A good internal system can pay for itself faster than a new hire.

Orders are piling up, invoices go out later than they should, and someone on the team has started working weekends. At this point a sentence gets said that sounds like plain common sense: we need another person.

Sometimes that is the right answer. More often it is an answer to a badly framed question. Hiring treats the symptom, the missing capacity, without asking why the capacity is disappearing in the first place. If half of the new capacity goes on re-keying data between email, a spreadsheet and the accounting system, you have bought an expensive solution to a cheap problem.

This is not about cutting jobs. It is about the fact that expensive people should not be doing cheap administrative work, and that one calculation is worth doing before you sign the contract.

The first response to growth is almost always a hire

Hiring has one clear advantage: it is easy to understand. You know what you are buying and what it costs each month. Redesigning a process is far less comfortable. It requires taking apart how the work actually flows through the business, not how everyone assumes it flows. The output is not a person you introduce to the team, but a change nobody notices three months later, because it works.

Both options draw on the same budget. Economics calls this opportunity cost: according to Econlib, the value of the best alternative use of that same resource. Money spent on an eleventh role is not available to stop the ten people you already have doing work they should not be doing.

First the question: why did that workload appear at all

Administrative burden is not a law of nature. It grows out of decisions nobody ever revisited. Two systems that do not talk to each other, so a person re-keys the data by hand. A report one client once asked for, which has gone out to everybody ever since.

According to the Anatomy of Work Index from Asana, knowledge workers spend roughly 60 percent of their time on work about work, meaning searching for information, switching between applications and communicating about the work itself, which leaves 40 percent for skilled work. Treat the figure with caution, because it is published by a software vendor that sells the solution to the problem it describes. So the question before hiring is not how many people are missing, but how much capacity is being eaten by work that adds nothing.

Split the work into five types, not into headcount

Write out the remit of the role in question across five categories and assign percentages to each:

• Judgment work. Decisions under uncertainty, assessing an exception, pricing an atypical case. The inputs differ every time and no rule can be written for it.
• Relationship work. Negotiation, complaints, building trust. The value comes from the fact that there is a person on the other side.
• Repetitive work. The same input, the same steps, the same output, all to a rule. You are not paying for judgment here, you are paying for clicking through.
• Data transfer. Moving figures from an email into a spreadsheet and from the spreadsheet onto an invoice. It creates no new information, it only adds the risk of a typo.
• Reporting. Assembling numbers that already exist, by hand, into a document that gets opened once a month.

The split is not arbitrary. Work by Carl Benedikt Frey and Michael Osborne of Oxford estimated in 2013 how susceptible 702 occupations were to computerisation and identified three bottlenecks where machines fall short: perception and manipulation, creative intelligence and social intelligence. Judgment work and relationship work both sit inside those bottlenecks. Re-keying data between spreadsheets does not.

What to eliminate, simplify, automate, and what to leave to people

The order matters, and most companies have it back to front. They start with automation and end up automating work that should not have been done at all.

Eliminate. The cheapest work is the work that is not done. An approval step that has stopped nothing in three years does not need a faster version, it needs scrapping. It is the cheapest intervention available and the one made least often.

Simplify. Fewer steps, fewer tools, fewer handovers between people. I have written separately about why more software is a worse answer than less chaos.

Automate. Only now, and only what survived the first two steps and runs to a stable rule. An automated bad process is still a bad process, merely faster and quieter, which goes double for artificial intelligence, as I set out in the piece on why AI will not fix a broken process.

Augment. Leave the rest to people and give them better tools. The goal is not a person without work, but a person who calls clients instead of chasing invoices.

The real cost of a role is higher than the gross salary

If you compare a gross salary with the price of a system, you are comparing two things that cannot be compared. The technical term is total cost of employment, and Eurostat calculates labour costs on the same logic: wages, employer social contributions, training costs and other expenditure, a category into which it explicitly puts recruitment costs.

Under the rates in force from 1 January 2026, an employer pays 25.2 percent of the gross salary in social insurance as standard, with the individual items varying by type of employer and employee. Add 11 percent for health insurance and the total comes to 36.2 percent. Every euro of gross salary therefore carries EUR 1.362 of real cost. According to Eurostat, the share of non-wage costs in total labour costs in Slovakia in 2025 was the third highest in the European Union at 28.6 percent, behind France and Sweden and above the Union average of 24.8 percent.

And that is only the beginning. Then come recruitment, onboarding, management time, hardware and licences. SHRM has reported an average cost per hire of almost 4,700 US dollars, although that is the American market and the figure was published in 2022, so treat it as an order of magnitude. Gross salary is the floor, not the answer.

On the other side of the scales sits a cost with a different curve

A system behaves the other way round. The OECD spells this out clearly in Vectors of Digital Transformation: physical products carry high fixed costs and substantial marginal costs, while digital ones carry mostly fixed costs and marginal costs close to zero.

The tenth invoice processed automatically therefore costs practically the same as the thousandth. That is the marginal cost of automation, the cost of processing one more unit of work. With an employee it rises in line with volume; with a system it is close to zero, until you hit an exception the rule does not cover.

A system also carries running costs that companies routinely underestimate: licences, maintenance and the time people spend during rollout. I put those at 15 to 25 percent of the initial investment per year, though that is my own estimate from practice, not a benchmark.

The payback period and how to calculate it conservatively

The payback period, according to the Corporate Finance Institute, shows how long it takes to recover an investment through the cash flows it generates. It can be worked out on the back of an envelope and it tells you how long the money is out of the door. The same source admits its weaknesses: it does not show the profitability of the investment and it discounts neither risk nor opportunity cost. Use it as a filter, not as a verdict.

A conservative calculation means four disciplines:

1. Count only the hours that genuinely disappear, not the ones that are theoretically saveable.
2. