“AI consultant” can currently mean everything from someone selling workshops to an engineer building custom infrastructure. For a small business, the useful distinction is much simpler: can your team set this up safely with the tools you already have, or has the workflow become complicated enough that implementation & maintenance expertise will save you more than it costs? At Unprompted, we think outside help should earn its way into the project.

What you build in-house vs. when outside help is worth it

A lot of the first steps should stay in-house.

Turning on a feature you already pay for is one. If your accounting software can send invoice reminders or your CRM can summarize calls, you should not need a consulting engagement just to test it.

Simple two-tool automations are another: form submission → CRM record, payment received → bookkeeping update, calendar booking → Slack notification. No-code tools were built for this kind of work.

Testing off-the-shelf software also belongs in-house. Name the job, put the product on real work, and decide whether it earns the subscription.

Even identifying your first opportunities can be done internally. A small team can list repetitive workflows, estimate the time they consume, and score the strongest candidates without hiring anyone.

Outside help becomes more useful when several things are true at once: the workflow crosses three or four systems, the connections are not straightforward, the logic is specific to your business, mistakes could affect customers or important records, nobody internally can maintain the technical side, or your team has already spent weeks on a pilot that is still “almost working.”

The original research for this article cites MIT’s 2025 work showing stronger reported success rates for AI solutions purchased from or implemented with external specialists than for internal builds. We would not read that as “outsource everything.” The more useful lesson is that experience starts to matter as the workflow gets harder.

How much does an AI automation consultant cost?

Before you talk to a provider, write a one-page brief.

Describe the workflow in plain English, the systems it touches, how often it runs, how much employee time it takes today, the important business rules, what the system may do automatically, and what is off-limits.

This prevents the conversation from starting with somebody else’s product demo.

It also makes proposals easier to compare. One provider may recommend an off-the-shelf product. Another may recommend a custom agent. You can judge both against the same workflow rather than comparing two completely different pitches.

Fixed-price projects work well when the workflow is contained and both sides can agree on what “done” means. Hourly or day rates can make sense for diagnosis or short technical work, but we would still want a cap or defined deliverable so discovery does not become an endless meter. A monthly retainer makes more sense once live workflows need monitoring, maintenance, and iteration. And a paid discovery phase followed by a separate build decision can be reasonable when the architecture genuinely is not obvious upfront.

Whatever the pricing model, ask what you own at the end: configuration, prompts, documentation, credentials, and data. Another provider should be able to maintain the workflow later if you choose to switch.

And ask about month four. Launch gets all the attention; the useful provider also has an answer for what happens when a login expires, the CRM changes, or your business updates the process.

Red flags of an unqualified AI consultant

A workshop can be useful. A 60-page “AI transformation roadmap” that hands every difficult step to somebody else is less useful for a 15-person business.

We would also be cautious if the provider has very little interest in your existing tools. Someone designing an operational workflow should want to know where the data lives, how often the job happens, and where people are manually handing work off. If the first conversation is a platform demo, you may be buying software rather than advice.

Guaranteed savings before a baseline exists are another warning sign. “Save 30 hours a week” is not credible until someone knows how long the process takes today and how much review the new workflow will require.

The same goes for proposals that never mention permissions, exceptions, or maintenance. Real workflows break, encounter weird cases, and need controls. If the entire story is the happy path, be careful.

A good consultant should also tell you when not to automate something. Maybe the job is too infrequent. Maybe the risk is too high. Maybe software you already pay for does it. Maybe the underlying process is such a mess that you should fix it before automating it.

And watch for scope that only moves in one direction. If every “could it also…” gets an immediate yes, the project is likely to become much bigger than the original problem.

References should look at least somewhat like you. A giant enterprise logo is not automatically useful evidence for a 12-person business with a completely different stack. Ask for a deployment with similar scale and complexity — then ask what went wrong.

Questions to ask before hiring an AI automation company

We would ask every candidate some version of these:

1. What would you not automate here, and why? This tests judgment immediately.
2. What does this look like in month four? You want to hear about monitoring, drift, exceptions, and ownership after launch.
3. What do we own at the end? Configuration, documentation, credentials, data, and the ability to switch providers all matter.
4. Which of our existing tools does this use, and what new software does it add? The architecture should be understandable.
5. How will we measure whether it worked? A provider who does not care about your baseline is difficult to hold accountable later.
6. What happens when the system cannot handle a case? Look for a clear fallback, escalation, and approval structure.
7. Show us a similar deployment and tell us what broke. The “what broke” part is often more informative than a perfect case study.

Unprompted sits in this category too: we help small businesses identify, build, and maintain workflows across the tools they already use. So hold us to the same standard. Send us the workflow and the first answer should be whether you need outside help at all.

The Bottom Line

• Do the simple AI work internally when you can.
• Bring in outside help when cross-system complexity, risk, or technical ownership justifies the fee.
• Scope the workflow before talking to providers so you can compare recommendations against the same job.
• Ask what happens after launch, what you own, and what the provider would not automate.
• Next: buy vs. build, how to maintain AI automations, or return to the full AI for small business guide.