Why compare four options?
Most decisions about getting work done have four realistic options: hire a person, engage an agency or contractor, hire an AI employee, or leave the work undone. Comparing only two of them, usually hire versus AI, misses the option the team may actually be living with, which is the fourth.
The framework below puts all four on the same five dimensions. It supplies no figures; those come from your recorded baseline and from the pricing page and workspace usage for the AI option.
What is the cost structure of each?
A hire is a salary plus on-costs, fixed regardless of output. An agency is a retainer or project fee, often with scope-based additions. An AI employee on ftes.ai is a metered block of work for the organization, shared by all roles, plus human approval time. Doing nothing is the value of the work foregone, which is real even when it is not on any invoice.
Write each down for the work in question. The structure matters as much as the amount: fixed cost is cheaper at high utilisation and more expensive at low; metered cost tracks the work.
What is the spending ceiling of each?
A hire's ceiling is soft: overtime, additional hires, and the cost of a bad hire. An agency's ceiling is scope: additions, change requests, and the next project. An AI employee's ceiling is hard: the wallet pauses work when empty, and nothing runs on credit. Doing nothing has no ceiling on the foregone value; it accumulates.
For a finance leader, the ceiling dimension is often where the AI option distinguishes itself, because it is the only one with a mechanical stop.
- Hire: soft ceiling, overtime and additional headcount
- Agency: scope ceiling, change requests and renewals
- AI employee: hard ceiling, wallet pauses work when empty
- Nothing: foregone value accumulates without limit
How do ramp and control compare?
Ramp time for a hire is recruitment plus onboarding, typically months. For an agency it is the engagement setup, typically weeks. For an AI employee it is the first-month setup: connections, remit, and close review of early proposals, with useful proposals in the first week. Doing nothing has no ramp and no output.
Control is the reverse of what people expect. A hire is controlled through management, which is high-touch. An agency is controlled through the contract, which is low-visibility. An AI employee is controlled through the approval gate, the remit, and the autonomy ladder, with every action logged, which is high-visibility and adjustable at any time. Doing nothing offers total control over nothing.
What happens when it ends?
When a hire leaves, their knowledge leaves with them unless it was documented. When an agency engagement ends, the work product is yours if the contract says so, and the relationships and methods usually are not. When an AI employee is stopped, its Memory tab, its approval record, and its audit trail remain yours, and credentials are revoked from one place.
Fill in the five dimensions for your specific work, using your baseline for the costs and the ftes.ai pricing page and workspace for the AI figures. The option that wins on most dimensions for that work is usually clear, and it is not always the AI employee; judgement-heavy and relationship-heavy work still points to a person.
What does the completed framework look like for one function?
As a labelled example for a support function, with your figures to fill in: for a hire, cost structure is salary and on-costs, ceiling is soft through overtime and additional headcount, ramp is months of recruiting and onboarding, control is management, and the end is knowledge leaving with the person. For an outsourced team, cost is a retainer or per-ticket fee, ceiling is scope, ramp is weeks, control is a service-level report, and the end is the provider's record staying with the provider.
For a Customer Support Digital FTE, cost is metered work from one wallet plus approval time, ceiling is hard because the wallet pauses work at zero, ramp is the first month's setup with useful proposals in week one, control is the approval gate and autonomy ladder with every action logged, and the end is an audit trail and confirmed memory that stay yours. For doing nothing, cost is the foregone value of tickets answered late, the ceiling is none, and the other dimensions are empty. Filled in with your baseline, the table usually makes the decision for the repetitive first line obvious and leaves the judgement work pointing, correctly, at people.