The three financial statements
Income statement (the month)
Balance sheet (month-end)
Cash flow (the month)
Read directly from the general ledger below: every posting is caused by an event (invoices, receipts, payment runs paying invoices as they fall due, payroll, close journals, a failed fraud control) or is an automatic or exogenous posting (the ERP’s month-end accrual and its reversal, agent provider charges). Effects that are assumed rather than simulated are listed separately. The recommended model adds agent running costs, fewer fraud and duplicate payment losses, and a one-off cash gain from faster collection in the first month. Released finance capacity is redeployed unless you choose to remove the cost below.
Financial assumptions
All figures are placeholders for a single-entity UK business of this size. Agent costs are illustrative and need confirming with your provider.
Journals and general ledger
Every transaction in the simulation posts a balanced journal: customer invoices and supplier invoices from the billing and AP systems, receipts from the bank statement, payment runs, the payroll journal and payments, close journals prepared by the team (or drafted by agents and approved by people), automatic reversals, and month-end accruals. Select a journal to see its lines. Journals posted after month-end are greyed and fall into next month.
Process board
| SOP | Playbook | Items | By rules | By people | Prepared by agents | Waiting for a person | Done |
|---|---|---|---|---|---|---|---|
| Total | 0 | 0 | 0 | 0 | 0 | 0 |
Every weekday arrival weight is zero, so arrivals are spread evenly across the week. Set at least one weekday weight above zero.
| Process | Items | By rules | By the team | Waiting for a person | Done |
|---|---|---|---|---|---|
| Total | 0 | 0 | 0 | 0 | 0 |
Who is doing the work
People: share of their time used so far. Dark is the modelled tasks (hours shown); light is their other work.
Agents: items handled
Playbooks: items prepared under each SOP’s playbook (select a board row or a playbook in the diagram for its content)
A bar turns red at 95% or more, when queues build and deadlines slip. During the close the accountants, Finance Manager and Financial Controller put most of their other work on hold; FP&A and the CFO do the same in reporting week.
The year, month by month
A full year runs twelve months in sequence, starting with the year-end close. Quarter-end work falls in the month after each quarter end, the annual budget in the last month of the year, and P11Ds, the insurance renewal and pension re-enrolment in July. Run the year under each model to compare them.
Task table: every task the simulation runs
Add a task
| Step | Person | Minutes per item today | Minutes per item with agents |
|---|---|---|---|
| 1 | |||
| 2 | |||
| 3 |
Model checks and back-test
Checks that the simulation behaves correctly: every item finishes, hours match the task table, the ledger balances, the same seed repeats, targets do not affect the result, and more people or more work move the result the right way.
Back-test against the actual timetable
Enter the actual close and pack days for the last twelve months, then run today's model for the same months and compare.
Sensitivity: which inputs move the answer
Moves each key input down and up on its own and shows how much the result changes, so you can see which assumptions need the most evidence.
Range of outcomes
Ranges for the inputs we are unsure of
Used by the range of outcomes (when uncertainty about inputs is on) and by the global sensitivity analysis. Set each range to what the evidence supports.
Global sensitivity: what drives the result, including interactions
Calendar and variability
The month picker and the full-year run use real dates: weekends and bank holidays are removed, weekly tasks fall on the chosen weekdays, and payroll, VAT and board work fall in the right months. Switch on real-world variability above to add the randomness below; the seed makes a run repeatable.
Real-world variability
Seasonality and planned leave by month
How agents take workload from each role
Workload, capacity and service: people, agents and systems
How each role spends its time
Finance organisation and organisation chart
An assumed structure for a single-entity UK business of this size: replace the titles, headcount and reporting lines with your own. Other work is the share of each role’s time spent on things the simulation does not model task by task, such as queries, meetings, analysis and projects; it is what makes the utilisation figures realistic.
| Role and job titles | Reports to | FTE | Other work (% of time) | Focuses on the close | Focuses on reporting | Main responsibilities today |
|---|
Today compared with the recommended model
Control points shown in the two models
- Today: FP&A maintains supplier and customer master data, including bank details, and also reviews spend. This is shown because it is how the team works now, not as a modelling error.
