← Back to blog

How Much Does a Custom AI Agent Cost? Build vs Buy

October 28, 2026

The question I get asked most often, right after "what is an AI agent," is "how much does one cost?" And I get why. You can find a hundred articles about what AI agents can do, and almost none of them will give you a straight number. Vendors hide the price behind a "book a demo" button. Consultants quote a range so wide it is useless. So owners are left guessing, and guessing is a bad way to make a budget decision.

I am going to do the opposite here. I will give you real ranges, explain what pushes a project to the cheap end or the expensive end, and help you decide whether you should build something custom, buy something ready-made, or do nothing yet. These are the numbers I actually see in 2026, working with small and mid-size companies. They are not quotes, and your project could land outside them, but they will put you in the right ballpark, which is more than most of the internet will do.

First, a quick reset on what we are pricing, because "AI agent" covers a lot of ground. In my last post I defined an agent as software that takes a goal, figures out the steps on its own, uses tools to do the work, and checks whether the job got done. The price depends almost entirely on which job, how many tools it touches, and how much judgment it needs. An agent that answers questions from your price list is a different project from an agent that books appointments, updates your CRM, and chases unpaid quotes. Same word, wildly different price tags.

The short answer, up front

If you want the one-paragraph version, here it is. Ready-made AI agents, the kind you subscribe to and turn on, start around $49 a month and go up to a few hundred a month depending on volume. A focused custom agent, built for one specific job in your business, usually costs a few thousand dollars to build, plus a monthly running fee in the low hundreds. A bigger custom system, one that spans several departments and several of your tools, can run into five figures to build. And a full "AI transformation" program from a big consultancy can run into six figures, which is almost never what a small or mid-size business needs, and I will explain why later.

That is the range. Now let us break down why each option costs what it does, so you can tell which bucket your project falls into.

Option 1: Buy a ready-made agent

This is the cheapest and fastest route, and for most businesses it is the right one. Ready-made agents are products built for a common job, sold as a subscription. The classic example is the one we build: AI voice agents that answer your phones, book appointments, and handle routine questions. The product already exists, so you are not paying anyone to invent it. You are paying to configure it for your business, which takes days, not months.

Typical pricing looks like this. A starter plan for a small business runs around $49 to $99 a month. That usually covers a set number of calls or conversations, with extra usage billed on top. A mid-tier plan, for a busier business or one with multiple locations, runs a few hundred a month. Enterprise plans go higher, but if you are reading this blog, you probably do not need one.

What you get for that money: the product, the hosting, the phone infrastructure, the maintenance, and the updates. When the AI model improves, you get the improvement without lifting a finger. When your tools change, the vendor updates the connection. Your main job is to keep your business information current, your prices, your hours, your policies, so the agent gives good answers.

What you do not get: a system shaped exactly around your weird process. If your booking flow has five exceptions and a spreadsheet that only one employee understands, a ready-made product will handle the standard 80 percent and escalate the rest to a human. For most businesses, that 80 percent is where the money is anyway.

The honest downside is the monthly fee never ends, and over years it adds up. A $99 a month product costs about $1,200 a year, and about $6,000 over five years. That is still usually cheaper than a custom build, but it is worth knowing the math before you sign up.

Option 2: Build a custom agent

Custom means someone builds an agent for your specific job, your systems, and your rules. This is the route when no ready-made product covers what you need, or when you have already proven the value with a ready-made product and want the full version.

Here is how the build cost usually breaks down. A simple custom agent, one job, one or two tool connections, clear rules, takes a few weeks and costs a few thousand dollars. Think of an agent that reads incoming quote requests, checks your calendar, and books the promising ones. Or a sales follow-up agent that watches your pipeline and sends the follow-ups nobody on your team has time to send.

A medium custom agent, one job but several tool connections, or a job with real edge cases, takes a month or two and costs in the low five figures. This is the agent that reads your CRM, checks inventory, books appointments across two calendars, sends confirmation texts, and logs everything back where your team can see it. Each connection adds work, and each edge case adds testing.

A large custom system, several jobs across departments, deep integrations, custom dashboards, approval workflows, takes months and costs well into five figures. This is the territory where you should already have proof the thing works, because you are betting real money on it.

