Business of One
[ ← Guides ]

Consultant, agency, build team or hire? How to choose AI help, with 20 questions for any vendor

Business of One Editorial

Hire a consultant when you need a decision, an agency when you need one defined project delivered, a build team when you need systems built and kept running, and an employee when the work is permanent and steady enough to fill a full-time job.

Two mistakes are easy to make: buying a project when you needed a decision, and buying advice when you needed someone to build. The table below is meant to stop both.

The four options side by side

Consultant Agency Build team Employee
What you buy Judgment on a decision A defined deliverable Ongoing building and running A person’s full working time
How it’s usually priced By the hour or by the project By the project, sometimes plus a monthly fee By the month Salary plus benefits
Best when You don’t yet know what to build, or whether to build The scope is clear and won’t change much The work keeps changing and the systems must stay up The work is core, steady and permanent
Main risk Advice with no one to carry it out A finished project nobody maintains Paying monthly for work you could own A slow hire, or one bet on one person’s skills
What you own at the end A plan, and whatever you learned Whatever the contract hands over Whatever the contract hands over, plus what they taught you The person’s work, while they stay
Your time needed High during sessions Low to medium Medium, ongoing High at first: hiring and managing

Two notes on the table. BLS says management analysts, “often called management consultants,” are paid directly by their clients when self-employed, typically by the hour or the project. And the “what you own” row depends on the contract, not the label. An agency can hand over everything. A build team can keep everything. Ask.

Start from your situation

If this is true for you Start with Why
You aren’t sure what AI should do in your business A consultant A decision is cheaper than a project built on a guess
You know exactly what you want, and it’s one thing An agency or a fixed-price project Clear scope is what project pricing needs
You want several systems, and they must keep working as tools change A build team Someone has to watch, fix and adapt them every week
The work is core to how you compete, and you’ll need it for years An employee, maybe after a build team proves it Skills in the building are hard to take away
You want to learn to run the systems yourself A consultant or build team that teaches You need the skills, not just the output
You have no budget for outside help yet Your own time, then a small advice block Learn where the value is before you pay to build

These start points can combine. A sequence that holds up: get a decision first, build the first system with outside help, then hire for the parts that turn out to be permanent.

Consultant or agency: the real difference

The label matters less than the question each one answers.

  • A consultant answers “what should we do?” Good ones leave you with a decision, a written plan, and enough understanding to judge the next vendor.
  • An agency answers “can you build this?” Good ones deliver what was scoped, on time, and hand it over cleanly.

Trouble starts when the questions are mixed. An agency asked “what should we do?” tends to recommend something it builds. A consultant asked “can you build this?” may not have builders. Neither is dishonest. Each is answering from what it sells.

What “done for you” leaves out

A done-for-you AI agency builds the system so you don’t have to. That is useful. But three things can’t be done for you:

  1. Deciding what the work is. Only you know which jobs matter, what good output looks like, and what must never happen.
  2. Checking the output. Someone inside the business has to review what the AI produces, at least until you trust it.
  3. Owning the change. When a model changes or a vendor raises prices, someone has to decide what to do.

If a vendor says you won’t need to be involved, count that as a warning. The journal post Rent the harness, keep the judgment makes the same case for AI infrastructure: you can rent the machinery, but the job description, the knowledge and the stop rules stay with you.

When to hire a team, or an employee

Hiring puts skills inside the company. It is also slow and expensive. BLS puts the May 2025 median wage for U.S. software developers at $135,980 a year, before benefits. One developer is one skill set, working normal hours.

A build team on a monthly fee gives you several skill sets and coverage, but the skills live outside your company unless teaching is written into the deal.

For the numbers on each, including BLS wages for comparable roles and a worksheet, see What AI help costs.

Many small firms using AI haven’t paid for help

The SBA’s Office of Advocacy, in a September 24, 2025 report, looked at Census survey data on businesses using AI (collected December 4, 2023 to February 25, 2024). About half of small firms using AI reported no investments in it, whether in employee training, physical capital or intangible processes. Hiring vendors or consultants to integrate AI was one of the areas where small firms lagged large ones most: a 5.2 percent difference, in the report’s words.

