The short answer: An AI consultant for a small business finds where your team's hours actually go, then works in three parts: advising on what to automate and what to leave alone, building automations inside tools you already own, and training your team to use them. A good engagement ends with a written roadmap and at least one workflow your team can run on its own.
An AI consultant helps a business figure out where AI tools can actually save time or money, then either builds that into the tools the business already uses or trains the team to do it themselves. The job splits into three parts: advise, build, train. Most engagements use some mix of all three, not just one.
That sounds simple, and the simplicity is the point. A lot of the confusion around this job comes from people expecting something more exotic: a data scientist, a machine learning engineer, someone who trains models. For a small business, that is almost never what you need. You need someone who can look at how your team spends its time, tell you honestly what is worth automating and what is not, and either build it or teach it.
The three kinds of work
Advise: find where the hours go, and what to skip
Before anything gets built, someone has to figure out where the time is actually going. That means sitting with the people doing the work, not just the owner, because the owner's mental model of the business and the front desk's actual week are often two different things. A consultant asks what takes longest, what gets redone, what customers complain about, and what nobody has time for.
Just as important is telling you what to leave alone. A consultant who recommends automating everything has not thought hard enough about your business. Some tasks are low-volume enough that automating them costs more than it saves. Some are sensitive enough - a first client email, anything touching medical or legal information, a price quote nobody double-checks - that they should stay a human's job for now. Part of the advisory work is drawing that line and explaining why.
This is also where a two-week assessment earns its keep: instead of guessing, someone spends real time mapping how the business runs and comes back with a written, prioritized list rather than a vague sense that "AI could probably help."
Build: automations inside tools you already own
The build part is where advice turns into something that runs. This usually means connecting or configuring tools you already pay for - your CRM, your calendar, your phone system, your inbox - so that a repetitive task happens automatically or gets drafted for a human to approve. It is rarely about buying new software. It is about getting more out of what is already sitting in your stack, half-used.
Examples of what this looks like in practice: an agent that answers a missed call with a text and books the caller onto the calendar, a system that sorts an inbox so urgent messages surface first, or a workflow that drafts a follow-up email after every job without anyone having to remember to write one. You can see live examples of this kind of work on the demos page, including a missed-call agent and inbox sorting, before you commit to anything.
Good build work is scoped tightly and reviewed by a human by default. A consultant should be able to tell you, specifically, what happens if the automation gets something wrong, and who sees it before a customer does.
Train: teach the team to use it on their real jobs
The most overlooked part of this work is training, and it is often the highest-leverage one. A tool that only the owner knows how to use is not really adopted. Training means sitting down with the actual people doing the actual jobs and working through their real tasks, not a generic slide deck about "the future of AI."
The difference between a training session that sticks and one that does not is usually whether people leave with something they used that same day. Sessions built around real work outperform lectures about AI in general, and the point is not that people nodded along, but that they are still using it a week later. If your team has been through training before and nothing changed, that is worth reading about separately: see why your team isn't using AI.
What a first engagement actually looks like
Step by step, a reasonable first engagement runs something like this:
- A discovery call. The consultant asks what is working, what is not, and what a good outcome looks like to you. This should be free and should not feel like a sales pitch.
- A short, bounded assessment. Something like two weeks of the consultant actually looking at your workflows, not months of open-ended "strategy." They talk to your team, look at real emails and processes, and come back with specifics.
- A written roadmap. Named tools, ranked by where the return actually is, in an order that makes sense given your team's capacity. Not a 40-slide deck of AI trends.
- A decision point. You choose what to do with the roadmap. A consultant worth hiring will let you take it and implement it yourself, or hand it to someone else, without hard feelings.
- Build and/or train, on the items you chose. This is where the actual automations get configured or the actual training happens, scoped to what the assessment found rather than a generic package.
- A walkthrough and handoff. You should end up able to run what was built without the consultant in the room, and knowing exactly who to call if something breaks.
If you want the fuller version of this process laid out end to end, including how the assessment itself is structured, see the workflow audit post.
