What work in an accounting practice is worth automating?
The work that is rule-bound, high volume, and always late.
Every practice has a set of jobs that are not difficult and are never done on time, because they lose to whatever is billable that week. Reconciliations, the sift of supplier invoices, the AML file for a client who started three weeks ago, the aged debtors listing nobody has opened since the last one. None of it needs judgement most of the time. All of it needs judgement some of the time, which is exactly why it cannot simply be handed to a junior and forgotten.
These are the five we have written up in detail. Each page says what automates, what does not, and what to measure.
| Workflow page | What it answers |
|---|---|
| Invoice processing | Can AI read and post our supplier invoices, and where does the exception queue sit? |
| Debtor follow-up | Can chasing be automated without it sounding like a robot to your best client? |
| Bank reconciliation | Can the feed be matched against the ledger, and will it catch what you would have caught? |
| Management accounts | How much of the month-end close and the pack can actually be automated? |
| Client onboarding | Can the AML file be assembled automatically, and who carries the risk if it is? |
Two of those have a date attached to them, which is unusual in this trade.
Revenue has published its implementation path for VAT modernisation. Phase 1 lands on 1 November 2028, when VAT-registered large corporates must send eInvoices to Irish business customers and report a subset of the invoice data to Revenue. Most practices are not in that phase and most of their clients never will be. The line that catches everybody is a different one: from that same date, all businesses in Ireland are required to be able to receive structured eInvoices. Revenue puts it plainly, that even if your own business is not yet required to issue eInvoices under the phased rollout, "you will need to have the capability to receive them in the required structured electronic format", and that issuing PDF invoices or scanned paper invoices will not satisfy the requirements. A practice that handles client invoices on PDFs today is going to rebuild that handling anyway. Doing it deliberately is cheaper than doing it in 2028. The detail, phase by phase, is on the Ireland e-invoicing page.
The second is the money already sitting in your ledger. The European Commission's EU Payment Observatory reports that in 2024 average Irish payment periods rose to 61 days for business-to-business transactions, a rise of nine days on the previous year, against average agreed terms of 42 days. That figure comes from the Intrum survey, and the Observatory is careful to say the 2024 Irish evidence is "somewhat conflicting", because a second source in the same chapter has on-time payment rising to 58%. Treat it as an order of magnitude rather than a national fact. The number that matters is your own, and you already produce it.
How do you decide which workflow to start with?
Three questions, asked in this order, and the third one decides more often than the first two.
What does this workflow cost you today, in a number you already have? Not an estimate of hours. Days sales outstanding. The date the pack actually went out versus the date it was due. The count of invoices that went round twice. If a workflow cannot be described in a figure the practice already produces, it is the wrong place to start, because there will be nothing to measure against afterwards.
What is genuinely repeatable in it? Ninety identical decisions and ten hard ones is a good candidate. Ten decisions that are all slightly different is not, whatever the software claims.
What happens if it is wrong? A misposted invoice is a correction. A wrongly assembled AML file is a regulatory matter, and a reminder sent to a client who paid last week is a relationship. Risk does not rule a workflow out, and it does set how much of it a person keeps. Debtor follow-up and invoice processing usually come first because they are frequent, measurable and reversible. Client onboarding is the one that needs the most conversation before anything is built.
We say which one out loud, and why, before the build starts. If the answer is that nothing here is worth automating yet, that is a real answer and it costs you a conversation to get it.
What does a pilot look like and how long does it take?
One workflow, one to four weeks, five steps in this order.
Understand. A session on the workflow itself: how it actually runs, not how the manual says it runs, and where the judgement sits.
Prioritise. The build starts on the best combination of value, feasibility and controlled risk. A clear business case is the condition for starting, and the rule is applied out loud rather than in a proposal.
Implement. The build fits the systems and the responsibilities already in place, so the practice keeps working the way it works. We do not ask you to change accounting package.
Embed. People are trained, ownership is documented, exception handling is defined and written down. Where the work touches an obligation you carry, the documentation is the record you would want to hand to a regulator.
Measure. The result is compared against the baseline agreed before anything was built.
Most builds spend a fortnight in draft-only mode first, where everything is generated and nothing is sent or posted, so a partner can read a fortnight of output before it touches a client. That step is not optional and it is where most of the corrections happen.
Then we hand it over and leave. You own what was built, there is no retainer, and changing how it behaves is a configuration you make rather than a call you book.
What stays with a person?
The judgement, the sign-off, and every decision a regulator or a client would ask you about.
Exceptions. A machine that posts unsupervised is not the offer. Anything that fails a check, crosses an approval threshold, or sets a precedent for how a supplier or a code is treated goes to a queue with a person's name on it.
Anything with your name on the output. Management accounts commentary is a case in point. A system can tell you a number moved and by how much. Why it moved is a conversation with the client, and the finance lead reads the pack before it circulates.
The AML risk rating and the decision to take the client on. This is the sharpest line in the practice and we state it plainly. We assemble the file, chase what is missing, extract what arrives and keep the record set in order. We do not sell watchlist or screening data, that comes from a licensed provider, and we do not make the acceptance call. The Anti-Money Laundering Compliance Unit's guidelines put the retention duty on the designated person: under Section 55(4) of the Act, records relating to the history of services and transactions carried out "shall be retained for a period of not less than 5 years". Automating the assembly of a file does not move that duty, and any consultant who implies otherwise is describing a product they have not read the rules for.
The relationship. Which clients get chased by a system and which get a phone call from a named person is a decision the practice makes before anything is switched on, and it is not ours to make.
What does it cost to find out?
The first conversation is a conversation, not an engagement.
It runs about an hour and it starts with your own numbers rather than with software: the aged debtors listing, the date the last three management packs actually went out, or the count of clients onboarded in the last quarter and how long each took from engagement letter to complete file. At the end of it you get a straight answer on which of the five is worth doing first for your practice, what it would take, and whether it is worth doing at all.
Build cost is quoted per workflow after that conversation, against a scope you have read. There is no retainer, no licence and no per-seat fee, because you own what is built. If your desk is recruitment rather than accounting, the same argument runs over the recruitment side, and the people behind both are on the about page.