Founders open LinkedIn, type AI automation engineer, and hire the first person who can demo an agent. Three months later the engineer is still waiting on a spec, and the missed follow-ups have not moved. The title was not the mistake. The sequence was. An AI automation consultant is paid to pick the job and get the first version live in accounts you own. An AI automation engineer is paid to build and keep what you have already decided. This is how to tell which invoice matches the mess you actually have.
What an AI automation consultant does all week
A useful week is not a workshop. It is spent inside work that already happens.
Monday is the queue. They sit with the person who does the job and watch one real cycle. A lead that arrived overnight. An invoice 11 days late. A booking confirmed in email that never reached the calendar. They write the trigger, the systems touched, the minutes, and what it costs when it slips.
Tuesday and Wednesday are the cut. Of the five jobs on the page, one or two are worth a build. The rest stay manual because they happen twice a month, or because the judgment is the product. A consultant who wants to automate all five is selling hours, not a result.
Thursday is the build, in tools you already pay for. n8n, Make, Zapier, the CRM, the inbox. Not a new platform "so there is room to grow". Friday is a live run on your records, with you watching, plus a note that says the workflow fails closed instead of failing silent.
If the proposal you received does not sound like that week, you are not hiring an AI automation consultant. You are hiring a strategist. A strategy with no workflow in an account you can log into is a PDF. What that engagement should cost, and which pages to refuse, is in business automation consulting, priced in plain numbers.
What you actually hire an AI automation engineer to build
An AI automation engineer is an in-house builder. People search the title more than they search for a consultant, and the job is narrower than the title sounds.
Hire one when the decision is already made. You know the trigger, the systems, and what done looks like. You need someone on the payroll who will still be there when the API changes, the CRM field gets renamed, and a salesperson invents a new status on a Thursday.
Good engineer work is boring to look at, which is the point. A workflow folder the team can open. A checklist for the three ways the job breaks. A log you can read without them on a call. A habit of saying no to the fifth integration until the first two have run for a month without a rescue.
Bad engineer work is a private n8n instance on their laptop, no notes, and a Slack thread called "do not touch". When they leave, the automation leaves with them. You did not hire a system. You rented a person.
Do not hire an engineer to discover the problem. That is the consultant's first week. An engineer asked to do both will freeze and wait for a spec, or automate the loudest request in the office, which is rarely the expensive one. If you are still choosing the job, start with where AI agents for business earn their keep, then come back to the hire.
Match the bill to the backlog. An in-house AI automation engineer is a salary, benefits, and tools, every month, whether the queue is full or not. A consultant is a project. One finished workflow is often $3,000 to $8,000, which is the range we see quoted for a single done-for-you build. The engineer is the cheaper shape only when there is enough work to fill the role past the first quarter. Three workflows do not do that. Idle builders invent platforms. Platforms are how a five-person company ends up maintaining software nobody asked for.
When an AI automation company is the better call
Sometimes neither a solo consultant nor a single engineer is the right shape. An AI automation company earns the fee when the path crosses desks and you do not want the knowledge in one person's head.
A lead comes in. Sales qualifies it. Ops sends the contract. Finance chases the invoice. That is four people and usually four tools. A company should bring someone who has shipped that path, a builder, and a person who writes the words a customer will read, then does not vanish after go-live. You are paying for the sequence and for a second pair of hands when the first is on leave.
Hire a solo AI automation consultant when the stack is small, the owner can decide in the room, and you want one named person accountable. A company adds process. Process helps around 40 people. It can waste a month at 8, because every decision waits on a standup that did not need to exist.
The practical test is the second workflow. If you already know there are four, and they share systems, a company or a retainer keeps the context. If you know there is one painful job, pay someone to finish that job and stop. How firms differ from a freelancer with a logo is in what AI automation agencies actually staff.
Either way, the non-negotiables are the same. The login is yours. A named owner on your side knows what "done" means. Failures tell a person before a customer does. What you should not buy first: a six-month transformation, a new CRM, or a chatbot that never touches the queue you complained about on the intro call.
What a business process automation consultant should hand you
A business process automation consultant is the same job with a colder name. Less "AI" on the slide, more "this form, this approval, this handoff". For a lot of small firms that is the honest scope. The model is one step inside the workflow, not the project.