- Today: supplier bank-detail changes are verified by the same team that makes them, and fake bank details on invoices rely on the AP clerk noticing.
- Recommended: supplier changes need Financial Controller approval and an independent AP call-back before FP&A keys them; customer set-ups move to the AR team; agents cannot change master data or payments.
Volumes, team capacity and time per task
Every result follows from these figures. They are placeholders for a single-entity UK business; replace them with your own. Changes apply when you restart the month.
Monthly volumes
Recommended model only
Engine settings
Service targets
Systems as resources
Minutes per item
| Task | Today | Recommended |
|---|
About this simulation
How the model works, how to use it, what its measures mean and how far to trust it. The SOPs and the process guide hold the full detail.
What it is for
It runs the same month of finance work twice on identical inputs: as the office works today, and with four agents preparing work for people to review and approve. The difference between the two runs is the effect of the way of working alone, because both use the same volumes, people, calendar, controls and random draws.
How to use the tabs
- Simulation run: choose today or with agents, pick a month, and run it or walk through one case. Click any box for details.
- Parameters: change headcount (click a role in the chart), task minutes and volumes, the calendar, variability and assumptions. Saved scenario files keep all of these.
- Results: run both models on the same month, then compare them: workload, service, statements, ranges of outcomes, sensitivity and the model checks.
- Documentation: this page.
The hierarchy
The organisation assigns roles. Each SOP (FIN-01 to FIN-16) is owned by one of those roles and sets what is done, by whom, when and who approves. Each SOP where agents prepare work has one playbook (PB-01 to PB-16), owned by the SOP owner, telling the agents how to prepare it; a playbook never sets an approver. Every task belongs to exactly one SOP. Processes (P1 to P9) group whole SOPs into end-to-end flows, used to measure service from start to finish, so every task rolls up cleanly: task, SOP, process.
How the engine works
- Every invoice, journal, reconciliation or payment is a separate item, following its task’s route through people, agents and systems.
- Items wait for a free person, agent or system. Close, payroll, payments and anything with a statutory deadline go first; people work in sessions of up to two hours and pick up urgent work between sessions.
- Nothing starts before its inputs exist; nothing posts or pays before the approval the SOP requires.
- Capacity is set each working day from FTE, leave, other work, overtime and outages. A year runs as one continuous timeline, so unfinished work carries on.
- Every posting in the ledger comes from a completed task or a labelled automatic posting.
What the main measures mean
- Close day and pack day: the working day the close is signed off and the management pack issued.
- Standard effort: planned task minutes for the month’s work, before rework and time variation.
- Service requested: effort including rework and time variation; savings use this.
- Delivered effort: hours actually worked within the month.
- Time used: delivered task work plus other work, as a share of the time available.
- Within target: items finished within their process’s service target, counting unfinished items as missed.
- FTE-equivalent: hours divided by a full-time month; capacity, not removable headcount.
Calibration and checks
Today’s model is calibrated with two explicit policies, overtime of 2 hours a day and 5% of other work kept during the close and reporting, so it reproduces the actual day 7 close and day 10 pack. The model checks (49 acceptance gates, in the Results tab) test the engine’s rules directly: capacity is never exceeded, effort is conserved, nothing runs before its inputs, every posting has a cause, deadlines and approvals are enforced, and the hierarchy holds.
Limits
- Task minutes, volumes and other work are assumptions until a two-week time log and 12 months of actual data are used.
- Roles stand in for named people, and controls are modelled as workflow steps, not checks of document contents.
- Agent minutes and acceptance rates are estimates for a pilot to measure.
- Freed capacity is not cash unless roles change; effects such as faster collection are external assumptions.
Related documents
- CFO office standard operating procedures (FIN-01 to FIN-16)
- CFO office process guide, version 25 (processes, systems, agent design, playbooks, simulation description, task inventory)
- What agents change in the CFO office (summary presentation)