That data was collected more than two years ago. It still shows that paying for help is not the only way in. Paying makes most sense once you know what you want the help to produce. For current adoption figures, see Small business AI adoption data.

20 questions to ask any vendor, including us

Ask these of any consultant, agency, build team or candidate. Ask them of Business of One too. Get the answers in writing.

Outcome and scope

  1. What will be different in my business when you finish, and how will we measure it?
  2. What is out of scope?
  3. How many hours a week will you need from me or my team?

People

  1. Who, by name, will do the work, and who is on call when something breaks?
  2. What have those people built that I can see working?
  3. Will any of the work be subcontracted, and to whom?

Ownership and exit

  1. Who owns the code, prompts, workflows and data when we stop working together?
  2. Which accounts and subscriptions will be in my name, and which in yours?
  3. What do I get on the last day: documentation, credentials, a handover session?
  4. How much notice does either side give to end the work, and what does ending cost?

Data and control

  1. Where will my data go, which AI providers will see it, and do they keep it?
  2. What can the systems do without a person approving it first?
  3. How will we see what the AI did, and can I read those logs myself?
  4. What risk framework do you work from, and where does it show up in the work? NIST’s AI Risk Management Framework, released January 26, 2023, is one public example.

Models and change

  1. When a better or cheaper model ships, who decides whether to switch, and who pays for the work?
  2. Do you have partnerships, referral fees or commissions with any AI vendor you’ll recommend?

Price and terms

  1. What is the full price, including model usage, software seats and overages?
  2. What happens to the price and the timeline if the work takes longer than planned?

Proof and learning

  1. Which of your claims about results can you back with evidence I can check?
  2. What will I be able to do myself at the end that I can’t do today?

Question 19 matters more than it looks. On September 25, 2024, the Federal Trade Commission announced Operation AI Comply, five law enforcement actions involving AI hype and deceptive AI claims. FTC Chair Lina M. Khan said at the time that “there is no AI exemption from the laws on the books.” A vendor who can’t support a claim shouldn’t be making it.

Red flags

  • Guaranteed results. Nobody can promise what a system will do in your business before it’s built.
  • No names. If you can’t learn who will do the work, you can’t judge it.
  • No exit. If leaving means losing everything, you’re renting, not buying.
  • “You won’t need to be involved.” See above.
  • One model forever. Models change often. A vendor with no plan for that is building something you’ll pay to rebuild.

This guide is education, not legal advice. Have a lawyer review any contract, especially the ownership, data and termination terms.

If you want to ask us the 20 questions, Business of One offers two ways in: Consult at $400 an hour in 5-hour blocks, and Build at $15,000 a month with build-with-you coaches, agents that work only on your systems, and a shared 24/7 on-call rotation with response times in writing.

  1. Mistral published the weights. Read the model-as-a-service line. · October 30, 2026
  2. GPT-4o and GPT-4.1 snapshots leave OpenAI's API on January 14. Find every process that still calls one · October 14, 2026
  3. Rent the harness, keep the judgment · September 24, 2026

Sources

  1. Management Analysts : Occupational Outlook Handbook: : U.S. Bureau of Labor Statistics · U.S. Bureau of Labor Statistics · 2026-08-27
  2. Software Developers, Quality Assurance Analysts, and Testers : Occupational Outlook Handbook: : U.S. Bureau of Labor Statistics · U.S. Bureau of Labor Statistics · 2026-08-27
  3. AI in Business: Small Firms Closing In · U.S. Small Business Administration, Office of Advocacy · 2025-09-24
  4. FTC Announces Crackdown on Deceptive AI Claims and Schemes · Federal Trade Commission · 2024-09-25
  5. AI Risk Management Framework | NIST · National Institute of Standards and Technology · 2023-01-26

Researched and drafted with AI assistance, checked against the sources above.

See how this could work in your day.

See a worked example →

Ask the Business of One advisor