"We'll do it manually for now" is one of the most sensible sentences in business. It defers a decision cheaply, keeps a human in the loop while a process is still changing, and avoids building something before anyone understands it. Used deliberately, it's good judgement.
It's also, quite often, the single largest unexamined cost in the company — because unlike software, it never appears on an invoice, so nobody ever reviews it. A £200-a-month subscription gets scrutinised at renewal. A task costing £9,000 a year in salaried time gets no review at all, because it isn't a purchase. It's just how things are done.
Do the arithmetic once
Take any recurring manual task and work out its annual cost properly.
Start with the obvious part: minutes per occurrence, times frequency, times the fully loaded hourly cost of whoever does it. Loaded means salary plus employer's NI, pension, holiday cover and overheads — typically somewhere between 1.25 and 1.4 times the raw salary, though your accountant will have a better number for your business.
A task taking ninety minutes a week is roughly 78 hours a year — about two full working weeks. If it's being done by someone whose loaded cost is £45 an hour, that's around £3,500 a year for one task. Most businesses have between six and a dozen such tasks, and none of them have ever been costed.
- Time — hours × loaded cost × frequency. The only part anybody estimates.
- Error — what one mistake costs, times how often it happens. Include the time spent fixing it, not just the direct loss.
- Delay — decisions and cash waiting on the task getting done.
- Opportunity — what that person would otherwise be doing. Usually the biggest of the four, and always the hardest to face.
The cost that hides best
Delay is the one people miss, and it's often larger than the labour.
When a report takes a day to produce, it doesn't just cost a day — it means the business is permanently looking at last week. Every decision runs on information that's a little stale, and nobody can point at the loss because it never shows up as an event. It shows up as a series of decisions that were slightly worse than they needed to be.
The same applies to anything queued behind a human. If invoices go out on Friday because that's when somebody does invoicing, then every invoice is up to four days later than it needed to be — and so is every payment. For a business invoicing £40,000 a month on 30-day terms, systematically shaving four days off the cycle is a permanent, free improvement in working capital. Nobody experiences that as a cost, because it's the absence of something rather than the presence of it.
Delay also compounds with error. A mistake caught the same day is an amendment. The same mistake caught three weeks later, after the customer has seen it, is a conversation.
The cost nobody wants to name
Opportunity cost is where the real number usually lives, and it's uncomfortable because it implicates how senior people spend their time.
In small businesses, manual work has a habit of landing on whoever is most trusted, which is frequently whoever is most valuable. The owner does the Friday numbers because only the owner knows where everything is. The best salesperson maintains the pipeline spreadsheet because they care most about it. The most experienced technician does the scheduling because they can see the dependencies.
Every hour of that is an hour not spent on the thing only that person can do. And unlike the labour cost, this one doesn't scale down when you get more efficient at the task — you just get more efficient at doing the wrong work.
When manual is genuinely right
Three cases, and they're worth defending properly, because the software industry has an obvious interest in pretending they don't exist.
When the process is still changing. Automating something that will be structurally different in two months is waste. Build the automation when the shape has settled — and "settled" means it's been the same for a couple of months, not that everyone has agreed it should settle.
When judgement is the point. Anything requiring a read of a relationship, a negotiation, a genuine exception. The right target there is the preparation, not the decision: have the software assemble everything the person needs and then get out of the way.
When the volume is genuinely tiny. A task done four times a year by hand is fine. Automating it will cost more than it saves and add a thing that can break. Leave it.
Notice what's not on that list: "because building it feels like a big job". That's a reason to build something small, not a reason to keep paying indefinitely.
The partial-automation option people forget
The choice is rarely between fully manual and fully automatic. Most tasks split into a mechanical portion and a judgement portion, and the mechanical portion is usually 70–90% of the time.
Automating just that part — assembling, checking, drafting, flagging — while leaving the human the decision and the send button, captures most of the saving with almost none of the risk. It also keeps the validation layer in place, which is what stops the error rate going up.
This is nearly always the right first move, and it's less impressive-sounding than full automation, which is probably why it gets skipped in proposals.
A worked example
A firm with fourteen staff. One task: producing the weekly sales and stock report, done by the operations manager on a Friday afternoon.
- Time. Three hours a week, 48 working weeks, at a loaded cost of £42 an hour — £6,048 a year.
- Error. A transcription mistake roughly monthly; most are caught, one or two a year lead to a purchasing decision being revisited. Call it £1,500, conservatively.
- Delay. The report describes a week that finished five days earlier, so every stock decision runs on a stale picture. Two avoidable stock-outs a year on fast lines at £2,000 of lost margin each — £4,000.
- Opportunity. Three hours of the operations manager's Friday, every week, not spent on supplier negotiation or scheduling. Hard to price precisely; not remotely zero.
That's £11,500 a year before opportunity cost, for one task nobody had ever costed, because it never appeared on an invoice.
Automating the mechanical portion — the assembly and the arithmetic, leaving the interpretation with the manager — is a small build. The payback period isn't measured in years.
Because the person best placed to do it is the person doing the task, and the conclusion is uncomfortable for them. It reads as an argument that their time is being wasted — which it is, but by the process rather than by them.
Framing matters: the question is not "why does this take you three hours", it's "what would you do with those three hours".
Which manual tasks hide best
The expensive ones are rarely the ones people complain about. Four patterns worth looking for specifically.
Tasks absorbed into a role. Nobody calls it a task because it's just "part of what Karen does". These are the most expensive and least visible, because there's no name to attach to them.
Tasks that only happen when someone remembers. The cost isn't the doing, it's the not-doing — the follow-up that didn't happen, the account that wasn't reviewed. Invisible in a time audit and often the largest loss.
Batched work. Anything done weekly that could be continuous carries an average delay of half the batch period on every item. Invoicing on Fridays adds days to every payment.
Checking. Somebody verifying that something happened — did the order go, did the payment land, did the file arrive. Pure mechanics, usually done by someone senior, and almost never counted as work at all.
The honest test
For each recurring manual task, write down the annual cost from the four parts above, and next to it what it would cost to have software do the mechanical portion. One line each, on a single page.
Most of the time, one of two things becomes obvious. Either the manual cost is far higher than anyone assumed — often by a factor of three or four once delay and opportunity are included — and the decision makes itself. Or it's genuinely small, and you can stop feeling vaguely guilty about it and go and do something that matters.
Both outcomes are worth the hour it takes. What isn't worth it is carrying on with "for now" for three years without ever putting a number on it — which is, in fairness, exactly what almost every business does, because the alternative requires admitting how the time is currently spent.