← Go back

Guide // MandateGuard

What spending controls do finance teams need for AI agents?

By the Ingenious AI team · Published 2026-11-14 · Reviewed 2026-11-14

▸ Run the free live demo See pricing

Free covers 1 agent and 1 policy with the full authorize flow, signed receipts and the approval inbox — no card required to start.

Short answer

Finance teams need agent spending to behave like other delegated spending: authorised in advance, bounded, evidenced at decision time and reconcilable. MandateGuard supplies the executable form — policies as delegations of authority, approval thresholds as countersignature bands, hash-chained receipts as the contemporaneous audit trail, verify-chain as the integrity attestation, and evidence export for period-end packs. Finance's programme around it: probationary delegations that widen on clean history, sampled approvals, a monthly three-way reconciliation whose charged-without-authorisation exceptions expose ungated paths, and caps governed as budget lines at planning time.

What does finance actually need from agent spending?

For finance teams, MandateGuard is a pre-commitment control system: it authorises agent spend against budget-like mandates before execution, records every decision in a verifiable ledger, and exports evidence in audit shape. Its controls map to finance concepts directly — caps are delegations of authority, approvals are countersignatures, receipts are the audit trail, and the chain verdict is the integrity attestation finance otherwise has to take on trust.

Translate the mandate into finance language

Finance does not need another dashboard; it needs agent spending to behave like other delegated spending: authorised in advance, bounded by authority limits, evidenced contemporaneously and reconcilable to the ledger. MandateGuard maps cleanly. The policy is the delegation instrument: this agent may commit up to the per-task cap, up to the daily cap, with these counterparties. The approval threshold is the countersignature band. Receipts are the contemporaneous evidence, created at decision time rather than reconstructed at month end. Verify-chain is the integrity check that lets finance rely on the record without re-performing it. When a deployment is described to finance in these terms — delegation, countersignature, evidence, integrity — ownership and budget conversations go noticeably faster.

Delegations of authority, written as caps

Finance already knows how to govern delegations for humans: authority scales with seniority and scrutiny. Apply the same logic to agents. A new agent starts with a probationary delegation — low caps, low threshold, every novelty approved — and earns wider authority through receipt history, exactly as a junior buyer's limits rise with a clean record. Cap changes are delegation changes: proposed with evidence (the receipts showing the bound binding real work), approved at the agreed level, recorded with the reason. This framing prevents the two failure modes finance fears: the shadow delegation (an engineer raises a cap in a config file) and the frozen delegation (caps never reviewed while prices and duties move).

Approvals as countersignature, with sampling

Countersignatures fail when they are universal or reflexive. Finance's interest is that the approval band represents genuine review, so govern it like a control: named approvers per band, a backup, latency expectations, and periodic sampling of granted approvals to confirm the cards contained decidable facts (vendor, amount, task reference) and that decisions show variance — a queue with a 100 percent grant rate at three seconds per item is a control in name only, and sampling is how finance finds out before an auditor does. Denied approvals are control evidence too and should be reported, not hidden: a quarter's report that includes denials reads as a functioning system.

Reconciliation: the control that validates all others

Monthly, reconcile three records: mandate receipts (what was authorised), payment records (what was charged) and the business ledger (what was booked). Authorised-without-charge and charged-without-authorisation are the two exception classes, and the second is the one that matters most — it enumerates spending paths outside the mandate, each requiring disposition: bring under control, envelope it, or accept it explicitly as a documented exception with an owner. Finance teams that run this reconciliation quarterly can state agent spend coverage as a percentage with a straight face. Those that do not are asserting coverage they have not measured. The reconciliation is also where task-reference discipline pays off: references that join to cost centres make the exercise hours, not days.

Budget cycles and cap governance

Bring caps into the planning rhythm. At budget time, agent delegations are forecast like other spend categories: expected volumes per agent from receipt history, cap headroom as a stated risk choice. During the year, the monthly review handles tuning; at year end, the evidence pack supports the audit and the delegation review resets authority consciously — including revoking delegations for retired agents, whose policies should be closed (and agents frozen) as part of offboarding, just as a departing employee's authority ends. Zombie delegations are the agent-era version of dormant vendor accounts: unglamorous, common and exactly what periodic governance exists to catch.

What finance should demand from day one

Four demands, cheap at the start and expensive to retrofit. Every agent spend path passes authorize — no exceptions by tool or team. Task references join to cost centres from the first receipt. The evidence pack is produced at period end as routine, chain verified in the same sitting. And delegation changes are recorded with reasons in finance's own register, mirrored against the live policies at each review. A deployment meeting those four can answer the board's eventual question — how much can our agents spend, who said so, and how do we know what they did — in one meeting, with artefacts. That answer is the entire business case for doing this before the first incident rather than after it.

Five actions to take now

  1. Agree the evidence standard with finance before the first audit: chain verdict plus export
  2. Set task-reference conventions finance can join to cost centres
  3. Schedule the monthly reconciliation: receipts versus statements, gaps actioned
  4. Define approval authority in finance terms: who may approve, up to what band
  5. Review caps as budget lines at each planning cycle, not as IT settings

Worked example: finance takes ownership of agent spend

A finance lead inherits a three-agent deployment with no controls and a quarter of unexplained tool spend. She implements MandateGuard as a finance programme, not an IT ticket: delegations written per agent (caps as authority limits), approval bands assigned to named managers, task references mapped to cost centres, and a monthly reconciliation ritual. The first reconciliation finds 96 percent of agent spend inside the mandate and two ungated paths — a legacy automation and a manual subscription — both brought under control or explicitly accepted.

At quarter end she produces the evidence pack with a verified chain: 864 decisions, approvals sampled clean, denial log demonstrating the controls operated. The external review conversation changes from how do you know to here is the remaining 4 percent and its disposition — which is the conversation finance wanted all along.

Frequently asked questions

Does MandateGuard replace our expense policy?

It enforces the machine-spendable part of it for agents. Human expense policy remains yours; the mandate is its executable form for agent purchases.

Can receipts map to cost centres?

Yes — through task references agreed with finance, receipts join to cost centres, projects or GL codes in your reconciliation.

What do auditors receive?

The evidence pack: receipts for the period with the chain verification, plus your policy context. Export is a Pro and Team feature.

Who should own cap changes?

Treat caps as delegations of authority: proposed by operations, approved at the finance-agreed level, recorded with reasons. The tool permits editing; governance decides who may.

How does this interact with card controls?

Complementarily. Card controls bound a payment instrument; MandateGuard bounds an agent's decisions across any instrument, with reasons and approvals attached.

Keep reading

Put the policy in force: spend envelopes · intent registry · approval inbox · how the proof works · all MandateGuard guides