Estimates of recoverable time vary wildly, and most of them come from companies selling automation. The number that matters is not the industry average. It is yours, and it takes about a fortnight to establish.
Ask a finance or operations team what they spend their week doing and you will get job titles. Ask them to keep a tally for five working days of every time they retype something that already exists in another system, and you will get a different conversation entirely.
The published figures are striking, and should be treated with appropriate caution given who publishes them. Surveys suggest the average finance or administrative worker spends somewhere between 40 and 60 per cent of the week moving data between systems. More than 40 per cent of workers report spending at least a quarter of their week on manual, repetitive tasks. Some vendor analyses claim 15 to 20 hours per employee per week is recoverable — a figure that should be read as a theoretical ceiling rather than a forecast.
The precise number is less important than the fact that almost no mid-market business has measured its own.
Why this cost stays invisible
Manual work does not appear anywhere in the management accounts. There is no line item called re-keying. The cost surfaces as headcount that feels necessary, as month-end taking longer than it should, as a backlog nobody can explain, and as capable people leaving because the work is dull.
It is also nobody’s problem in particular. The person doing the re-keying is not empowered to change the systems. The person who could change the systems does not see the re-keying. The gap between those two facts is where the cost accumulates, year after year, entirely undisturbed.
Nobody ever asked for this process. It is the residue of four system decisions taken years apart by people who never met.
The disconnection is structural
Mid-market technology estates are typically assembled rather than designed. A finance system chosen in one era, a CRM added when sales grew, a bespoke tool built for one department, a payroll platform inherited through an acquisition. Each was a sound decision. None was made with reference to the others.
The connective tissue between them is human. Someone exports, someone reformats, someone imports, someone reconciles the differences. The volume of that work rises with growth, which is why businesses often find that scaling revenue by 30 per cent required scaling administrative headcount by rather more.
Recent research found 74 per cent of manufacturers describing themselves as being in a state of “data chaos” despite increasing technology spend — which captures the pattern neatly. More systems, bought to solve the problem, frequently make the connective work worse.
Measure before you automate
The instinct at this point is to buy an automation platform. It is the wrong next step, and expensively so.
Automating a process without redesigning it industrialises whatever was wrong with it. If the reconciliation exists because two systems hold conflicting customer records, a bot that performs the reconciliation faster has entrenched the conflict and made it harder to see. The underlying fault is now automated, invisible, and load-bearing.
The sequence that works is duller and considerably cheaper. Measure first, over a short window, with the people doing the work rather than to them. Then ask of each task, in order: can this be eliminated, can it be simplified, and only then — should it be automated?
A two-week exercise that produces a real number
- Pick two teams where you suspect the problem, not ten — depth beats coverage here
- Have people tally instances, not estimate hours; recall is unreliable, tallies are not
- Record what triggers each task and what system the data came from — the trigger is where the fix usually lives
- Separate genuinely repetitive work from judgement work that merely looks repetitive from a distance
- Cost it at fully loaded rate and express it as annual capacity, not annual pounds — capacity is what a board can act on
- Ask each team what they would do with the recovered time; if there is no good answer, the saving is theoretical
What to do with the answer
The output is a ranked list, and the ranking usually surprises. The most painful task is rarely the most expensive one; the most expensive is typically something small repeated a great many times by several people, none of whom regard it as significant.
A meaningful share of what surfaces can be eliminated outright. A report nobody reads. A double-entry that exists because a field was never mapped during an implementation. An approval step introduced after an incident in 2019 that has never been reviewed. These fixes cost days, not quarters, and they build the credibility that funds the larger integration work.
The remainder splits into genuine integration — connecting two systems properly, once — and genuine automation, applied to a process that has first been simplified. Both are worth doing. Neither should be attempted before the measurement, because without it you are automating an anecdote.
The connection to everything else
This work is usually framed as efficiency. Its more interesting effect is on customer experience, because the same disconnection that makes internal work manual is what makes external experience inconsistent.
When a customer is asked for information the business already holds, when an order status cannot be answered without a phone call, when a complaint is handled without visibility of the last three — that is the same structural fault, seen from outside.
Fixing the internal version fixes the external one. That is a considerably better business case than headcount reduction, and it is the one worth putting to a board.