get the win. Then expand.

What to Do With the Answers

If you get five solid answers, you may already have the right partner. Give them the project.

If you get five vague answers, your MSP may still be excellent at the job you hired them to do. Keeping infrastructure stable and building custom AI workflows are both valuable skills. They are not interchangeable.

Mixed answers are more common (and more workable) than either extreme. Your MSP can own security, access, infrastructure, and operational context while an AI specialist builds your workflow. That is not a failing relationship. It is partners recognizing that not every vendor needs to cosplay as a full-stack innovation lab.

AI automation for small business succeeds when someone owns the workflow, someone can name the business result, someone agrees to a real price, and someone puts a go-live date on a calendar. Everything else is just another “strategic initiative” slowly fossilizing in meeting notes.