What AI consultants actually deliver for small businesses, what it costs, and how to tell if you need one.
Article text
What Does an AI Consultant Actually Do? (And Is It Worth It for a 30-Person Company?)
A plumbing company owner told me last month that he'd hired an AI consultant the year before. Paid $12,000. Got a 47-page PDF with the word "roadmap" in the title. It recommended 11 tools, 4 integrations, and a "phased approach" that would take 18 months.
He never opened the PDF again after page 6. The tools sat unbought. The phases never started. Twelve grand, gone.
"So what do
you
actually do?" he asked me. "Because that guy did a lot of talking and nothing changed."
Fair question. And it's one I hear in some version from almost every business owner who calls us. The title "AI consultant" covers everything from someone who hands you a strategy deck and disappears, to someone who sits with your team, watches how work actually moves through your company, and builds the thing that fixes it.
The gap between those two experiences is the entire ballgame. Here's what the job actually looks like when it's done right.
Who This Post Is For
If you're running a company with 15 to 100 employees and you've been wondering whether an AI consultant is worth the money, or even what they'd do if you hired one, this is the breakdown. No jargon. No upsell. Just what the engagement looks like from your side of the table.
If you're already thinking about this, we do free 30-minute discovery calls where we'll tell you honestly whether you need outside help or whether you can handle it yourself.
The Job Has Three Parts (And Most Consultants Only Do One)
The worst version of AI consulting is pure strategy: someone analyzes your business, writes up recommendations, and leaves. That's the $12,000 PDF. It fails because the person writing the recommendations never has to live with the consequences.
The best version has three parts that can't be separated:
1. Diagnosis: Figuring Out What's Actually Broken
This is the part that most people skip, and it's the part that matters most.
Before we talk about AI tools, models, or automation, we need to understand how work moves through your company right now. Not how you think it moves. Not how the org chart says it moves. How it
actually
moves, including the workarounds, the sticky notes, the "just ask Janet" steps that nobody's documented.
We wrote a full breakdown of our diagnostic process if you want the details. The short version: we shadow your team, map every step, and find the spots where time, money, or accuracy is leaking.
This phase usually takes 1 to 2 weeks. It's unglamorous. It involves a lot of watching people do their jobs and asking "why do you do it that way?" But it's where 80% of the value gets created, because most companies don't have an AI problem. They have a process visibility problem.
A good AI consultant will sometimes tell you during this phase that you don't need AI at all. Maybe you need a better spreadsheet. Maybe you need to eliminate three redundant approval steps. If the consultant can't name a specific process they'd automate and explain why, they're not done diagnosing.
2. Build: Making the Thing That Fixes It
This is where strategy meets reality. A consultant who can diagnose but can't build is like a doctor who can read an X-ray but can't set the bone. Useful, but not enough.
Building means different things depending on what the diagnosis reveals:
Workflow automation
: Connecting the tools your team already uses so data moves between them without someone copying and pasting. A staffing agency we worked with had recruiters manually entering the same candidate info into three different systems. The build took 4 weeks and freed 18 hours a week .
AI agent implementation
: Building an AI system that handles a specific, repeatable task. An auto repair shop we worked with had service advisors spending 22 minutes per parts order cross-referencing catalogs. We built an agent that cut that to under 3 minutes.
Internal tools
: Dashboards, intake forms, automated reporting. Not fancy. Just functional. The kind of thing that replaces the spreadsheet everyone hates but nobody has time to fix.
Data pipeline work
: Getting your data from where it lives to where it's useful. Most small businesses have data trapped in 6 to 12 tools that don't talk to each other. This is the plumbing (no pun intended for the plumbing company owner) that makes everything else work.
The build phase typically runs 2 to 6 weeks for a focused project. We wrote about how we scope these projects so nothing balloons into a six-month odyssey.
3. Measurement: Proving It Worked
Here's a dirty secret about the consulting industry: most engagements never measure outcomes. The consultant finishes, sends an invoice, and moves on. If someone asks whether it worked, they point to the deliverable ("we built the thing!") rather than the result ("the thing saved you $94K").
Measurement means defining a specific number before you build, then checking it 30, 60, and 90 days after. Not "improved efficiency." A number. Hours per week recovered. Error rate before and after. Revenue that was leaking and isn't anymore.
Every project we do has a baseline metric established in week one and a measurement checkpoint at 60 days. If the numbers don't move, something's wrong, and it's our problem to fix.
What the First Two Weeks Look Like
If you've never hired an AI consultant, here's what should happen in the first 10 business days. If you're evaluating proposals and someone can't describe something like this, that's a red flag.
Days 1-3: Process observation.
The consultant (or team) watches your staff work. Not in a conference room looking at a slide deck. In the actual workspace, following the actual flow of work. They're looking for repetitive manual steps, data that gets re-entered, decisions that wait on one person, and handoffs where things fall through cracks.