What a good consultant hands you at the end
By the end of an engagement, you should have all of the following, not just a good feeling about the conversation:
- A written record of what was found and what was recommended, that you could hand to a different implementer if you wanted to.
- Ownership of every account, login and piece of data involved. Nothing should live only in the consultant's tenant.
- A plain answer to "who finds out if this breaks, and how." Automations fail quietly more often than loudly.
- At least one person on your team who understands how to use or adjust what was built or taught, so you are not permanently dependent on one outside person.
- A clear sense of what was intentionally left alone, and why.
If none of that shows up by the end, the engagement was a conversation, not a delivery.
Red flags
A few patterns are worth watching for before you sign anything:
- They cannot name anything they would not automate. Every real business has tasks that should stay human for now. A consultant with no answer here has not thought about your specific business.
- The accounts end up in their name, not yours. If the automations live in a tenant you cannot access without them, you do not actually own what you paid for.
- Everything is described in terms of hours saved with no source for the number. Be wary of specific-sounding statistics with no attached case or client. Ask where a figure came from.
- No training is offered at all. A build with nobody on your team able to run or adjust it is a dependency, not a system.
- They will not tell you the actual person doing the work. If your calls are with a salesperson and the delivery team is a mystery, ask to meet them before you commit.
When you do not need one
Plenty of businesses do not need to hire anyone yet. Skip it, or wait, if:
- Nobody on your team has spent real time with ChatGPT or a similar tool on their actual work. Do that first. It is free and it will change what you would even ask for.
- Your problem is not really about information or workflow. If the issue is pricing, staffing or a partnership disagreement, AI will make the wrong process happen faster, not fix it.
- You already know exactly what integration you need. That is a scoped development task, not a consulting engagement.
- Your records are a mess and you know it. AI on top of bad data produces confident, wrong answers. Sometimes the honest first project is unglamorous cleanup.
A free what to automate first tool or an ROI calculator can help you get a first read on your own before you talk to anyone.
How to tell if it is working
Set the measure before the work starts, not after. Reasonable signs it is working: people are still using what was built or taught a month later without being reminded, a specific recurring task now takes noticeably less time, and you can explain in one sentence what changed. Signs it is not: the tool was demoed once and never opened again, nobody but the consultant can explain how the automation works, or you are still waiting for a written summary of what was actually done.
For a broader comparison of consultants, agencies and other ways to get help, read consultant, agency, fractional, or nobody. And if you are trying to figure out exactly what to ask before you hire anyone, see questions to ask before hiring an AI consultant.
Frequently asked questions
What does an AI consultant actually do day to day?
A good one spends most of their time watching how your business actually runs, not talking about AI in the abstract. That means interviews with your team, reading the emails and spreadsheets that make up a normal week, and then either recommending changes, building something inside tools you already own, or teaching your people to use what you have. The mix depends on what your business needs.
Is an AI consultant the same as an IT consultant?
No. An IT consultant typically manages infrastructure, networks and security. An AI consultant focuses on how work gets done: which tasks can be handled by AI tools, which automations are worth building, and how to get a team actually using them. The two overlap at security and account setup, but the core job is different.
How long does an AI consulting engagement take?
It depends on the shape of the work. An assessment of where the hours go is usually a couple of weeks. A single build, such as a missed-call text-back or an inbox triage, is often a few weeks from first call to running. Training can be one afternoon. Ask for the timeline and the deliverable in writing before you commit.
Do I need an AI consultant, or can I just use ChatGPT myself?
If nobody on your team has spent real time with the tools yet, start there. It costs nothing and it changes what you would even ask a consultant for. Bring one in once you have a specific recurring problem, a stalled first attempt, or something that needs to run reliably enough that someone has to own it.
What should an AI consultant hand me at the end of an engagement?
A written roadmap or a working build, not just a slide deck. That includes what was found, what was implemented, who owns the accounts and data, and how you will know if it keeps working after the consultant leaves. If nothing is handed over that you could give to someone else, the engagement was not really finished.