Procedures
End-to-end audit procedures, run over every row.
Describe the procedure in plain words. Deixio proposes the tests from the evidence, you shape and approve them, the approved tests run over the whole population, and the workpaper is built from what happened, for you to sign.
End-to-end procedures
From intent to signed workpaper.
Deixio takes each procedure from start to finish, in the steps you see on screen. The approved tests run over every row, every document in scope is read or listed with the reason, and each figure links to where it is printed.
- 01
Intent
Say what the procedure should prove, in your own words, and upload the client’s documents.
- 02
Tests
The AI drafts each test from the evidence: what it proves, its population, its sources and what counts as an exception. You edit, add and remove.
You approve the plan - 03
Run
Only the approved tests run, over every row. Code checks that each figure is printed where the AI points, and does all the arithmetic.
You approve figures below your line - 04
Results
Each test is done, or “not testable” with the reason. Every finding shows its sources, side by side.
You decide each finding - 05
Workpaper
Built by code from the plan, the results and your decisions, with exports to Excel, PDF and an archive.
You sign off, in order
Start next period’s audit from the plan you approved. The AI fits each test to the new period’s data and keeps your words.
Human approval point
Not fixed templates
Your intent, your tests, your decisions.
A procedure starts from what you want to prove, for example “three-way match the Q1 supplier invoices to their orders and goods receipts”. The AI looks through the evidence and drafts each test: what it proves, its population, its sources and what counts as an exception.
You edit, remove and add tests in your own words, and set standing instructions such as tolerances, files and the period. Then you approve. What you wrote stays marked as yours, and you can watch the AI work as it goes.
Built and tested in depth
Five procedure families, built and tested in depth.
Each has its own methodology, tests the AI can propose and sample audits it is tested on. They are examples, not the limit: describe another procedure in your own words and Deixio proposes the tests from the evidence.
Three-way match
Hold every recorded invoice to its order and its receipts.
Line by line, not a sample. Each exception shows the values compared and opens on the invoice line it came from.
- The AI reads the invoice lines and proposes the join keys and the tests.
- Code checks each figure is printed there and does every comparison, in decimal.
- You set the tolerance, approve the plan and decide each exception.
Bank reconciliation
Pair every cash-book line to the bank statement.
Reconciling items are listed with the statement line they come from, and anything left unpaired says why.
- The AI proposes pairing on reference, then on amount within a clearing window.
- Code does the matching and the sums, and checks statement figures are printed there.
- You approve the window, the plan and each reconciling item.
Journal entry testing
Test the whole journal population, not a sample.
Choose the risks; every entry is tested against them, and each flagged entry names its rows and the rule it broke.
- The AI reads your instructions and proposes tests for the rules they state.
- Code runs every filter, count and total over the whole population.
- You choose the risks, approve the plan and decide each flagged entry.
Payroll
Recompute every payslip line from approved hours and rates.
Gross pay recomputed line by line, plus validity tests: roster, contracts, employment dates and paid twice.
- The AI proposes the recompute and the employee validity tests.
- Code does the recompute itself, in decimal. The AI never adds or multiplies.
- You approve the plan and decide each difference.
Trial balance reconciliation
Tie every account to its lead sheet and schedule, both ways.
Lead sheets, schedules, sub-ledgers and roll-forwards, with prior-year closing to current-year opening checked by code.
- The AI proposes the ties, the footings and the roll-forward checks.
- Code does every tie and footing, reconciling both ways.
- You approve the plan and decide each difference.
A procedure we have not listed?
Procedures are not fixed templates. Describe another one in plain words and the AI proposes tests from the evidence and your intent, for you to shape and approve. Start from a plan your firm has approved, or publish your own to the firm library; a second person publishes it.
What changes for the engagement team.
Each figure sourced
Every figure links back to the exact place on the page it was read from.
Recalculation by code
Every sum, comparison and recompute is the engine’s, kept with the table version it read.
Warnings listed
What could not be tested, read or matched is listed with the reason.
A review trail
Every correction, decision and signature records who and when, in an append-only log.
Workpapers and sign-off
Workpapers built from what happened.
The workpaper is composed by code from what was stored, never reworded by a model: objective and scope, the tests as approved (yours marked “Yours”), the answers, results with evidence links, findings, other observations, the conclusion and the sign-off. The system writes an objective summary of the results; the conclusion is the auditor’s.
- Signed in order
- Preparer, then reviewer, then partner. Signing is blocked while any finding is open.
- Independence, if you want it
- An optional policy makes one person sign only one role on a workpaper.
- Exports
- Excel with live links, PDF with highlights and an evidence appendix, and an archive with a SHA-256 manifest.
Watch one procedure run, figure by figure.
In 30 minutes we run a procedure end to end on sample data and show you where every figure comes from.
- A source for every figure
- Human in the loop
- Append-only, hash-chained audit trail