27 August 2026 · 6 min read
TL;DR
Most of the admin load in a small company is not one big job. It is a few dozen small ones that recur, and the largest single category is the same information being moved by hand between two systems that don't talk to each other. Count the hours before you buy anything, start with whatever recurs most often, and keep a named person approving the output.
The work that never appears on anyone's job description
Ask an owner what their team does and you get the real work: the installs, the clients, the orders out the door. Ask the same team what ate their week and you get something else entirely — the re-typing, the chasing, the reconciling, the version of the report that had to be rebuilt because the export changed.
That second list is the interesting one, and it is almost never written down. It has no owner, so it lands on whoever is nearest. It has no budget, so it never gets questioned. And because each individual piece takes eleven minutes, nobody ever adds it up.
The ranges below come from discovery conversations with Irish SMEs, not a survey. Treat them as a prompt to count your own, not as a benchmark. The point is the shape, not the number.
Where the hours go
Money in and money out
12 – 30 hrs / month- Invoices typed from a PDF into the accounts package
- Purchase orders matched to deliveries by eye
- The same three customers chased for payment, again
- Expense receipts coded at the end of the month in one long sitting
- What a machine can take
- Reading the document, pulling out supplier, date, lines and totals, matching it against the order, coding it and queuing it for payment with anything odd flagged.
- What stays with a person
- Approving the batch before payment. Deciding what happens with the ones the machine flagged.
Enquiries and quotes
8 – 20 hrs / month- Enquiry details re-typed from an inbox into a spreadsheet or CRM
- The same twelve questions answered by hand in slightly different words
- Quotes rebuilt from an old one because nobody can find the template
- Follow-ups that never happen because the week got busy
- What a machine can take
- Capturing the enquiry once, qualifying it against your own criteria, drafting the reply and the quote from your real pricing, and scheduling the follow-up.
- What stays with a person
- Reading the draft before it goes, and setting the price where judgement is involved.
Reporting and the Monday morning pack
6 – 16 hrs / month- A report rebuilt every week from the same four exports
- Figures copied between a system and a spreadsheet, then checked twice
- A board or funder update written from scratch each time
- What a machine can take
- Pulling the exports, reconciling them, writing the commentary in your house style and showing its workings so the numbers can be traced.
- What stays with a person
- Reading the commentary and owning the version that gets circulated.
People and paperwork
4 – 12 hrs / month- New starter packs assembled by hand from last time's folder
- Training records kept in someone's head or a stale spreadsheet
- Policy documents that exist in four versions with no clear latest
- Certificates and renewals noticed only when they lapse
- What a machine can take
- Assembling the pack, keeping one record of who did what and when, and raising renewals before they expire.
- What stays with a person
- Signing the record. Deciding what the policy says in the first place.
Suppliers, scheduling and the small stuff
6 – 14 hrs / month- Diary tetris across three calendars to find one hour
- Job sheets re-keyed from a WhatsApp message
- The same weekly order placed by email
- Notes from a site visit typed up that evening
- What a machine can take
- Turning voice notes and messages into structured records, proposing the slot, drafting the order, and keeping one thread per job.
- What stays with a person
- Confirming the slot and the order before it commits money or a promise.
The biggest single category of admin in a small company is one person acting as the integration between two systems.
Frequency beats size
Owners instinctively point at the biggest job. It is usually the wrong place to start. A fifteen-minute task done every weekday costs about sixty hours a year. A two-hour task done monthly costs twenty-four. The small, boring, daily one is both the more expensive problem and the easier thing to hand over, because it is repetitive enough to be described precisely.
How to pick the first one
- Write down every job anyone does more than twice a week, with rough minutes beside it. One week of noting is enough.
- Circle the ones that produce a record — an invoice, a quote, a report, a log. Those are the ones a machine can be checked on.
- Name who would approve the output. No name, no workflow. This is the step that fails most projects, and it fails them after the build rather than before.
- Cost it: hours a week × loaded hourly cost × 46 weeks. Compare that against the build, net of any LEO or Enterprise Ireland support.
- Build one. Not four. One workflow live and trusted teaches you more about the next three than any amount of planning.
What this is not
It is not a plan to remove people. The work being handed over is largely work nobody had time to do properly — the follow-up that never went out, the reconciliation done at speed on a Friday. And it is not a plan to remove judgement: every workflow keeps a named person approving before anything leaves the business. Outputs are decision-support until a human signs them.
If you can name one recurring job and roughly how long it takes, you already have enough to have a useful conversation about whether it is worth automating — including the answer where it isn't.
Questions owners actually ask
- How do I find out how many hours admin is really taking?
- Pick one week and have each person note the recurring jobs they do more than twice, with rough minutes beside each. You are not after precision, you are after the shape. Most owners are surprised less by the total than by how much of it is the same information moved between two systems that do not talk.
- Which admin task should I automate first?
- The one that recurs most often, produces a record you can check, and has a named person willing to approve the output. Frequency beats size: a fifteen-minute job done daily costs more in a year than a two-hour job done monthly, and it is usually far simpler to hand over.
- Will automating admin mean letting staff go?
- That is not how it plays out in a 10 to 250 person company. The work being handed over is usually work nobody was doing properly for lack of time — chasing, reconciling, following up. The realistic outcome is the same team getting to the work they were hired for, with a machine doing the reading and drafting first.
- Does someone still have to check the output?
- Yes, and that is deliberate. Every workflow Artellis builds keeps a named person approving before anything leaves the business or hits a ledger. Budget 15 to 45 minutes a week for it. A workflow that removes the human gate is not saving you time, it is moving your risk somewhere you cannot see it.
- What if the numbers say don't automate?
- Then that is the answer and it costs you nothing to have found it. If the annual cost of a task is smaller than the cost of building the workflow, leave it alone. A discovery conversation that ends in a don't-build is a good outcome, and it happens often enough to be worth saying out loud.
Bring one recurring job and a rough hour count
Twenty minutes is enough to work out whether it is worth automating. You'll get a straight answer either way, including a don't-build where the arithmetic says so.