Latest episode
AI in financial close for the reconciliations and mappings, signed by a person.
An agent proposes the matches and the mappings, the platform records what it used, and your group financial controller signs the period. We check the close can carry that before we fix the price.
AI in the financial close does the matching and the mapping, and a person keeps the signature.
AI in the financial close means a model proposing the reconciliation matches, the ledger mappings and the first draft of the flux commentary. A named person signs each result before it reaches a period the auditor will test. Constancia is an EPM consultancy, and we put those models on the close of a multi-entity group whose board wants AI by a date.
A model matching intercompany balances reads the 66 counterparty pairs that 12 entities make, and each pair keeps the tolerance the group financial controller set. The model clears the pairs inside tolerance and a person signs the rest.
Where an acquired entity brings 800 accounts of its own, an analyst maps them to the group chart by hand, in a workbook. The model proposes the same mapping in 1 pass, and the analyst corrects and signs the accounts it placed wrongly.
5 stages put a model on the close, and your controller signs the parallel run before it goes live.
Awaiting image
- 01
Survey
We read the close as your team runs it and list every reconciliation, mapping and commentary draft a model could take on.
- 02
Price
We fix the price on that list, and a ledger whose balances a model cannot trace is priced as data work first.
- 03
Design
The design names the rule each agent follows, the tolerance it works within, the record it writes and the person who signs its output.
- 04
Build
The consultants from the survey switch on the matching, the mapping and the commentary agents on the platform, 1 agent at a time.
- 05
Parallel run
Each agent's proposals run beside the manual work for a period, and your group financial controller signs the agent in when the 2 agree.
A person signs at every stage, and the platform records which balances the model read to reach each proposal.
A model works at 3 places in the close, the mapping, the match and the record, and a named controller signs each.
01
The mapping puts a new ledger's accounts onto the group chart, and an accountant signs the proposal.
The model reads the acquired entity's chart, proposes a group account for each line from earlier mappings and marks the lines it is unsure of. The entity accountant corrects those lines and signs the mapping, and the platform stores who signed and when.
- A proposed mapping, with the earlier mapping it was learnt from beside each line
- A confidence flag per line, and the flagged lines go to a person first
02
The match pairs balances across entities, and the controller signs the exceptions it could not pair.
The model matches intercompany and bank items on the rules and tolerances the group set, and it clears the pairs that agree. The group financial controller reviews the unmatched items and signs the period once each has a reason.
- Matching on counterparty, amount, currency and date, inside the platform
- Every cleared pair kept with the rule that cleared it
- An exception queue with 1 owner, and nothing cleared by the model alone
03
The record shows which balances the model read, so the auditor can test the proposal.
Every proposal the model makes carries the balances it read, the rule it applied and the version of the model that ran. The auditor reads that record in the platform, and the person who signed stands behind it.
- Inputs, rule and model version stored against each proposal
- A signed proposal that can be rerun and gives the same answer
- A period lock, so a change after sign-off is a new dated entry
- 1 named signatory for each agent's output, per period
4 conditions hold before we switch a model on in a close.
- 01
A close that already ties out
The reconciliation ladder agrees at every rung before a model reads it, because a model on a close nobody trusts proposes matches nobody can sign.
Read about close automation - 02
Ledgers a model can read
Each ERP releases its balances at account level, with a date, a currency and an entity on every line.
Read about the assessment - 03
A board with a date on its AI ask
The board has asked for AI in finance and set a date, and the close is where the first checkable result can come from.
Read about agents - 04
A person who will sign
The group financial controller has agreed to sign a model's proposals by name, on the same terms as a junior's work.
General AI advisory outside the finance data estate, and a single entity closing 1 ledger, sit outside this work.
The model reads the platform the close runs on, and the platform decides which agent you get.
OneStream (EPM software)
SensibleAI agents reading the consolidated close
- For a group that closes and consolidates on OneStream
- Matching, mapping and commentary on the balances the consolidation already holds
- The record of each proposal kept in the same platform the auditor reads
Abacum (FP&A software)
Abacum Intelligence on the actuals once the period locks
- For a group whose close is settled and whose commentary is written after it
- The first draft of the variance commentary from the closed actuals, edited by an analyst
- Abacum does not run the close, and the matching stays in the ERP or the EPM platform
Constancia is certified on both platforms, and on either 1 a person signs before a number reaches the board.
5 questions about a model in the close, answered from the platform up.
How is AI currently used in finance?
Finance teams use AI today on the repeating tasks with a checkable answer: matching reconciliations, mapping ledgers, drafting commentary and forecasting from drivers. Each use survives where the model can show which balances it read and a person signs the result.
Will AI take over the financial sector?
The Companies Act 2006 still makes the directors sign the accounts, and a model cannot hold that duty. A model takes the routine work beneath the signature: the matching, the mapping and the first draft of the commentary.
How to automate month-end closing process?
Automate the month end 1 rung at a time, starting where the ledgers already agree. Schedule the loads first, put the reconciliations on rules, then let a model propose the matches and mappings. A person signs each rung before the next 1 is automated.
Can ChatGPT do my bookkeeping?
A general chatbot can draft a journal from a description, and it cannot read your ledger, apply your tolerances or record what it used. Bookkeeping in a group needs a model inside the platform that holds the balances, with a person signing its output.
Can ChatGPT do financial analysis?
ChatGPT can analyse a spreadsheet you paste into it, and the analysis is as good as that 1 file. Financial analysis in a group reads the consolidated actuals with their lineage, so the model has to sit on the platform that holds them.
See more on our
Subscribe on YouTube
Every Sunday, 6am
The Sunday Close.
Our weekly newsletter on AI in finance.
Yours to enjoy with your Sunday morning coffee.
- AI in finance observations from Alex (CEO) and Brad (COO)
- Newest video from our YouTube channel
- AI in finance trends we have spotted each week
Tell us the date you are working to.
The people who would do the work read every message and reply inside 2 working days.