Days 4-5: Process mapping.
They document what they observed and show it back to you. This is where you usually say "I had no idea it took that many steps" or "wait, they're still doing
that
manually?" The map should be visual and specific, not a generic flowchart from a template.
Days 6-8: Opportunity ranking.
Not every broken process is worth automating. The consultant should rank opportunities by three things: how much time or money the problem costs you, how hard the fix is to build, and how likely your team is to actually use it. The highest-ROI, lowest-friction opportunity goes first.
Days 9-10: Scope and proposal.
A clear document that says: here's what we're building, here's what it will cost, here's the timeline, and here's the specific metric we'll use to prove it worked. No ambiguity. No "phase 2 TBD."
If you get a proposal that skips straight to tool recommendations without this observation phase, the consultant is guessing.
What It Actually Costs
Let's talk money, because that's what you're really wondering.
According to The AI Consulting Network's 2026 data , here's the range:
Readiness assessment only
: $2,000 to $8,000. You get a diagnosis and a priority list. Useful if you have internal people who can build. This takes 2 to 4 weeks.
Strategy plus pilot
: $15,000 to $50,000. You get the diagnosis, a focused build on your highest-priority process, and measurement at 60 days. This is where most 20-to-80-person companies land. Timeline: 6 to 12 weeks.
Full implementation
: $50,000 to $150,000+. Multiple processes, multiple integrations, staff training, ongoing optimization. This is for companies that have validated AI works for them and want to go deeper. Timeline: 3 to 6 months.
For context, most of our engagements with companies in the 15-to-60-employee range fall in the $15,000 to $40,000 bracket. The ROI target is 3 to 5x the consulting fee within the first year , and for the projects we've published case studies about, we've consistently hit that.
A fractional Chief AI Officer model is also gaining traction: $2,000 to $8,000 per month for 8 to 20 hours, giving you ongoing strategic guidance without the $200K salary. For a 30-person company, this often makes more sense than a one-time project if you have multiple processes worth automating.
When You Don't Need an AI Consultant
I'd be doing you a disservice if I didn't say this: not every business needs to hire outside help.
You probably don't need a consultant if:
You have one specific, simple workflow you want to automate and it only involves one tool. Most AI-powered CRM, scheduling, and customer service tools have good enough setup wizards that your team can handle it.
Your real problem is that you haven't picked a tool yet. That's a research problem, not a consulting problem. Spend a few hours comparing options, pick one, try it for 30 days.
You have a technical person on your team who's already experimenting with AI tools. Give them 10 hours a week and a small budget. Internal knowledge of your business is worth more than external AI expertise if the technical skills are there.
You probably do need a consultant if:
The workflow you want to fix spans multiple tools, departments, or data sources. Integration work is where DIY falls apart.
Customer data is involved and you need to get it right (HIPAA, financial data, PII). The compliance and security layer adds real complexity.
You've tried the DIY approach and it stalled. Usually because the initial tool worked fine but connecting it to the rest of your workflow required technical skills nobody on your team has.
You don't know where to start. If you're staring at 15 processes and can't tell which one to automate first, the diagnostic phase alone is worth the investment.
Red Flags When Evaluating AI Consultants
Because the plumbing company owner's $12,000 PDF experience is more common than it should be, here's what to watch for:
They lead with tools, not problems.
If the first meeting is about which AI model to use or which platform is "best," they're working backwards. The tool should follow the diagnosis, not the other way around.
They can't show you specific outcomes from past work.
"We helped companies improve efficiency" means nothing. "We helped a 35-person insurance agency reduce hold times and protect $170K in annual revenue" means something. Ask for specifics.
The proposal is vague on measurement.
If the deliverable is "implementation of AI solution" with no defined success metric, you'll have no way to know if it worked.
They don't want to observe your team.
If they can diagnose your problems from a conference room without watching how work actually flows, they're applying a template, not solving your problem.
The timeline is longer than 12 weeks for a first project.
First engagements should be focused, not ambitious. The right first project is a single process with clear before-and-after metrics. If someone's pitching a 6-month transformation for your first engagement, they're over-scoping.
The Question That Actually Matters
The question isn't "what does an AI consultant do?" The question is "what changes in my business after they leave?"
If the answer is "you got a PDF and a list of tools," that's not consulting. That's a report.
If the answer is "your dispatching process went from 14 steps to 4, your team got 20 hours a week back, and you can see the impact in your P&L," that's the job done right.
We've done this for HVAC companies , law firms , dental practices , and a dozen other industries . Every engagement starts with the same question: where is time or money leaking, and what's the simplest thing we can build to stop it?
If you're weighing whether it's worth the call, book a free 30-minute discovery session . We'll look at one process together and tell you whether outside help makes sense, or whether you can handle it on your own. No pitch. Just an honest answer.