AI Software for Small Business: What to Actually Pay For

The short version: most AI software for small business costs far less than the sales pitches imply, and the biggest money mistake is not overpaying for one tool, it is paying three times for the same capability without noticing. The payments data backs this up: the typical small business that pays for AI at all spends around $28 a month on it. Before you buy anything new, audit what you already own, because Microsoft and Google have both folded AI into plans you are probably already on. Then buy one thing, monthly, and give it a number to hit.

This guide is about the money decision specifically. Which tool to use for which job is a separate question, covered in our roundup of AI tools organized by the job you need done. This one is about what to actually pay for, how to read a 2026 pricing page without getting surprised, and how to tell whether a subscription earned its place on your card.

What does AI software for a small business actually cost?

Less than almost everyone assumes, and the gap between the assumption and the reality is where most bad purchases happen.

The JPMorganChase Institute analyzed payments data from millions of small businesses through the end of 2025 and found that median monthly AI spending peaked around $80 in 2022 and then fell to roughly $28 by 2025. Typical entry-level subscriptions landed around $20 to $30 a month. Adoption climbed over the same period, from 1.7% of firms in January 2019 to 17.7% by December 2025, so this is not a case of spending falling because interest fell. More businesses are paying, and they are paying less.

The composition shifted too. The Institute sorted AI spenders into three tiers: $1 to $40 a month, $41 to $150, and above $150. The bottom tier grew from 38% of AI-paying small businesses in 2019 to 63% in 2025. Nearly two thirds of small businesses that pay for AI are spending under $40 a month on it.

Hold that number next to the pitch you are most likely to hear. A reputation and messaging suite quoting you $400 a month is not wrong to charge that, it is bundling review management, a multi-channel inbox, and text automation into one line item. But it is roughly fourteen times what the median AI-paying small business spends, and you should know that before you sign, not after. The question is not whether the product is good. It is whether you have proven you will use fourteen times the median.

What AI am I already paying for without realizing it?

This is the single highest-return thing to do before spending another dollar, and most owners skip it.

Over the past two years, the software you already run has absorbed AI features into base plans rather than selling them separately. Google Workspace now includes Gemini across all paid plans at no additional charge, with Business Standard at $14 per user per month carrying Gemini through Gmail, Docs, Sheets, Slides, and Meet. Google previously sold that as a separate add-on and folded it in. If you pay for Workspace, you already have an AI assistant, and you may also be paying $20 a month for a second one.

Microsoft now sells Microsoft 365 Business Standard with Copilot at $23.50 per user per month and Business Premium with Copilot at $32, both with Copilot included rather than bolted on. Business Basic at $7 does not include it, and the standalone Copilot Business add-on runs $18 per user per month promotionally against a $21 list price. The lesson is not which plan is best. It is that the answer to “do we have AI” now depends on which SKU your reseller put you on eighteen months ago, and almost nobody knows that off the top of their head.

The same absorption happened lower down the stack. Accounting platforms including Wave, Xero, and QuickBooks Online read receipts and auto-categorize transactions inside plans you already hold. Scheduling tools like Calendly and Acuity added AI notetaking and smarter routing. Canva put its generative features inside Pro rather than charging separately.

So the first move is an inventory, not a purchase. Open your card statement or your accounting software, list every software subscription with its monthly price, and next to each one write what AI features it already includes. It takes about forty minutes. Owners who do this routinely find one of two things: a capability they were about to buy is already sitting inside a plan they own, or a subscription nobody has opened in five months is still billing.

When is a free tier enough, and when is it a trap?

Free tiers in this category are usually real, and they are usually capped in a way that matters more than the price does.

Read the limit, not the label. Zapier’s free plan allows 100 tasks a month and two-step workflows, which is genuinely enough to test whether an automation is worth building and not enough to run a business on. Make gives 1,000 credits a month free. Assistant free tiers from ChatGPT, Claude, and Gemini are usable for real work, with limits that tighten as you lean on them. We looked at how quickly one of those windows can move in our piece on free AI tools and the unlimited ChatGPT window.

Free automation tiers are worth a special mention, because they are usually enough to answer the only question that matters at that stage. If you cannot build a useful workflow inside 100 free Zapier tasks a month, the problem is almost never the task limit, it is that the workflow is not yet clear enough to automate, which is the argument in our guide to what to automate first and in what order.

The trap is not the cap itself, it is building a habit on a ceiling you have not read. A free plan that runs out in week two of a thirty-day trial does not tell you the tool failed, it tells you nothing at all, and you will probably conclude the wrong thing. Before you start a trial, find the specific number: how many tasks, minutes, conversations, or seats. Then estimate whether your real weekly volume fits under it. If it does not, either pay for the month deliberately so the test is valid, or pick a smaller test.

