A 2027 AI budget for 50 to 100 people

The short answer

For a 50-person company, plan on about $52,000 a year once it's all running. At 100 people it's about $99,000. That's around 1.4% of payroll, before agents, which get priced one at a time from your own volume.

We started from Michael Domanic's 2027 budget from Section. His is built for 1,000 people and comes to about $6.6 million, and it's worth reading the whole thing. But his biggest line assumes you pay for every use on top of each seat, which is how big companies buy. Most of the companies we work with buy seat plans, so we rebuilt it that way.

What we assumed

For the examples we figured about $75,000 a person once you count taxes and benefits. That's on the conservative side. The Bureau of Labor Statistics puts the average across private employers closer to $97,000, and Section's example works out to about $110,000. Payroll matters less here than in Section's version, but you'll still want your own number for the payback math.

We also figured 1 in 10 people are technical, and that about 1 in 10 of everyone else will use AI heavily. If some of your team never sits at a computer, like drivers or a plant crew, leave them out.

Seats

Seats are the easy part, because the vendor tells you the price. Here's what each one listed on 4 October 2026.

Copilot and Gemini aren't apples to apples. Copilot sits on top of a Microsoft 365 plan you pay for separately. Gemini comes with Google Workspace, so if you're already on Business Standard or Plus, you may have it already. We left ChatGPT off because OpenAI's pricing page wouldn't load for us, and we didn't want to guess.

Usage

Section's biggest line is usage, the cost every time someone asks a question or runs a task. He budgets it at 2 to 5% of salary for desk roles and 5 to 7% for technical people. At $75,000 that's around $200 a month per person on top of the seat.

That's real for a company on an enterprise plan, where use is billed on top. On a seat plan, most of it is already in the price. People who use AI all day move up to a premium seat instead, which is $125 a month on Claude Team. So we put your technical people and your heaviest users on premium seats, everyone else on standard, and added 25% for people moving up a tier as they get better at it.

Where your company lands depends a lot on the work. A farm, a plant or a trucking company has a lot of people who'll never need a seat. A law firm, an accounting firm or an agency has nearly everyone at a desk reading and writing all day, so expect more premium seats there.

Agents

Domanic prices agents one business process at a time, at $1,500 to $2,000 a month each, and says every one should launch with a cost cap and a clear idea of what it's supposed to return. His are big agents that work across several departments of a 1,000-person company. Most of what a 50 to 100 person company needs runs one process, and that usually costs a lot less.

We left agents out of the totals because the real number depends on how much work each one does, and we'd only be guessing at yours. The model is usually the cheap part anyway. Building the agent and the person who checks its work cost more.

Training and outside help

Two of Domanic's lines don't fit here. He budgets $200,000 to $400,000 for a Head of AI and about $200,000 for a coaching platform. Either one is more than this whole budget.

Someone still has to own it, though. At this size that's usually the owner or an ops lead, and their time is already on payroll. Keep them to one or two processes at a time. Domanic's point is that AI run as a side project doesn't change much, and that holds at any size.

Three things carry over. A few training events, which he puts at about $2,000 each. Something for the handful of people already ahead on AI, since he says paying them gets you more than asking them to volunteer. And outside help, which he counts on this line. We put our own Discovery price in, since that's the number on this line we can stand behind.

The ramp

Don't ask for the whole year on January 1. Domanic suggests 40% of the full amount in the first quarter, 80% in the second and all of it from the third, with each step tied to a target you agree on up front with whoever signs off. That puts the year at about 80% of the full number.

The 50-person budget

About $52,000 for the full year, or $43,000 in 2027 with the ramp. At 1.4% of payroll, AI needs to save or earn about two thirds of one person's pay a year to cover it. Domanic says Section is about 50 people and expects its AI spend to buy at least two people's worth, so that's a bar most teams should clear.

The 100-person budget

About $99,000 for the full year and $82,000 with the ramp, 1.3% of payroll. In both budgets, how many premium seats you need moves the number more than anything else.

Paying for it

Somebody's going to ask what you're cutting. Domanic's biggest lever is not refilling some of the jobs that open up. At 50 to 100 people you only lose a few a year, so the version here is smaller: before you post the next job, ask whether AI plus someone already on the team could cover it.

His other levers work at any size. He expects outsourced content, research and support costs to drop 15 to 25%. He says about a third of software licenses go unused. And he pays for training out of the training budget you already have. A smaller company has a shorter software list, so that's usually the quickest place to look.

Saving more than you spend

The budget is the easy half. Making sure it comes back to you is the harder half, and it's most of what we do.

Here's the math on one process, with round numbers.

Hand it to an agent and someone still checks each invoice, say a minute apiece. That's 20 hours a month instead of 120, or about $43,000 a year back. One process like that covers most of the 50-person budget, and a second one covers the rest.

We've seen it in real work. At Genesco Sports Enterprises, a month-end reconciliation that took their CFO about four days now runs in about 60 seconds. At Blue Sky Farms and McHur Care, replacing software they paid for every year now saves each of them six figures a year. That's more than the 100-person budget above, from one piece of work.

Here's how we keep it pointed that way.

We start where the hours are. The audit is free and takes about 15 minutes. You talk through how the work gets done, and it scores each process on how often it runs, how settled it is and what it's worth, then tells you which ones to hand over first. You keep the report either way.

We count before we quote. Discovery is $7,500 over two weeks for your top three processes, and $2,500 for each one after that. We sit with the people doing the work and count how often it happens, how long it takes, how long checking it takes, and what those hours cost. A guess doesn't count, including ours. The build quote then shows the price, what each agent should cost to run each month, a cap 25% above that, and the payback after running costs, all worked out from your own numbers.

We tell you when it won't pay. If an agent would cost more to run than it gives back, you'll hear that from us before you spend anything on it.

We check after. Before anything changes, we write down the number you'll judge the build by. Once it ships, we measure it again and compare it to where you started. Domanic says one real before-and-after carries more weight in a budget meeting than any projection.

We look for what you can stop paying for. Some tools you pay for every year can be replaced outright with something you own. That's where the Blue Sky and McHur Care savings came from.

Frequently asked questions

Why is this so much lower than Section's number?

Section's method assumes you pay for every use on top of the seat, which is how large companies on enterprise plans buy. Most companies with 50 to 100 people buy seat plans where everyday use is already in the price. Run his percentages on our 50-person example and you'd get about $209,000, or 5.6% of payroll. Treat that as the high end if your team ends up paying by use.

What if we don't have any AI wins yet?

Ask for less. Domanic says to start small, name the work you'll go after first, and tie it to numbers your leadership already watches. Fund the seats and one agent, show what happened, then ask for more.

Do we need a Head of AI?

Not at 50 to 100 people. Section budgets $200,000 to $400,000 for one, which is more than this whole budget. Someone still has to own it, though. At this size that's usually the owner or an ops lead.

Why isn't there a number for agents?

Because it depends almost entirely on how much work the agent does. Section's figure is for big agents running across a 1,000-person company, at $1,500 to $2,000 a month each. An agent that handles one process at a smaller company usually costs far less, but the honest number comes from counting your own volume, which is what Discovery does.

How do we know if it's paying off?

Divide the total by payroll. In our 50-person example that's 1.4%, about two thirds of one person's pay a year in time or savings. Then pick one piece of work and measure it before and after, using a number you already track.

Is this on top of what we already pay for software?

Yes. Section keeps your existing software, cloud and data costs separate. The servers your agents run on count as AI spend.