Ask most business owners in the UK what’s holding them back and you’ll hear about margins, hiring, or the cost of borrowing. Ask them what they actually did last Tuesday and you’ll get a very different answer: chasing an invoice, re-typing a customer’s details into a second system, forwarding a quote request to someone who was on annual leave, and copying figures out of a spreadsheet into a report nobody read.
That’s the admin tax. Nobody sends you a bill for it. It just quietly eats the week.
The shape of the problem
There’s a particular kind of work that shows up in almost every small and mid-sized business we speak to. It has a signature:
- It’s repetitive but not quite identical each time
- It moves information from one place to another
- It’s low-judgement, but it can’t be skipped
- And it always lands on the person you’d least like to be doing it
A practice manager at a dental surgery in Leeds spends her mornings turning voicemails into diary entries. A quantity surveyor in Bristol rebuilds the same cost summary every Friday from four different spreadsheets. A three-person letting agency in Manchester re-keys tenant references from PDF to portal, one field at a time.
None of these are strategic problems. They’re plumbing. But plumbing has a way of consuming the day, and the people doing it are usually the ones you hired for something else entirely.
Why it’s got worse, not better
Software was meant to fix this. In a sense it did — and in another sense it made things worse.
The average UK SME now runs on somewhere between eight and twenty separate tools. A CRM. Xero or QuickBooks. A booking system. WhatsApp Business. A shared inbox. Something for project management. A payment provider. Each one is good at its own job. Almost none of them talk to each other properly.
So the humans become the integration layer. Every gap between two systems is filled by a person copying, pasting, forwarding, and remembering. The more software you buy, the more gaps you create, and the more of your team’s week gets spent as glue.
What automation actually means now
Here’s where things have genuinely changed in the last two years, and why the conversation is worth having again even if you dismissed it before.
Traditional automation — the “if this, then that” kind — could only handle work that was perfectly structured. If an email arrived with a subject line in exactly the right format, a rule could fire. If it didn’t, the rule broke. That covered maybe 20% of the admin tax and left the rest untouched, because most real business input is messy: a customer writes in prose, a supplier sends a PDF, a site manager fires off a voice note.
Language models handle messy. That’s the whole point of them. An AI system can read an email that says “Hi, we’re after a quote for the Sheffield job, similar to what you did for us in March but we’ll need it a fortnight earlier” — and correctly extract the client, the reference job, the deadline shift, and the intent. Ten years of automation software couldn’t do that. This can.
That single capability moves an enormous amount of work from the “needs a human” column to the “needs a human to approve” column. And that’s the shift worth understanding, because those two columns have very different costs.
The maths, roughly
Take a business with ten staff on an average salary of £34,000. Fully loaded — with employer’s NI, pension, holiday, and the rest — you’re at roughly £42,000 a head, call it £22 an hour.
Now suppose each person loses six hours a week to the admin tax. That’s conservative; in our audits it’s usually higher.
- 10 people × 6 hours × £22 = £1,320 a week
- Across a 46-week working year: roughly £60,000
That’s not a line on your P&L. There’s no invoice for it. But it’s a full salary and a half, spent on copying things between boxes.
The point isn’t that you’d fire anyone. The point is what those hours could have been instead — quotes turned around in an hour rather than three days, follow-ups that actually happen, a sales pipeline that gets worked rather than merely maintained.
What we’d tell you to do first
Not “buy AI.” That’s a vendor’s answer, not an operator’s.
The honest first step is much duller and costs nothing: for one week, get your team to note down every task that made them think “a machine should be doing this.” No apps, no consultants. A shared note will do.
At the end of the week you’ll have a list. Sort it by two questions:
- How often does this happen? Daily beats monthly, every time.
- How much does it hurt when it goes wrong? A missed enquiry costs more than a mis-filed receipt.
The tasks that are frequent and costly when dropped — that’s where automation pays for itself in weeks rather than years. Everything else can wait.
You’ll probably find three or four candidates. That’s normal. It’s also enough. The businesses that get value from this don’t automate everything; they automate the four things that were quietly bleeding them and then stop.
The bit people get wrong
The failure mode we see most often isn’t technical. It’s that someone automates a broken process and makes it broken faster.
If your quote process is slow because three people have to approve it and one of them is always in a meeting, an AI won’t fix that. It’ll just generate the quote quicker and then it’ll sit in the same queue. Automation is an amplifier. Point it at a good process and you get leverage; point it at a bad one and you get a mess at scale.
So before anything gets built, the process itself has to be honest — written down, agreed, and stripped of the steps that only exist because someone added them in 2019 and nobody’s questioned it since. In our experience that exercise alone recovers a surprising amount of time before a single line of automation is written.
Sonwa AI Lab builds AI automation for UK businesses that would rather their team spent the week on the work they were actually hired to do. If you’d like a straight answer on where your admin tax is going — and whether it’s worth doing anything about it — get in touch.
