What does back-office automation actually cost? Straight numbers, for once

Ask "what does it cost?" in this industry and you'll get a discovery call, a workshop, and a proposal with a number that appears to have been chosen by darts. Nobody publishes prices.

There's a half-decent reason — every business's mess is different — and a worse one: vagueness lets sellers price by how worried you look. So here are the actual shapes of the numbers, because you shouldn't need to book a call to learn what a market costs.

The three cost shapes

1. Off-the-shelf software: tens of pounds a month, per thing, forever. Invoice-chasing add-ons, inbox tools, document-scanning apps — typically £20–£300 a month each. This is the right answer when your process fits the tool. The catch is the "per thing, forever" part: businesses quietly accumulate five or six subscriptions that each do 70% of a job, and the office still bridges the gaps by hand. The subscriptions were cheap; the remaining 30% is where the hours still bleed.

2. A custom-built system: a one-off build cost, then small running costs. The order-entry, invoice-chasing, paperwork-extraction systems we write about — built around your process and connected to your software. Depending on complexity, builds like these generally land somewhere between a few thousand and the price of a decent used van. Running costs are modest and honest: the AI itself is pennies per document processed, plus normal small hosting fees. Not per-seat, because machines don't have seats.

3. The enterprise project: six figures and a steering committee. ERP replacements, "digital transformation programmes." Real, occasionally necessary, and almost never what a £5M–£50M business drowning in retyping actually needs. If someone's first suggestion is replacing your ERP, get a second opinion before signing anything.

Most growing trade businesses belong in shape two, sometimes mixed with a bit of shape one.

What actually moves the price

When two builds differ in cost by 5x, it's nearly always one of these:

• How many systems it touches. Reading an inbox and writing to one place is simple. Coordinating inbox + ERP + courier portal + accounts package is plumbing across four things, each with its own quirks.
• The exception rate. If 95% of your orders are clean, the build is cheap. If a third of them are "the usual, but ring me about the odd sizes," handling the mess is the work.
• The state of your data. Duplicate accounts, product codes only Karen understands, prices living in someone's head — the system can't match against chaos, so someone has to fix the chaos first. (Worth doing regardless; the chaos was already costing you.)
• What "wrong" costs you. A system suggesting draft replies can tolerate the odd miss. One entering orders needs checking workflows and audit trails. Tolerance is a dial, and precision costs.

Notice what's not on the list: how impressive the AI sounds. The intelligence is the cheap part now. You're paying for plumbing and edge cases.

The only arithmetic that matters

Ignore percentages in vendor decks. Do this on the back of an envelope:

1. Hours per week the process eats (count honestly — including checking and fixing).
2. × the real hourly cost of the people doing it.
3. × 48 weeks. That's the annual cost of the status quo.
4. Put the build quote next to it. A one-off cost against a recurring one.

For comparison, we've written before about the true cost of an admin hire — realistically £32–35k a year, every year, rising. A build that removes most of a hire's workload pays for itself in months and then keeps paying. A build that saves ninety minutes a week does not, and anyone who'd sell it to you anyway is not your friend.

Red flags when you're buying

• Day rates with no defined outcome. You're funding an open-ended adventure.
• Per-seat pricing for machine work. Seats are for humans.
• "AI transformation roadmap" before anyone has watched your office work. You can't price a fix for a process you've never seen.
• Nothing about exceptions, checking, or what happens when it's wrong. The demo always works. Ask what Tuesday-afternoon-when-it's-confused looks like.
• Reluctance to give ranges. Custom work can't be priced to the pound up front, but anyone experienced can give you a shape. Refusing to is a choice.

Where the walk-through fits

This is why we work the way we do: a free day walking your operation, then a ranked plan where every item carries its own price and its own payback arithmetic — this fix costs X, saves roughly Y a year, pays back in Z months. You see the numbers side by side before committing to anything, and the plan is yours to keep even if you build none of it (or have someone else build it — it happens, and it's fine).

We'd rather compete on arithmetic than adjectives.

What to do this week

Pick the one process everyone complains about. Time it honestly for a week — hours, not vibes. Multiply out to a year at real wage cost.

Now you know your budget: whatever fixes it for meaningfully less than that number, with payback under a year, is worth doing. Whatever costs more isn't — whoever's selling it, including us.