The other honest use of a free tier is as a permanent answer. Plenty of small businesses run bookkeeping on Wave’s free tier and a general assistant on a free plan indefinitely, and that is a complete AI stack for a solo operator. Free is not a lesser version of buying. Sometimes it is the correct final decision.

How do I read an AI pricing page in 2026?

Three pricing models are now mixed together across this market, and they fail in different ways.

Per seat. A fixed price per person per month. This is the easiest to forecast and the easiest to waste, because seats keep getting provisioned and almost never get removed. The failure mode is paying for four seats when two people log in.

Per usage or per outcome. You pay for what the software does, not who holds a license. Fin, the AI support agent, charges $0.99 per outcome with a minimum of 50 outcomes a month, and no seat cost when paired with a non-Intercom helpdesk. That is elegantly aligned: you pay when it resolves something. It is also uncapped by design. A quiet month costs you the $49.50 floor, and a viral product complaint costs you considerably more, at exactly the moment you least want a surprise bill. If you buy on this model, find out whether the vendor offers a spend cap or an alert, and turn it on the day you sign.

Per volume, inside one plan name. This is the one that catches people. Zapier’s Professional plan is quoted from $19.99 a month, and the same plan name runs up to $3,299 a month as your task volume climbs. Nothing about the tier is misleading, the price simply scales with usage inside a single label. When someone tells you they are “on Zapier Pro,” that sentence contains no price information at all.

Two habits handle all three. First, before signing, write down what you expect this to cost in month twelve at your realistic volume, not month one at your trial volume. Second, check the date beside the price. This market repriced repeatedly through 2026, in both directions, which we covered in why the number on an AI pricing page now matters less than the date beside it. Promotional rates carry expiry dates, and the Microsoft Copilot add-on price above is a live example of one.

What is actually worth paying for?

Run a new subscription through three questions before it goes on the card.

Is it doing a job that costs me real money right now? Not a job that sounds valuable, one you can name a number against. Missed calls, quotes going out two days late, invoices nobody chased, hours lost to retyping the same address into three systems. If you cannot name the leak, you are buying a capability, not a fix, and capabilities do not pay for themselves.

Is this the cheapest thing that does that job? The temptation is to buy the platform that also does six other things you might want later. Later rarely arrives, and the six other things are why the price is four times higher. Buy narrow. You can always upgrade into the suite once one lane is proven, and you will negotiate better having used it.

Would anyone notice if I cancelled it tomorrow? Ask this at the ninety-day mark, in writing, about every AI line item. It is a blunter and more useful question than any ROI framework, because the honest answer arrives instantly.

There is a fourth question worth asking about anything customer-facing, and it is not about money. Does this tool hand off cleanly to a person when it should? A support agent that guesses instead of escalating, or a phone tool that traps a caller in a loop, costs you more than its subscription in a single bad interaction. Our guide to AI customer service and the handoff rules that make it work covers what that scope should say before you buy.

How do I stop subscription creep before it starts?

Subscription creep is not caused by buying bad software. It is caused by nobody owning the renewal.

Four practices, none of which take long:

• Keep one ledger. A single sheet with every software subscription, its monthly cost, its renewal date, and one named person responsible for it. Not a folder of emails. One sheet somebody actually opens.
• Default to monthly for the first ninety days. Annual plans typically save 15% to 20%, which is real money and a bad trade on a tool you have used for three weeks. Switch to annual once a tool has survived a quarter, not before.
• Put renewal dates on a calendar with a reminder two weeks out. The reminder is the entire mechanism. Renewals that arrive unannounced get paid; renewals you see coming get a decision.
• Cull once a quarter. Thirty minutes, the ledger open, one question per line: did anyone use this in the last month? Cancel what fails. Nothing on that list is load-bearing enough to survive nobody using it for a month.

One specific overlap to watch for, because it is the most common expensive one: paying for a standalone assistant subscription while also paying for a Workspace or Microsoft 365 plan that includes one, while also paying for a third tool whose main selling point is AI writing. That is three bills for drafting text. Pick one.

How do I know a subscription paid for itself?

Do the arithmetic on the specific leak, with your own numbers. It is simpler than it sounds, and it is the step that turns an opinion into a decision.

The shape is always the same: what does the leak cost per month, and what does the fix cost per month? A local service business counting missed calls for a week, and knowing roughly what an average job is worth and what share of callers become customers, can multiply those together and get a monthly figure for the leak. Set that against the subscription price. If the leak is worth several times the fix, buy it and check again in ninety days. If the two numbers are close, the tool is not obviously wrong, it is just not obviously right either, and you should solve a bigger leak first.

