Every article about AI costs for small businesses either quotes enterprise figures that are irrelevant to you, or dodges the question entirely with “it depends”. Both are useless when you are trying to work out whether to have the conversation at all.
So here are actual ranges, what drives them, and — more importantly — the costs that do not appear in anyone’s proposal.
The three cost tiers
Almost everything a small business might do falls into one of three bands.
Configured off-the-shelf tools: $500–$3,000 setup, $50–$400 per month. You are buying existing software and having someone connect it properly to your systems. Document capture, automated client reminders, a scheduling workflow, invoice routing. This is where the large majority of small business automation lives, and it should be where you start.
Connected multi-system workflows: $3,000–$12,000 setup, $150–$800 per month. Several tools stitched together with logic between them — a process that pulls from your CRM, checks something in your accounting software, makes a decision, and writes back to both. More moving parts, more testing, more to maintain.
Custom-built agents and integrations: $10,000+ and up. Something built specifically for you because nothing off the shelf fits. Genuinely warranted sometimes. Warranted far less often than the people selling it suggest.
If a vendor quotes you the third tier for a problem that lives in the first, that is worth noticing.
What actually drives the number
The price is rarely about the AI. It is about four things.
How many systems have to talk to each other. One system is cheap. Two is manageable. Four, one of which has a bad API and one of which is a desktop application from 2014, is where budgets go to die.
How well documented your current process is. If nobody can tell us exactly how the task is done today, including the exceptions, the first chunk of the project is discovering that. It is worth paying for, but it is a real cost.
How badly it can go wrong. A workflow that drafts internal emails needs light testing. A workflow that touches your general ledger needs review gates, parallel running and an audit trail. The safety work often costs more than the automation.
Volume. Not because software charges more, but because low volume rarely justifies any of it. If a task happens twice a month, the honest answer is usually to leave it alone.
The costs nobody quotes you
This is the part that catches people out.
Ongoing subscriptions. Automation platforms, AI API usage, and connector fees are monthly forever. A $2,000 setup with a $300/month running cost is a $5,600 first-year commitment, not a $2,000 one. Always ask for the twelve-month figure.
Your own time during the build. Someone in your business has to explain the process, review the output, and test the result. Budget 5–15 hours of internal time for a straightforward workflow. It is not optional and it is not free.
Maintenance when something upstream changes. Software vendors change their interfaces. Your process changes. Something that worked in March needs attention in September. Expect a few hours a year per workflow, and ask who is on the hook for it.
Training and adoption. An automation your team does not trust gets quietly bypassed. This is the single most common way automation money is wasted — not a technical failure, but a workflow everyone routes around within two months.
How to work out whether it is worth it
The arithmetic is simpler than the sales process suggests.
Take the task. Estimate the hours it consumes each month — genuinely estimate it, do not guess, because owners are consistently wrong in both directions. Multiply by the loaded hourly cost of whoever does it. Multiply by twelve.
Now compare that against the first-year cost including subscriptions and your own time.
If automation costs less than half of what the manual process costs annually, it is almost certainly worth doing. Between half and roughly equal, it depends on secondary benefits — fewer errors, faster turnaround, less key-person risk. Above the annual manual cost, it is not a financial decision and you should be clear-eyed about why you are doing it.
One task consuming five hours a week is about 250 hours a year. At a $35 loaded hourly cost, that is $8,750 annually. Against a $2,000 setup and $200 a month, the payback lands somewhere around month six. That is the shape of a good automation candidate.
Where the money is usually best spent first
Across the small businesses we talk to, the cheapest meaningful wins cluster in the same places: document capture and filing, chasing people for information, moving data between two systems that should already talk, and assembling the same report every month.
None of it is exciting. All of it is boring, repetitive, high-volume work with clear rules — which is exactly what automation is good at, and exactly what humans are bad at staying accurate on.
The exciting projects tend to be the expensive ones with uncertain payback. Start with boring.
A note on doing nothing
Sometimes the right answer is to wait. If your process is genuinely in flux, if volumes are low, or if you are about to change the underlying software anyway, automating now means automating something that will not exist in six months.
We tell people this reasonably often. It is cheaper to hear it in a fifteen-minute call than to discover it after spending $6,000.
If you want real numbers for your specific situation rather than the ranges above, that is what the $999 AI Tools Assessment produces — named tools, ranked by impact against effort, with the hours and dollars they give back each month. Or start with a free 15-minute call if you want to sanity-check whether there is anything here first.