Ask for four artifacts, and do not start until they are named in the proposal.
- A one-page map. Trigger, steps, systems, and the human who still approves anything a customer will read.
- The workflow in an account you own. n8n, Make, or Zapier. If it lives only in theirs, you are renting.
- A failure note. Timeout, blank field, CRM down. What the system does, and who hears about it.
- A handover you can replay. Thirty minutes, recorded, plus the page that says who to ping on a Monday.
If the proposal is a RACI chart and a tool shortlist, you are talking to someone who does not build. That can be a fair two-week audit, if the price is an audit price. It is not a $20,000 build. The line between a service that ships and a workshop that does not is in business process automation as a service.
Questions for a business automation consultant on the first call
Bring five repeating tasks. Times per week, minutes each, and what a miss costs. Then ask these, and write the answers down. A business automation consultant who cannot answer them on the call is not ready to price the work, and you are not ready to fund a discovery phase to find out.
- Which of these five would you refuse to automate, and why? A good one crosses at least one off. Yes to all five means they are pricing hours.
- What will I be able to open on the Friday of week one? If the answer is a deck, you are in a strategy sale.
- Whose login will the workflow sit in? The only acceptable answer is yours.
- What does the customer see on day one? Nothing sent automatically. Drafts, then a person, then the boring send once the tone is trusted.
- What is the number? Hours back per week, or cash. "Improved efficiency" is not a number.
- Who fixes it when a tool renames a field in October? If the answer is "not included", price a small retainer now, or accept that you are the on-call engineer.
Two more filters. Do they ask to see the real queue, or do they pitch a package? Packages fit a newsletter tool. They fit badly on the spreadsheet your ops lead built because the official report was wrong. And do they want to rip out tools the team already uses? A migration is sometimes right. It should never be step one. The jobs that usually pay back first for a small team are in what small businesses automate before they buy another platform.
The first 30 days, whoever you hire
The title on the invoice should not change the month. If it does, someone is stalling.
Days 1 to 5. Watch the work. No new software unless a tool you already have cannot do the job. Write the five tasks down if that page is still blank.
Days 6 to 10. Pick one job. Define done in a sentence a new hire could check. Example: every new lead from the site is in the CRM within 10 minutes, with a draft follow-up waiting for the owner to send.
Days 11 to 20. Build on real records, not a demo account. You approve any sentence a customer might read. Nothing customer-facing sends itself yet.
Days 21 to 30. Run it. Count the runs. Count the failures. Pick a second job only after the first has survived a week without a rescue. Starting the second one early is how a zap farm begins. Forty automations, nobody sure which still fire, a new hire afraid to edit them.
Day 30 also includes the folder, the logins in your password manager, and a 30-minute recording. Consultant or engineer, same list. If either person treats documentation as a later phase, it will not happen. Later is where handovers go to die.
A simple way to choose
Choose an AI automation consultant when you can name the pain and cannot yet name the workflow. You need the decision and the first build. After that you might not need anyone for a quarter.
Choose an AI automation engineer when the workflows already exist, they break, and someone has to live with them every week. You also need a backlog that still exists in month four. If you do not, you will invent work to justify the salary.
Choose an AI automation company when the path crosses several teams and the context should outlive one person. Pay for sequence and handover, not for a cleverer zap.
Choose none of them this month if you cannot fill the five-task page. Write that page yourself. It is free, and it is the brief every serious hire will ask for on the first call. Showing up without it is how you buy a workshop.
- An AI automation consultant picks the job that should stop being manual, then ships the first version in accounts you own.
- An AI automation engineer builds and maintains what you already specified. Searched more often. The wrong first hire if the decision is still open.
- An AI automation company fits when the workflow crosses teams and should not live in one person's head.
- A business process automation consultant owes you a map, a login you own, a failure note, and a recorded handover. A deck is not a deliverable.
- One finished workflow is often $3,000 to $8,000. An in-house engineer is a salary. Match the bill to the size of the backlog.
Want the five-task page read before you hire anyone? Book a free 20-minute walkthrough. We will say which line is a consultant job, which is an engineer job, and which you should leave alone.