On top of the build cost, there is a monthly running cost. This has three parts, and vendors sometimes only mention the first one. First, the platform or hosting fee, usually in the low hundreds a month. Second, the AI model usage, which charges per conversation or per task. For a typical small business this is modest, tens to low hundreds a month, but it scales with volume, so a business handling thousands of calls will pay more. Third, maintenance. Models get updated, your tools change their connections, your prices change, and someone has to keep the agent current. Budget for it, or buy a support plan that includes it. An agent nobody maintains slowly goes stale, and a stale agent giving out last year's prices is worse than no agent at all.

So why would anyone pay for custom when ready-made is so much cheaper? Three good reasons. One, your process is genuinely unusual, and no product covers it. Two, the job touches systems no product connects to, like a niche industry tool. Three, you have already proven the value with a pilot and you want to own the system rather than rent it. If none of those apply, custom is probably not your first move.

Option 3: The pilot, which is what I actually recommend

If you are torn between build and buy, there is a third option that settles the argument with data instead of opinions: run a pilot. Pick one painful, repetitive job, run an agent on it for 30 to 60 days, and measure what happens. A pilot with a ready-made product might cost you one or two months of subscription, a few hundred dollars. A small custom pilot costs a few thousand. Either way, you come out knowing two numbers no vendor can give you: what it actually costs in your business, and what it actually saves or earns.

I recommend this so often that it probably sounds like a dodge. It is not. It is the cheapest way to buy certainty. Owners who skip the pilot tend to make one of two mistakes. They buy a big system nobody uses, because it solved a problem nobody actually had. Or they decide AI is too expensive, because they priced the biggest possible version instead of the smallest useful one. A pilot protects you from both.

Here is how to run one well. Pick a job with clear rules and a measurable outcome, missed calls answered, follow-ups sent, hours saved. Write down the current cost of that job before you start, in hours and in lost business. Run the agent for at least 30 days, with a human reviewing its work in the first weeks. Then compare. If the numbers work, expand. If they do not, you spent a small amount to learn something valuable, and you can drop it without regret.

Build vs buy, side by side

Here is the comparison in one place, because I know this is the table most people came for:

Read that table with one caveat. The cheapest option is not always the cheapest outcome. A $49 a month product that nobody configures properly, fed with an out-of-date price list, will cost you customers. A custom build that solves a real, expensive problem pays for itself in months. The price tag matters less than the gap between what the job costs you today and what the agent costs you tomorrow. That gap is the whole decision.

What pushes a custom project to the expensive end

If you do go custom, it helps to know what makes builders quote high, so you can control it. The biggest driver is integrations. Every system the agent has to talk to, your phone system, your calendar, your CRM, your booking tool, your payment system, is a connection that has to be built, tested, and maintained. One connection is simple. Five connections is a project. Before you ask for a quote, list the systems involved, and ask yourself whether the agent really needs all of them on day one. Often it does not, and starting with two connections instead of five cuts the price sharply.

The second driver is unclear rules. "Use your best judgment" is not a specification, and a builder who has to guess what you meant will either quote high to cover the guessing, or build the wrong thing and charge you to fix it. The cheapest thing you can do before any AI project is write down how you handle the job today, step by step, including the exceptions. That document becomes the agent's playbook, and it makes quotes tighter and results better. I have seen owners cut a quote nearly in half just by showing up with a clear process doc.

The third driver is scope creep, and this one is on you as much as the builder. A project that starts as "answer the phones" and quietly grows into "answer the phones, and also handle quotes, and also do the scheduling, and also write the newsletter" is three projects wearing one budget. Keep the first version small. Ship it, measure it, then decide what version two should be.

The fourth driver is data cleanup. An agent is only as good as what you feed it, and if your price list lives in three spreadsheets, two of them out of date, someone has to sort that out before the agent can give a reliable answer. That work is not glamorous, but it is real, and it is often the biggest hidden cost in a project. Ask any builder how they handle your data, and be suspicious of anyone who says it is not a factor.

The costs nobody puts in the quote

Beyond the build and the subscription, there are a few costs owners forget to budget, and I would rather you hear them from me than discover them later.

Your time is the first one. Even a ready-made product needs you, or someone on your team, to provide business information, review early calls or messages, and correct mistakes in the first weeks. Budget a few hours a week at the start. That drops fast as the agent proves itself, but it is not zero, and pretending otherwise is how pilots get abandoned.

Training is the second one. If your team is supposed to work alongside the agent, they need to know how it works, what it can and cannot do, and when to step in. This does not need a big program. A short session and a one-page guide usually does it. But if you skip it, your team will either ignore the agent or fight it, and both waste the money you spent.

