Last week Meta added another AI agent to your options. On September 29 it launched Muse for Small Business, which connects to tools like Asana, Zoom, Intuit, Box, Canva and Slack, and is free with usage limits (a subscription covers more). Meta's own pitch to owners: "They told us they're short on hours, not ideas."
The tools keep arriving, and most of them are cheap to try. So why aren't more small businesses using them?
The Census Bureau has been asking. Between December 2025 and May 2026, AI use rose among firms with at least 20 employees but didn't change significantly among firms with fewer than 20. By May, 37% of firms with 250 or more employees used AI, against less than 20% of firms with 4 or fewer. A Federal Reserve note published in April found that larger firms had the highest adoption rates in the business surveys it examined.
Big companies have people whose job is to test a tool, set it up and write the rules. A small business has an owner and a full inbox. That gap is what an AI automation consultant is supposed to close, and it's why choosing one carefully matters.
This guide is designed to:
- Explain what an AI automation consultant does, in plain terms.
- Help you decide whether you need one yet.
- Give you the questions to ask before you sign anything.
- Show you how to run a first project you can measure.
After reading it, you will be able to:
- List the 2 or 3 jobs in your week most worth automating.
- Write a 1-page brief a consultant can quote against.
- Tell a careful consultant from one who just resells software.
- Set up a 30-day pilot with clear measures of success.
The basics: what an AI automation consultant does
An AI automation consultant looks at how work moves through your business and finds the steps a tool can do for you. Then they set those tools up, connect them to the software you already use and show your team how to work with them.
In practice that usually means some mix of:
- An audit. What you pay for, what overlaps, where data lives and who can reach it.
- Connections. Linking an AI assistant such as ChatGPT, Claude, Gemini or Muse to your email, files, calendar or accounting software, with the right permissions.
- Automations and agents. Workflows that move information between apps without someone retyping it, and agents that draft, sort or answer on your behalf.
- Training and rules. Showing staff what good use looks like, and writing down what must never go into an AI tool.
Most of that list is about connections and permissions. Muse's selling point is that it plugs into your existing tools. Every one of those connections is also a door into your customer data, so somebody has to decide which doors open and for whom.
OK, so now you know what the job covers. But do you need to hire someone for it?
Step 1: Lay the groundwork
Before you talk to anyone, find out where your hours actually go. It's easy to guess wrong, because the jobs that eat the most time are the small repeated ones nobody tracks.
For 1 week, keep a simple tally of tasks that are repeated, rule-based and done on a screen. Copying order details into an invoice. Answering the same 5 customer questions. Following up on who still owes you a form, a signature or a payment.
Pause and reflect: Which 2 tasks took the most time this week? Who did them, and what software were they working in? Write them down before Step 2.
Ask yourself:
- Does this task follow the same steps most of the time?
- Does the information already exist somewhere else (an email, a form, a spreadsheet)?
- Would a mistake here be annoying, or would it cost you a customer?
- Who would notice first if the task stopped getting done?
The tasks with "yes, yes, annoying" answers are your best first candidates. Leave anything where a mistake reaches a customer for later.
Step 2: Decide whether to try it yourself first
Plenty of first automations are a free trial and an afternoon away. If you're comfortable with your software's settings, try 1 small workflow yourself before paying anyone. You'll learn what you want, and you'll ask better questions later.
| Try it yourself when | Bring in help when |
|---|---|
| The task lives inside 1 app | The task crosses 3 or more apps |
| A mistake is easy to spot and undo | A mistake reaches customers or your books |
| Only you need to use it | The whole team has to change how they work |
| No customer or employee data is involved | Personal, health or payment data is involved |
| The vendor has a ready-made template | You'd need to connect accounts and set permissions |
If most of your answers sit in the right-hand column, you're past the point where trial and error is cheap.
Step 3: Write a 1-page brief
Large organizations don't call a consultant and ask what they should do. They write a statement of work first: the problem, the scope, how success gets measured and who signs off. You don't need the 20-page version. 1 page does the same job and makes quotes comparable.