The same arithmetic works on hours. A retailer spending six hours a week writing product descriptions and email campaigns can price those hours at what the person doing them costs, or at what the owner’s hour is worth, and compare it to a $20 assistant subscription. That comparison is usually lopsided enough to decide in one line. What it does not tell you is whether the recovered hours went anywhere useful, which is the follow-up question owners skip. Two hours back is only a return if those two hours went into follow-ups, estimates, or leaving earlier. If they quietly refilled with other busywork, you rearranged your week rather than improving it.

A professional practice has a third variable worth pricing: rework. If an AI notetaker means intake details stop getting re-asked and files are complete before the second meeting, the saving is the follow-up emails that never had to be sent, not the transcription itself.

None of this requires a spreadsheet model. It requires one number written down before you buy and the same number collected again ninety days later, which is the measurement discipline in our 30-day path for getting started with AI.

What should a small business AI budget actually look like?

Three sensible shapes, matched to the tiers the payments data actually shows.

Under $40 a month. Where roughly two thirds of AI-paying small businesses sit. One general-purpose assistant at around $20, plus whatever is already bundled in your existing plans, plus free tiers for notes and light automation. For a solo operator or a very small team, this is not a starter budget you are supposed to outgrow. It is frequently the right permanent answer.

$41 to $150 a month. The assistant, plus one or two paid tools aimed at specific proven leaks: a scheduling platform with reminders, a paid automation plan, a chatbot trained on your own content, or the AI tier of software you already run. Everything in this band should be traceable to a named problem you measured before buying.

Above $150 a month. Usually one bundled platform doing several jobs at once, most often in the missed-call, messaging, and reviews lane. This can be entirely correct for a business where a single lost lead is worth hundreds of dollars. It is the wrong first purchase for a business that has not yet proven it will change one habit for a month, because you are buying capacity you have no evidence you will use.

The order matters more than the number. Audit what you own, spend nothing new for a week, then buy the cheapest tool that fixes your most expensive leak, monthly, with a date on the calendar to judge it. Once one tool is genuinely earning, the next decision is easier and better informed. That sequencing, one task at a time, is the through-line of the practical guide to AI for small business this playbook belongs to, and it applies to spending just as much as it applies to workflows.

One closing note on what you are buying, because it shapes the budget more than any price comparison. None of this software is priced as a substitute for a person, and buying it as one is how businesses end up disappointed on both counts. What a $20 assistant or a $150 platform actually buys is the work that was already falling through: the after-hours call nobody was going to answer, the follow-up nobody had time to write, the receipts that would have sat until March. Priced against that, the numbers in this guide are small. Priced against a salary, they will always look either suspiciously cheap or wildly insufficient, because that was never the comparison.

Frequently asked questions

How much should a small business spend on AI software per month?

Start under $40 and only go higher when a specific measured leak justifies it. That is not a cautious recommendation, it is where the market actually sits: the JPMorganChase Institute found median monthly AI spending among small businesses at roughly $28 in 2025, with 63% of AI-paying small businesses in the $1 to $40 band. If a vendor implies that number should be several hundred dollars, ask what specifically it fixes and what that leak costs you today.

Should I buy an annual plan to get the discount?

Not in the first ninety days. Annual billing typically saves 15% to 20%, which is worth taking on a tool that has proven itself over a quarter and is a poor trade on one you have used for three weeks. AI pricing and free tiers also moved repeatedly through 2026, so a twelve-month commitment signed this quarter can look wrong by the next one.

Is free AI software good enough for a small business?

Often, yes, and permanently so for a solo operator. Free tiers cover real bookkeeping, real note-taking, real drafting, and light automation. The thing to check is never the price, it is the specific cap: tasks, minutes, conversations, or seats. A free plan that runs out mid-experiment will teach you the wrong lesson about the tool.

What is the most common way small businesses waste money on AI?

Paying more than once for the same capability. The usual version is a standalone assistant subscription running alongside a Google Workspace or Microsoft 365 plan that already includes one. Google now bundles Gemini into all paid Workspace plans, and Microsoft includes Copilot in its Business Standard and Business Premium SKUs, so a subscription you signed two years ago may have quietly acquired the feature you are separately buying.

What is the difference between per-seat and usage-based AI pricing?

Per seat charges a fixed amount per person per month, which is predictable and easy to over-provision. Usage-based charges for what the software does, such as Fin’s $0.99 per resolved outcome, which aligns cost with value and removes the ceiling on a busy month. If you buy usage-based, set a spend cap or a billing alert on day one rather than discovering the model during your worst week.

Do I need a bundled platform or several separate tools?

Separate tools first, in almost every case. Buy narrow against one proven leak, prove it earns, then consider consolidating into a suite once you know which lanes you actually use. Buying the bundle first means paying for five capabilities to fix one problem, and it removes the evidence you would have used to negotiate.

What is the oldest AI subscription on your card, and would anyone notice if you cancelled it tomorrow? Tell us in the comments.