Ask what AI automation costs and you will get a range so wide it is useless. I have seen the same job quoted at $1,500 and at $30,000, and both firms were being honest, because they were selling genuinely different things.
So rather than give you one number, here is the shape of the market, what the pieces actually are, and the questions that turn a vague estimate into a real one.
There are four separate costs, and most quotes only mention one
This is the single most useful thing to understand before you talk to anybody.
The build. One-off. Designing the workflow, writing it, connecting it to your systems, testing it, handing it over. This is the number everyone quotes.
The running cost. Ongoing, and real. Every message a language model writes costs a fraction of a cent, every text message costs a fraction of a cent, every phone minute costs money. Small numbers, but they never stop, and they scale with how well your business is doing.
The software it sits on. Hosting, a database if it needs one, sometimes a third-party platform licence. Often a few dollars a month. Occasionally not.
Maintenance. Things break. Providers change their APIs, a system you connected to gets updated, a workflow that worked fine in January starts behaving oddly in June. Somebody has to notice and fix it.
A quote that mentions only the first of those four is not a complete quote. Ask about the other three before you sign anything, and be suspicious of anyone who says the running cost is nothing. It is never nothing.
What the market actually charges
Broadly, and from what I see in this market rather than from a survey, here is the shape:
Audits and discovery: $0 to about $1,000. Some firms charge for the first conversation, some do not. Paid discovery is a legitimate model, particularly for larger companies where the discovery genuinely is the work. For a small business it means paying a meaningful sum to find out whether a build is worth doing. Mine is free, which I mention because it is the part worth comparing.
A single defined workflow: roughly $2,000 to $8,000. One thing, like missed-call text back or automated follow-up, built and connected to one system. The spread is mostly about how much integration is involved and how much of your process has rules that only exist in somebody's head.
Multi-step builds: $3,500 to $20,000 and up. Several systems talking to each other, decision logic, multiple user types. This is where the range gets genuinely wide, and where scoping properly matters most.
Custom assistants and agents: $5,000 upward. Something that answers, decides or acts, with the permissions, evaluation and escalation work that responsible versions of these require.
Voice and AI receptionists: often a setup fee plus a monthly. The monthly covers keeping it tuned and answering, and telephony is usually billed separately on top.
Retainers: anywhere from a few hundred to a few thousand a month. This varies more than anything else on the list, and it is worth being clear what you are buying. Monitoring and small fixes is one thing. Defined development capacity is another. An open-ended "strategic partnership" is a third, and for a business under about fifty people I would want a very specific reason to buy it.
My own prices sit inside those ranges and are published on the site rather than quoted on a call, which is unusual in this market and is deliberate.
What actually moves the number
If you want to predict where in the range you will land, four things do most of the work:
- How many systems it has to touch. One is straightforward. Four is a different project, because every connection is a place things can break.
- Whether your process is written down. If the rules live in one person's head and change depending on the day, someone has to extract and formalise them first. That is real work and it belongs in the quote.
- Whether anything can act without approval. A workflow that drafts for a human to send is much cheaper than one allowed to act alone, because acting alone requires permissions, evaluation and escalation handling.
- How wrong it is allowed to be. Testing against the awkward cases costs time. Skipping it costs more later.
The running cost, honestly
This is the part I wish more people asked about, because it is where surprises live.
For most small business workloads the AI usage itself is genuinely small, often a few dollars a month, occasionally a few tens of dollars if it is working hard all day. Text messaging is cents per message, which adds up if you send thousands. Phone minutes cost more than texts.
Two things are worth insisting on.
Ask to see the provider's bill. If someone marks up your usage without telling you, you will not find out for a year. I show the invoice and add nothing to it, and you can put the accounts in your own name if you would rather I never touched them.
Ask what happens when you grow. Some pricing scales with your success in an unpleasant way. A chat tool that charges by conversation bills you more precisely when your marketing starts working, which is backwards.
Build or buy
Sometimes the honest answer is that you should just pay for a product.
Off-the-shelf wins when your process is normal, the product does most of what you need, and the subscription is small next to a build. If a $60 a month tool does the job, spending $4,000 to own a slightly better version of it is a bad trade unless you are going to use it for years.
Custom wins when the workarounds have become the job, when you are paying for four products to do one thing, or when the thing you need does not exist. It also wins when you want to own the result rather than rent it.
That comparison is most of what a good first conversation should be about, and any consultant unwilling to tell you to buy something instead is not really advising you. There is a longer version of this argument, including the running costs, on the AI tools page.
Questions that turn an estimate into a price
Take these to whoever you are talking to, including me:
- Is this a fixed price or an estimate? If it moves, what moves it?
- What are the running costs, and will you show me the provider's invoice?
- What happens if it breaks in six months? Who fixes it and what does that cost?
- Do I own the code and the data? Can I export everything?
- What is explicitly not included?
- What would you tell me to buy instead of building this?
The last one is the most revealing question on the list. Anyone with a good answer to it is worth listening to.
The short version
For a small service business, a first automation that genuinely earns its keep usually lands somewhere between two and four thousand dollars to build, plus a small monthly running cost, and it should pay itself back inside a year or it was the wrong first build.
If you want a specific number for your situation rather than a range, the free audit exists for exactly that, and it ends with "not yet" often enough that I mention it on every page.