Employee AI assistants all promise the same outcome, fewer tickets and faster answers, but they bill for it in at least six different shapes. One vendor quotes per employee, another per agent seat, a third by consumption, and a fourth says it is "free with your suite." Put those quotes side by side in an RFP and you are comparing apples to oranges before you have even reached the demo.
This guide defines the six pricing models you will actually encounter, explains who each one favors, and gives you a decision framework by headcount so you can model a real year-one cost. The aim is to walk into procurement able to normalize every quote to the same basis and spot where a low headline hides a high total.
How it works: You pay a flat rate for every employee in scope, regardless of how much each one uses the assistant. Your bill is a function of headcount.
Pros and cons: Highly predictable and easy for Finance to forecast; fair when most employees benefit. The trade-off is that you pay for the whole population even if adoption starts low.
Best for: Mid-market HR and IT teams that want budgetable operating cost and broad employee access.
Contract watch-outs: Pin down the definition of a billable employee (see the RFP section below) and confirm whether contractors and seasonal staff count.
How it works: You pay per licensed user or per support agent, similar to most SaaS tools.
Pros and cons: Familiar and simple to reason about; efficient when only a defined group uses the tool. It gets expensive and awkward for an employee-facing assistant meant for everyone, since you either license the whole company or ration access.
Best for: Agent-facing tools used by a bounded support team rather than the whole workforce.
Contract watch-outs: Watch for minimum seat commitments and true-up rules as usage spreads.
How it works: You pay for usage, measured in tokens, messages, queries, or credits. Microsoft's Copilot Studio agents, for example, bill on consumption at $0.01 per credit or in prepaid packs of 25,000 credits for $200 per month (alphavima, 2026).
Pros and cons: Efficient at low or uneven volume and you only pay for what you use. The downside is volatility: a busy enrollment window or a chatty agentic workflow can push the bill up unpredictably, and forecasting gets harder as adoption grows.
Best for: Narrow, low-volume use cases, or teams comfortable managing a usage meter.
Contract watch-outs: Model a peak month, not an average one, and confirm what one unit actually represents.
How it works: A fixed platform fee covers access, and usage charges stack on top.
Pros and cons: Combines a predictable base with pay-for-what-you-use flexibility. The risk is the worst of both worlds if the platform fee is high and usage is also metered aggressively.
Best for: Organizations that want a stable floor plus room to scale usage.
Contract watch-outs: Get the usage rate and any included allowance in writing, and model where the two lines cross.
How it works: The assistant is included with, or added onto, a suite you already own. Microsoft 365 Copilot is the clearest example: a $30 per user per month enterprise add-on that requires a qualifying base license, so the true all-in cost typically lands between about $34 and $87 per user per month depending on tier (GoSearch, 2026).
Pros and cons: Convenient procurement and one vendor. But "included" often means a general assistant, not end-to-end employee support, and the cost scales with headcount rather than outcomes. As one analysis puts it, if 2,000 employees are licensed you pay for 2,000 whether 200 use it daily or not.
Best for: Teams whose needs are met by the suite's native capability and who value single-vendor simplicity.
Contract watch-outs: Separate what the base license already covers from what the AI add-on costs, and confirm whether the bundled assistant actually resolves HR and IT requests or just drafts and summarizes.
How it works: You pay per resolved ticket, per deflection, or against another outcome metric.
Pros and cons: Attractive because cost tracks value delivered. The catch is definitional: what counts as a "resolved" or "deflected" ticket, and who audits the count? Misaligned definitions can inflate the bill.
Best for: Buyers who want cost tied to results and are willing to negotiate airtight metric definitions.
Contract watch-outs: Nail down the exact outcome definition, the measurement method, and whether re-opened tickets are excluded.
These are directional patterns, not quotes. Actual costs depend on your vendors, tiers, and negotiated terms.
Around 800 employees. At smaller scale with modest volume, consumption can look cheap on paper, but the admin overhead of watching a meter often outweighs the savings. PEPM tends to win on simplicity, and a suite bundle can make sense if the native capability genuinely covers your needs.
Around 2,500 employees. This is where predictability starts to matter. Consumption volatility becomes a Finance headache across a larger population, and seat-based pricing for an all-employee tool gets expensive. PEPM usually offers the cleanest forecast, and outcome-based can be attractive if you can agree on definitions.
Around 8,000 employees. At scale, suite bundles look convenient but the per-user math compounds fast, and you may be paying for thousands of light or non-users. PEPM keeps the forecast flat and defensible to the board, while heavy consumption models require active cost governance.
The pattern across all three: the larger and broader your deployment, the more a predictable per-employee model tends to beat a variable one, unless your usage is genuinely low. For a deeper look at what drives the usage line, see our explainer on employee AI token costs.
Add these to every RFP so quotes normalize to the same basis:
MeBeBot uses transparent per-employee-per-month pricing, published in bands on the pricing page rather than hidden behind a discovery call. For mid-market HR and IT teams, the appeal is a forecastable operating cost that does not spike with usage, does not require rationing access to a subset of employees, and does not need a usage meter watched month to month. It is not always the lowest headline number in every scenario, but it is the one Finance can budget with confidence.
No. At genuinely low or uneven volume, consumption can cost less because you only pay for what you use. PEPM tends to win as usage grows and broadens, because it removes volatility and the cost of watching a meter. Model both at your real volume, including a peak month.
It varies by vendor. Some bill by full-time employees, others by total headcount including contractors and seasonal staff. This single definition can swing your cost significantly, so pin it down in the RFP before comparing quotes.
Copilot is rarely free in practice. The enterprise add-on is $30 per user per month and requires a qualifying base license, pushing the true all-in cost higher, and it scales with headcount rather than outcomes. A "free chat" tier exists but does not connect to your internal content, so it is not the same as governed employee support.
Whichever owns the tool should own the line, but decide before launch. Predictable PEPM usually sits cleanly with the HR or IT budget. Consumption models often need Finance watching the total, so unclear ownership leads to surprise-invoice friction.
Six pricing models, one job: making sure the way you buy employee AI matches how you will actually use it. Normalize every quote to the same employee definition, model a peak month rather than an average, and run the numbers at your real headcount before the demos dazzle you. For most mid-market teams deploying broadly, a transparent per-employee model is the one Finance can defend without a spreadsheet full of caveats.
Want your real numbers modeled? Book a demo and we will walk your headcount and volume, or see the bands on the pricing page.
Vendor pricing and Microsoft's rates change frequently. Figures here reflect public prices at the time of writing; confirm current terms directly before you buy.
.svg)
.svg){kind=icon}
.svg){kind=icon}