Month-end is where good firms lose evenings and bad firms lose accuracy. The mechanics haven't changed much in twenty years — accruals, prepayments, intercompany, FX, variance commentary, sign-off — but the volume of source data has, and the patience of finance directors hasn't. An intelligence brain doesn't replace the close. It removes the parts of the close that should never have been manual in the first place: chasing missing invoices, reconciling supplier statements, drafting variance commentary, and answering the same five questions from the same three people every month.
This is how I run a month-end workflow with the Brain on-prem, what it actually does, and where it stops. No magic. Just plumbing done properly.
What the Brain is, in a finance context
The Michael English Intelligence Brain is an on-premise organisational intelligence layer. For an accounting team that means: it ingests your ledger exports, your bank feeds, your supplier inboxes, your contract folder, and your prior-period working papers, and it builds a private index that the close team can query in natural language. It runs inside your environment — your tenancy, your VPC, or a box on your premises — and the data does not leave.
That last point matters more in Irish accounting practice than people admit. Between client confidentiality, GDPR, and the practical reality that most firms are now handling data for entities across multiple jurisdictions, sending ledger detail to a public model is a problem you don't want. On-prem removes the question.
The Brain is not a bookkeeping engine and it is not a replacement for Sage, Xero, BrightBooks, or whatever you post journals into. It sits beside them and reads.
The shape of a Brain-assisted close
A traditional Irish accounts close — whether for a SME client or an internal management accounts pack — has roughly the same skeleton every month. Cut-off, accruals, prepayments, payroll journal, depreciation, intercompany, FX, bank rec, debtor and creditor review, variance analysis, commentary, pack production, sign-off. The Brain doesn't change the skeleton. It changes which steps need a human keystroke.
Here is the workflow I actually use:
- Day -3 to -1: Brain ingests the latest GL export, bank statements, AP inbox, and any contract changes. It pre-flags the cut-off list — purchase orders received but not invoiced, recurring contracts with no posting this period, prepayments approaching their amortisation point.
- Day 0 (period close): Brain produces a draft accruals schedule and a draft prepayments schedule by reading contracts, prior journals, and supplier patterns. The accountant reviews, adjusts, posts.
- Day 1: Bank rec and intercompany reconciliation. The Brain matches and surfaces the unmatched. Humans deal with the unmatched.
- Day 2: Variance analysis and commentary. The Brain drafts the "why is heat & light up 18% this month" paragraph by reading the GL detail and the supplier invoices. The accountant edits.
- Day 3: Pack production, partner review, sign-off.
The compression isn't dramatic for one client. It's dramatic when you're running fifteen of them.
How the accruals and prepayments inference actually works
This is the part most people want to understand and most vendors hand-wave through. Let me be specific.
The Brain holds three things in its index: your contracts (rent, insurance, software subscriptions, service agreements), your historical postings (twelve to twenty-four months of GL detail), and your AP activity (invoices received, POs raised, supplier statements). When the close period ticks over, it runs a recurrence check across each of those three sources.
If a contract says "monthly licence fee, €X, billed in arrears" and there's no invoice in the period, that's a candidate accrual. If a supplier has posted invoices on or around the same day for the last eleven months and this month is silent, that's a candidate accrual. If an annual insurance premium was paid in March and we're closing September, the unexpired portion is a candidate prepayment with a calculable balance.
None of these are posted automatically. They land in a draft schedule with the supporting evidence — the contract clause, the prior twelve postings, the invoice PDF — linked beside each line. The accountant approves, edits, or rejects. The journals are posted by the human into the ledger system the way they always were.
The reason this matters: most accruals errors at month-end are not errors of judgement. They are errors of memory. Someone forgot the quarterly maintenance contract. The Brain doesn't forget.
Variance commentary that isn't generic
Management accounts AI gets a bad name because most of it produces commentary that reads like a horoscope. "Revenue increased due to higher sales activity." Useless. The reason it's useless is that the model doesn't have access to the underlying transactions — it's been given the summary numbers and asked to explain them.
The Brain does it the other way around. It reads the GL detail behind each variance, identifies which transactions are driving the movement, and writes commentary at that level. So instead of "professional fees up €4,200" you get "professional fees up €4,200, primarily driven by the legal invoice from [firm] dated [date] relating to the [matter] — non-recurring, no equivalent in prior period." That's commentary a finance director can actually use.
The technical piece that makes this work is the link between the trial balance variance and the underlying journal lines and source documents. The Brain maintains those links in its index so when it explains a variance it can cite the specific invoice or journal that caused it. If you want a deeper look at how this is wired into a practice's existing systems, the accounting vertical page has the integration detail.
Bank rec, supplier statements, and the boring 40%
If you actually time a month-end honestly, somewhere around forty percent of the elapsed hours go on three tasks: bank reconciliation, supplier statement reconciliation, and chasing missing documents. None of these are intellectually interesting and all of them are necessary.
The Brain handles the matching layer. Bank statement lines match to ledger postings on amount, date, and reference fuzziness. Supplier statements match to AP postings the same way. What's left after the match is the work — the genuine differences, the timing items, the disputes. That's where a human needs to be. The matching itself is mechanical and the Brain does it at machine speed.
The chasing is more interesting. The Brain knows which suppliers normally invoice by which date, which approvers are sitting on which POs, and which clients owe which documents. It can draft the chaser emails — to suppliers for missing invoices, to internal staff for missing approvals, to clients for missing bank statements — in a tone you've configured, with the specific reference numbers and amounts. You review and send. The cycle of "I'll just pop them an email" that consumes two hours every period becomes fifteen minutes of review.
Where it stops, and where the auditor sits
Three things the Brain does not do, and should not do.
It does not post journals automatically. Every journal is human-approved and human-posted. This is not a technical limitation — it's a deliberate design choice. Audit trail and professional responsibility require it.
It does not make judgement calls on impairment, going concern, or recognition. It can surface the indicators — a debtor ageing past a threshold, a contract with unusual termination clauses, a customer with declining purchase patterns — but the call is the accountant's and the partner's.
It does not replace the working papers. It adds to them. Every draft schedule, every variance explanation, every reconciliation export is captured with the underlying evidence and stored in the period's working paper folder. When the auditor arrives, they can see what the Brain proposed, what the human accepted or changed, and why. That trail is more complete than most manual closes produce, which is a side benefit nobody anticipated when I built the first version.
For the broader picture of how this layer fits across other regulated workflows, the overview page covers the architecture and deployment options.
Where to start this week
If you run a finance function or a small practice and you want to see whether this is real, don't try to deploy a month-end workflow on your next close. Start with one boring task. Pick supplier statement reconciliation for your three largest creditors and run it through the Brain in parallel with your normal process for one period. Compare what it found against what your team found. If it's useful, expand to accruals next period. If it's not, you've lost a few hours and learned something. That's the only honest way to evaluate a tool that claims to change a process you've been doing the same way for twenty years.