How to Budget for Digital FTEs: A Finance Guide

Most AI agent budgets are built the wrong way round: a license or usage line is estimated from model prices, a savings figure is attached, and the plan is approved. Six months later the usage line is over, the savings are unmeasured, and the exception queue is staffed by people whose time was never budgeted. Digital FTEs belong in the workforce plan, budgeted as roles with volumes, cost per task, oversight, and a share of the platform. This guide walks finance partners and operations leaders through that process. It builds on Digital FTE cost and the Digital FTE workforce planning whitepaper.

Where do Digital FTEs sit in the plan?

In the operating budget of the function that owns the workflow, alongside human headcount, with the shared platform funded as a central service and allocated. This placement does three things: it makes the process owner accountable for the role's economics, it stops the platform from being charged to the first team that used it, and it lets the workforce plan show human and digital capacity together.

What are the budget components per role?

Component	How to estimate	Source
Volume	Eligible cases per period, with seasonality and growth	Process data
Inference	Steps per task × tokens per step × model price	Agent profile from shadow mode; contract prices
Retrieval and tools	Tool calls per task × unit costs	Agent profile; platform prices
Infrastructure	Runtime, sandboxes, storage, observability	Platform team
Oversight	Exception and approval rates × handling time × loaded cost	Process data; autonomy level
Platform share	Central platform cost ÷ roles, or by usage	Finance allocation rule
Maintenance	Engineering hours for specs, evaluations, migrations	Engineering estimate
Reserves	Model migration, incident response, volume variance	Policy

Before shadow mode, the agent profile is an estimate; after two to four weeks of shadow data it is observed, and the budget should be re-based on it. The unit economics method is in the AI agent unit economics whitepaper.

How should the budget be phased?

Digital FTE costs follow the autonomy curve, and the budget should too.

Phase	Typical duration	Budget characteristics
Build and shadow	Weeks	Engineering, evaluation build, full human cost continues; no savings
Suggest	Weeks to months	High oversight; inference at full volume; first quality evidence
Act with approval	Months	Oversight falls to consequential steps; exceptions shrinking
Act with sampling	Ongoing	Oversight at sample plus exceptions; cost per task at its floor

Tie the phase transitions to evaluation evidence rather than dates. A budget that assumes act-with-sampling economics from month one will be over by month two; a budget that funds the curve will be accurate and defensible. The evidence for each transition is described in Digital FTE performance review.

How is oversight capacity budgeted?

Oversight is the component most often left out and the one that determines whether the process speeds up or stalls. Budget it as people: the exception handlers and approvers, their hours per period at each autonomy level, and where that time comes from. In most cases it comes from the same team that previously did the standard-path work, redeployed; the budget should show that redeployment explicitly, along with where freed time goes as autonomy rises. Under-staffing the exception queue is the single most common cause of a Digital FTE that "works" but makes the process slower.

How is the platform allocated?

The gateway, integration layer, evaluation tooling, and security controls serve every role. Fund them centrally and allocate by a simple rule, usage or an even split, revisited annually. The economics of the fleet depend on this: the first role should carry a fair share, not the whole platform, or the pilot looks uneconomic and the fleet never gets approved. Platform components are described in the LLM gateway architecture whitepaper.

What controls keep the budget honest?

1. Per-role budgets enforced at the gateway, with soft alerts and hard limits.
2. Cost per task tracked next to quality on one dashboard; cost is never optimized blind. A build guide is in how to build an AI cost dashboard.
3. Quarterly re-forecast from actual cost per task and volume.
4. Autonomy advances approved by the owner with evidence, which is also when the budget steps down per task.
5. Change gates that check cost per task on every prompt, model, or tool change.
6. Annual review of each role against the alternative, retiring roles that do not earn their place.

How does the business case fit the budget?

The business case compares the role's cost per task over three years with the fully loaded alternative, human or outsourced, at equal quality, including errors, rework, and delay. The budget is the first year of that model at the planned phasing. Keeping the two consistent, same volumes, same phasing, same alternative, prevents the familiar gap between the approved case and the actual spend. The comparison method is in Digital FTE vs human FTE.

What are the common mistakes?

1. A software line instead of a role.
2. End-state economics from day one.
3. Unbudgeted oversight.
4. The whole platform on the first role.
5. No maintenance or migration reserve.
6. No budget enforcement, so volume grows because the agent works.

How does FISTA Solutions help?

FISTA Solutions builds Digital FTE budgets with finance and operations partners as part of its AI enablement practice, measures the real agent profile in shadow mode through forward deployed engineers, and delivers every role as a governed AI agent with the cost telemetry the budget needs. FISTA has delivered 150+ projects for 50+ companies across 12+ countries with 47% average efficiency gains.

To build a budget for your first roles, message FISTA on WhatsApp, or read how to launch your first Digital FTE for the sequencing.