The pattern of the accounting firm in 2026 — what's actually broken in the day
I've sat in enough partner meetings over the last two years to draw the shape of it. The firm has more work than it can staff. Trainees won't stay past their exams. The senior managers spend half their week answering questions that were already answered in an email three months ago, sitting in someone else's Outlook. The partner who knows how a particular client's group structure actually works is the partner who keeps getting pulled into it, because the knowledge never left their head.
That's the bit AI helps with. Not the audit opinion. Not the judgement call on a deferred tax position. Not the conversation with a Revenue inspector. Those are yours and they should stay yours. What AI helps with is the part of the day where you're searching, summarising, drafting, reconciling, and explaining the same thing for the fourth time. If a junior is spending Tuesday afternoon reading three years of correspondence to figure out what was agreed on a directors' loan, that's where the time goes. That's where a properly built accountant AI earns its keep.
Where it doesn't help: anywhere the answer needs a signature, a professional view, or a regulator-facing assertion. The Intelligence Brain is a research and drafting layer underneath your people. It is not a replacement for them and I won't sell it as one.
The seven workflows that pay for the project in month one
These are the ones I see firms recoup the build cost on inside the first cycle. Pick any three and you'll feel it.
- Client intake and KYC packing. The brain reads the engagement letter, the prior-year file, the CRO filings, and the AML documentation, and produces the intake pack a manager would otherwise build by hand. You review and sign — you don't assemble.
- Year-end file pre-population. Trial balance lands, prior-year working papers are already indexed, and the brain drafts the lead schedules, flags the movements above your materiality threshold, and writes a first-pass commentary. The senior starts at 60%, not zero.
- Corporation tax computation drafting. Disallowables, capital allowances, R&D claim narrative, group relief working — drafted from the file, with every figure citing the source document. A tax AI in Ireland is only useful if it shows its workings, so it does.
- VAT return review and anomaly flagging. The brain reads the bookkeeping ledger and surfaces the entries that don't match the pattern of the last twelve returns. Reverse charge errors, mis-coded EU acquisitions, the property transactions that shouldn't be on a standard return.
- Client correspondence drafting. A partner dictates two lines; the brain produces the email in the firm's voice, with the right attachments referenced, and the right caveats. Nothing sends without the partner clicking send.
- Practice management AI for WIP and recovery. The brain reads time entries, billing narratives, and engagement scope, and tells you on a Monday morning which jobs are overrunning and why, in plain English. Not a dashboard you have to interpret — a paragraph you can act on.
- Technical research with citation. Ask it about Section 110, about the close company surcharge, about a CGT relief, and it answers from the legislation, Revenue eBriefs, and your own internal precedent library — with the citations attached. No hallucinated case names.
The data-residency posture — what an Irish accounting firm actually needs
This is the part I won't compromise on, and neither should you. An Irish accounting firm holds personal financial data, beneficial ownership data, payroll records, and in many cases medical or family information that arrived sideways through estate planning. GDPR isn't theoretical here.
The Intelligence Brain is deployed on-premise or in a tenancy you control. Client data does not leave your environment to be processed by a third-party model. That matters for three reasons:
- GDPR Article 28 and your processor obligations. If you're sending client data to a US-hosted general-purpose model, you need a defensible answer for your data protection notice. On-prem removes the question.
- Chartered Accountants Ireland and ACCA confidentiality rules. Both bodies require client confidentiality as a condition of practice. Sending working papers into a public model is a conversation you don't want to have at your next practice review.
- Revenue and the Central Bank, where relevant. If you act for regulated entities, your audit trail needs to show where data was processed and by what. An on-prem brain produces that trail by default.
The accounting intelligence brain runs against your document store, your practice management system, and your tax software. It doesn't phone home.
The deployment cadence — what you see at each gate
Thirty-two weeks, four gates of eight weeks each. I've found shorter doesn't stick and longer loses momentum.
Weeks 1–8: Ingest
I connect to your document management system, your email archive, your practice management system, your tax software exports, and whatever shared drives still hold the institutional memory. By the end of week eight you can ask a question across every client file the firm has ever held and get a sourced answer. Nothing is being drafted yet — this gate is about retrieval.
Weeks 9–16: Structure
Now we shape the retrieval into the firm's actual taxonomy. Engagements, entities, periods, working paper types, correspondence threads. At the end of this gate, a manager can pull a year-end pack template populated from prior years in under a minute.
Weeks 17–24: Swarm
This is where the seven workflows above go live, one or two at a time, with a named partner sponsor on each. I don't switch them all on at once. We pick the two that hurt most and prove them on live files, with review.
Weeks 25–32: Audit
The final gate is about defensibility. Every output the brain has produced is reviewable, every source is cited, every prompt is logged, and your data protection officer has a report they can hand to a regulator without flinching. This is the gate that lets you tell a client, on the record, how their data was handled.
What to have ready for the assessment call
If you book a call, three things make it useful rather than exploratory:
- A rough headcount and engagement count. Partners, managers, seniors, trainees, and the number of active engagements. I'm not going to quote on a vibe — the shape of the firm changes the build.
- The names of your core systems. Practice management, document management, tax software, payroll. I need to know what I'm connecting to.
- The two workflows that h
Frequently asked questions — Accounting
Is the Intelligence Brain on-premise or cloud?
Default is on-premise — the firm's own server, the firm's own data, the firm's own model weights. We support private-cloud (your AWS, your GCP, your Azure tenant) when on-prem hardware isn't a fit. We do not run a multi-tenant SaaS.
How long is the rollout?
About six months from kick-off to live use. Four eight-week stages — ingest, structure, swarm, audit. The swarm runs in shadow mode for the first ninety days alongside your team; only at day ninety, with the audit logs to back it up, does the swarm earn the right to run a tool live.
What does it cost?
Per-firm engagement, scoped from a free thirty-minute assessment. Firms vary too widely for a public list price — a five-partner law firm and a forty-person SME need different scoping. Book a slot via Calendly and we will scope it together.
Can it write contracts / draft accounts / produce clinical letters automatically?
It can produce a first pass that a qualified human reviews before anything is signed, filed, or sent. Tool-layer authorisation is a hard architectural boundary in the brain — the swarm reads everything and signs nothing.
What about hallucination?
The auditor agent's job is to catch hallucination before output reaches a human. Every claim in every output is required to be cited; every cite has to be reachable; every cite has to load. If the auditor cannot verify, the output is rejected as a build-failure signal — not corrected.
What's specific about accounting firms in the rollout?
The accounting vertical brings its own data-residency, professional-body, and audit-trail constraints. The methodology is the same; the structure-stage and swarm-stage prompts are vertical-specific.
Do you understand the accounting regulatory environment in Ireland?
I have worked with firms in this vertical and I bring the regulatory posture into the architecture from day one. The compliance pack at delivery includes DPIA, LIA, and EU AI Act tier-mapping, all reviewed against the vertical's specific framework.