Financial controllers do two jobs once agents arrive: they approve the agent's held journals and reconciliations, and they set the limits the agent works within. Both need tested understanding before go-live. A controller who sets a materiality threshold without knowing how the decision chain uses it is configuring a control they cannot explain to the auditor.
What does an agent do in the close?
The month-end close is a sequence of repeatable tasks with judgement at the edges. Reconcile bank accounts and balance sheet control accounts. Post accruals and prepayments. Match intercompany balances. Investigate variances. Prepare the management pack. Sign off.
A governed agent can carry much of the repeatable part. At depth 1 it reads the general ledger, sub-ledgers and bank feeds. At depth 2 it drafts reconciliations, accrual journals and variance commentary. At depth 3, once Attest has proven it on the business's own past closes, it may post routine journals that sit inside rules the controller sets, such as recurring accruals below a threshold or reconciling items that match exactly. Anything above the threshold, any new type of journal, and every manual adjustment to a control account is held for a named person.
The controller then signs off a close that the agent largely prepared. That is a very different sign-off from one where their own team did every step.
Why is the controller part of the control, not just a reviewer of it?
In a governed system, every action an agent takes must clear the minimum chain: role scope, its certificate, its sources, the customer's own limits, and the platform guardrails. The lowest bound wins.
The customer's own limits are, in a finance function, largely the controller's decisions. What journal value may the agent post without approval? Which accounts may it never touch? Which reconciling items count as routine? When the controller sets those, they are writing the bound that will decide many of the agent's verdicts.
That makes the controller's understanding of the system part of the control environment. If they set a threshold too high because they did not realise it would let the agent post accruals across entities, the agent will do it, correctly within its bounds, and the error will be the controller's. Certification for controllers therefore has to cover setting limits as well as reviewing held items.
What should a controller be able to show?
A certificate for a financial controller is worth having if it demonstrates:
- Reading the chain. For any held or posted journal, they can say which bound decided the verdict.
- Setting limits deliberately. They can explain each threshold they set, what it allows, and what it holds.
- Reviewing at the right depth. They sample posted depth 3 journals, not only the held ones, because routine work is where silent errors accumulate.
- Reading provenance. A bank balance from the feed is sealed. An accrual estimate from a budget holder is declared. The controller weighs them differently.
- Knowing the floor. Some actions are withheld for every agent at the platform level. The controller does not need to design around those; they need to know they exist.
The practicum log template for financial controllers captures these decisions in a way an auditor can read.
Does the controller need a certificate to approve agent work?
The business decides, but the case is strong. External auditors will test the close, and increasingly they will ask how journals posted by an agent were authorised. The answer "within limits set by the controller, who holds current certification" is far stronger than "within limits someone configured at launch". The detailed argument is in whether financial controllers need a certificate to approve agent actions.
A business can adopt the practice of tying the right to approve held journals to a current certificate, and tying the right to change the agent's limits to a higher one. That separates the person who reviews from the person who configures, which also keeps segregation of duties intact.
What is the trade-off?
Controllers are senior and stretched, especially near close. Asking them to sit an exam and complete a practicum before go-live will meet resistance, and some will argue that their professional qualification already covers judgement.
It does cover accounting judgement. It does not cover how a governed agent reaches a draft, what its verdicts mean or how the limits they set will behave. Those are new, and seniority does not supply them.
The practical route is to time it. Run the agent at depth 1 and 2 for a close or two, so the controller reviews every draft and learns the system on real work. Use that as the practicum. Raise the agent to depth 3 on routine journals only after the controller is certified and has set the limits deliberately.
The accountability point is decisive for most finance leaders. If a misstatement traces back to a journal the agent posted inside limits the controller set, the question of whether financial controllers are accountable for an agent's mistakes will be put to them directly.
How does this sit alongside other finance roles?
The controller is the last check in a chain that starts with operational finance. Credit controllers propose write-offs that the controller provisions for. In lending businesses, loan officers originate the assets the controller reports. Certifying each role means the agent's work is checked by tested judgement at every stage, rather than relying on the controller to catch everything at the end.
What should a finance director do differently?
Certify controllers before raising any finance agent to depth 3. Separate the right to approve from the right to set limits. Ask controllers to document the reason for every threshold. Keep certificates and practicum logs in the audit file. The guide to AI literacy and certification for people who work with agents explains which level fits which job.
Where the Academy fits
Controllers who review held journals take Pryme Intelligence Academy's Operator track, which ends in a written exam and a two-week practicum of real review decisions signed off by a Pryme assessor. Controllers who also design or adjust the agent's brief and limits can go on to the Builder track, whose certificate needs a proctored exam and a capstone agent certified in Attest at depth 2 or higher.
Questions people ask
Should an agent ever post a journal to a control account?
Most businesses hold every adjustment to a control account for a person, whatever the agent's depth elsewhere. Routine reconciling entries may be allowed within limits, but the controller should decide that deliberately.
Can a controller delegate review of agent journals to their team?
Yes, to certified team members, with the controller sampling their decisions. The controller remains responsible for the limits and the sign-off.
What happens at close if the agent's certificate lapses?
Its depth drops to 1 on the affected capabilities and the work arrives as held drafts. Plan for it by knowing renewal dates and keeping certified reviewers available near close.