Your brief should cover:
- The problem in your words. "We retype every web order into QuickBooks, about 6 hours a week."
- The tools involved. Every app the work touches today.
- What done looks like. Orders reach QuickBooks without retyping, and someone reviews a daily list of exceptions.
- The data involved. Customer names, addresses, payment details, anything sensitive.
- Who owns it afterwards. The person on your team who will look after it once the consultant leaves.
A consultant who reads this and asks good follow-up questions is a better sign than one who replies with a price in 10 minutes.
Step 4: Ask the right questions before you hire
Ask every candidate the same questions and compare the answers side by side. Careful consultants and software resellers answer them very differently.
Ask yourself, and then ask them:
- Whose accounts will this run under? Automations should live in accounts your business owns, not the consultant's personal ones.
- What happens to our data? Which tools will see customer information, and where is it stored?
- Are you paid by any of the vendors you recommend? Some consultants earn referral fees. You want to know before you take their advice.
- Is it a fixed quote or open-ended hours? A defined pilot can usually be quoted at a fixed price.
- What do we get when you're done? Written instructions, a list of every connection and login, and a way to switch things off.
- How will we know it worked? The answer should be a number you can check, not a feeling.
Pause and reflect: If your consultant disappeared tomorrow, could someone on your team find every automation they built and turn it off? If not, add that to your brief.
Step 5: Start with a pilot, not a platform
Big companies rarely roll a new system out to everyone at once. They pilot it with 1 team, agree on success criteria in advance and decide afterwards whether to expand. Copy that habit.
Pick 1 workflow from Step 1. Run it for 30 days. Measure it before and after with a short scorecard:
| Measure | Before | After 30 days |
|---|---|---|
| Hours a week spent on the task | ||
| Errors or rework found | ||
| Time from request to done | ||
| Staff who use it without being reminded |
Fill in the "Before" column first, during Step 1. Without it, you can't tell whether the pilot helped or just felt new.
At the end of the 30 days you have 3 choices: keep it, fix it or switch it off. Any of them is a fine result. A pilot you switch off has cost you a month.
Step 6: Write the rules before you scale
Once 1 automation works, the temptation is to connect everything. Pause here. Larger organizations manage AI with written frameworks. The best known in the U.S. is the NIST AI Risk Management Framework, a voluntary framework first released in January 2023. Its 4 core functions are Govern, Map, Measure and Manage.
Here is the small-business version (adapted from the NIST AI Risk Management Framework):
- Govern: Who decides which AI tools are allowed, and where is that written down?
- Map: Which tools touch customer, employee or financial data?
- Measure: How do you check the output is right (a weekly spot check, an exceptions list)?
- Manage: What do you do when it gets something wrong, and who does it?
A good consultant will help you answer these on a single page. A great one will insist on it.
Pause and reflect: What is 1 kind of information your team should never paste into an AI tool? Customer card numbers and employee health details are a good place to start.
Questions owners ask
How much does an AI automation consultant cost?
Pricing models vary: hourly rates, fixed project fees and monthly retainers are all common. For a first project, ask for a fixed quote against your 1-page brief. That keeps the scope clear and makes quotes easy to compare.
Do we have to switch software?
Usually not. Most of the work is connecting the tools you already pay for. Be wary of anyone whose first recommendation is a new platform.
Will automation replace my staff?
That depends on the task. First automations tend to take over retyping and checking, and the hours that frees up go back to the person who did it. Your pilot scorecard will show how many hours that is.
Concluding thoughts: start with 1 workflow
The tools will keep coming. This week it was Muse, and next week there will be another launch with a long list of connections. What closes the gap between your business and the 250-person company down the road is a method. Know where your hours go, try 1 thing, measure it and write the rules.
Go back to Step 1 and start your tally this week. If you'd rather have someone run the method with you, our audit starts with exactly that list, and our automation work picks up from the pilot.