Search for what AI costs and you get a wide range, mostly from people selling it. Some pages say a few hundred dollars. Some say tens of thousands. Few say why.
This post does not give you a Di-Hy price. Every engagement we do is scoped on a call, and you get a fixed price in writing before any work starts. What it does give you is a way to think about cost: the pieces, what the software vendors themselves publish, where bills surprise people, and the questions that turn a vague quote into one you can compare.
This is part of our guide to AI implementation for small business. Read that first if you want the full picture of what implementation includes.
The four buckets of cost
Almost every AI project has four kinds of cost. A quote that leaves one out is not cheaper, it is just hiding something.
1. Software subscriptions
The tools themselves. Usually monthly, usually per person. This is the part you can look up.
2. Setup and integration
Someone has to choose the tools, write the instructions the AI follows (those are called prompts), connect the tools to the software you already use (that is an integration), and test it on real work. This is labor. It is where most of the variation in price lives, because it depends on how many jobs you want covered and how messy your current tools are.
3. Training
Your team has to learn to use it on their real work. Surveys keep landing on this. The Federal Reserve Banks' Small Business Credit Survey found that among small firms planning to adopt AI, 37% named the time to implement or train staff as a challenge. Thryv's 2026 survey, reported by CPA Practice Advisor, found about 70% of owners said they need more training to use AI effectively. That one is a vendor survey, so weigh it accordingly. Either way, if a quote has no training line, ask why.
4. Upkeep
AI setups drift. Tools change, connections break, people forget. Someone has to check. That can be you, an employee, or a monthly plan. It is a real cost either way, and it is the one owners forget to budget. We wrote about it in what managed AI is and why it needs maintenance.
What software vendors publish
Here are three public price lists, read directly from the vendors' own pages in November 2026. Prices change, so check the page before you plan around any number.
- Claude (Anthropic): the Team plan is listed at $20 per seat per month billed annually, or $25 billed monthly, for teams of 2 to 150 people. Source: claude.com/pricing. The individual Pro plan is listed at $17 a month on an annual subscription or $20 billed monthly.
- Zapier (an automation tool that moves information between your apps): a free plan with 100 tasks a month. The Professional plan with 750 tasks a month is listed at $19.99 a month billed annually, or $29.99 billed monthly. Source: zapier.com/pricing.
- Google Workspace: Business Standard is listed at $14 per user per month, and the page says it includes the Gemini assistant in Gmail, Docs and Meet. Source: workspace.google.com/pricing.
Two things to notice. First, the software for a small team is often a few dozen to a few hundred dollars a month, not thousands. Second, that is the cheap part. The money in a project goes to the time it takes to make the tools do your specific job.
What moves the price
When someone scopes an AI project, these are the questions that change the number.
- How many jobs. One automation is a different project than five. We recommend starting with three, ranked by hours saved.
- How many tools it has to connect to. A calendar and an inbox are easy. An old industry-specific system with no connection options is not.
- How clean your information is. If your services, prices and policies live in people's heads, someone has to write them down first.
- How many people need training, and where. Ten people in a room costs differently than ten people on video.
- Whether you need ongoing upkeep. A one-time build and a monthly plan are two different products.
- How much risk. Anything that talks to your customers needs more testing than something that drafts an internal report.
Where bills surprise people
Usage-based pricing catches people. Pax8, which sells software to small businesses, reported in its Q3 2026 pulse report that 49% of AI purchasers got an unexpectedly high bill or usage charge. It also found 35% of AI-engaged firms discovered staff using AI tools the owner did not know about (often called shadow AI). It is a vendor survey of 1,150 leaders across four countries, so treat the numbers as a warning sign, not a benchmark.
The cheap fix is to ask three things before you sign up for anything:
- Is this priced per person, per use, or both?
- What happens when we go over?
- Can we set a cap or an alert?
A payback formula you can run yourself
Here is the math we would want to see before spending anything. It is simple on purpose.
Payback in months = total up-front cost divided by (hours saved per month x your hourly cost) minus the monthly cost of the tools and upkeep.
A worked example, with made-up numbers. These are illustrative, not a benchmark and not a Di-Hy price.
Say a service business spends 12 hours a month on appointment reminders and follow-up emails. The office manager's loaded cost is $30 an hour. If AI cuts that by 8 hours, that is $240 a month. If the tools and upkeep cost $60 a month, the net is $180 a month. An up-front project cost of $1,800 pays back in 10 months. An up-front cost of $6,000 pays back in about 33 months, and you should probably not do it for that one job.
The lesson is not the numbers. It is that cost only means something next to hours saved. Our free time-savings calculator does this math with your own inputs, and find your biggest time-sinks shows how to pick which jobs to count.
And be honest about the benefit. In the Fed survey, of firms using AI, 71% reported increased productivity but only 31% reported higher sales. Count time first. If you count revenue, count only what you can trace.
What to ask for in any quote
A quote you can compare has all of these in writing. If a vendor or consultant will not provide them, that tells you something.
- The jobs. Which specific tasks will be set up, named in plain words.
- What is included. Setup, connections, testing, training, written instructions.
- What is not included. Extra jobs, extra people, changes after launch.
- Software costs and who pays them. Your account or theirs? Who owns it if you leave?
- Ownership. Do you keep the prompts, the written playbook and the logins?
- Training. Who, how long, in person or video, and is there a follow-up.
- Upkeep. Month to month, yearly or none, and what it covers.
- A fixed price or an estimate. If it is hourly, ask for a cap.
- A way to measure. What number will you look at in 30 days to know it worked?
- An exit. What happens to your accounts and data if you stop.
We cover how to compare the types of providers in AI consultant vs AI agency vs doing it yourself.
Cheap and free options
Before you spend on outside help, know what is free. The SBA says SCORE mentors give area-specific advice at no cost, and Small Business Development Centers provide counseling and training. You can also start with one subscription and one job. Many owners never need more.
If you have several jobs, no time, and want training and upkeep included, that is when a firm earns its fee. Our free AI audit tells you which of those you are, and you keep the results. If you want to talk through scope for AI Implementation, book a strategy call.