The third one is the cost of getting it wrong early. An agent will make mistakes, especially in the first weeks, and a mistake that reaches a customer costs goodwill. That is why guardrails matter: limit what the agent can touch, require approval for anything irreversible like refunds or cancellations, and log everything it does. Guardrails cost a little to build, and they save a lot in damage control.

None of these are reasons not to do it. They are reasons to budget honestly and start small, so the surprises are small too.

How to decide, with real numbers

Let me give you a simple framework you can run on a napkin. It is the one I walk owners through when they ask me whether a custom agent is worth it.

Step one: pick the job. Not "AI for my business," one job. Missed calls after hours, quote follow-ups, research that eats your mornings, appointment reminders. One job.

Step two: price the job today. How many hours a week does your team spend on it? Multiply by a rough hourly cost, including the fact that those hours could have gone to work that actually earns money. Add the business you lose when the job does not get done, the calls that go unanswered, the follow-ups never sent. Be honest, not optimistic. This number is your monthly cost of doing nothing.

Step three: price the agent. For ready-made, that is the monthly subscription plus a little of your time. For custom, that is the build cost spread over, say, two years, plus the monthly running cost. If you are unsure, use the pilot cost instead, because the pilot will give you real numbers.

Step four: compare. If the agent costs clearly less than the job costs you today, and the pilot backs that up, do it. If the numbers are close, run a pilot and let the data decide. If the agent costs more than the job, do not do it yet, or pick a different job. There is no shame in that. Not every job is worth automating, and the owners who do best with AI are the ones who automate the expensive, repetitive jobs and leave the rest to humans.

One more thing about payback time, because owners always ask. A ready-made agent that captures even a few jobs a month you would otherwise have missed usually pays for itself in the first month. A custom build typically pays for itself in months, not weeks, and you should be suspicious of anyone promising otherwise. If a builder tells you a five-figure project pays for itself in 30 days, ask them to show you the math for a business like yours, not a best-case slide.

What I would not spend money on

I want to be direct about the things I see owners waste money on, because avoiding a bad spend is as valuable as making a good one.

I would not buy a six-figure "AI transformation" program before running a single pilot. Transformation is a word consultants use when they want to sell you everything at once. Your business does not need transforming. It needs one painful job handled better, then another, then another. The transformation, if it happens, is the sum of those small wins, and it happens on your schedule, not a consultant's.

I would not build a custom agent for a job a ready-made product already does well. I have watched owners spend five figures building a phone-answering agent that a $99 a month product would have handled, because they liked the idea of owning it. Ownership is nice, but it is not worth ten times the price for the same outcome. Own the things that make you different. Rent the things that are the same for everyone.

I would not pay for features nobody asked for. Dashboards nobody opens, integrations nobody uses, a second language your customers do not speak. Every feature is a cost, in build and in maintenance, and the best agents I have seen are boring in the best way: they do one job, reliably, around the clock.

And I would not sign a long contract before a pilot. A vendor confident in their product will let you try it for a month or two. A vendor who needs a year-long commitment up front is telling you something, and it is not confidence.

So, build or buy?

Here is my honest answer, after building these systems and watching owners use them. For most small and mid-size businesses, the right order is: buy first, pilot it, and build custom only when you have proof. Buy a ready-made agent for the common job, usually answering calls or chasing follow-ups, because it is cheap, fast, and proven. Run it for a month or two and measure it. If it works and you hit its limits, then talk about custom, with real data about what the custom version needs to do differently.

That order saves you money twice. It saves you the custom build cost on jobs a product already handles, and it makes the custom build cheaper when you do need one, because you already know the process, the data, and the outcome you are buying.

If you are still unsure, that is fine, and it is actually a good sign. It means you are thinking about the money, not the hype, and that is exactly the mindset that gets good results from AI. Start with one job, price it honestly, run a pilot, and let the numbers decide. If you want help thinking through which job to start with, or what a build would actually cost for your specific process, that is the kind of conversation AI consulting is for, and I would rather have that conversation before you spend anything than after.

The technology is real, the prices are knowable, and the decision is simpler than the industry wants you to believe. It is not "should my business use AI." It is "which job, what does it cost me today, and what does the agent cost." Answer those three questions, and the build vs buy decision mostly makes itself.

Frequently asked questions

How much does a custom AI agent cost?

Is it cheaper to buy a ready-made AI agent than to build one?

What ongoing costs come with an AI agent?

What drives the price of a custom AI agent up?

How do I know if a custom AI agent is worth it?

Should I start with a pilot or go straight to a custom build?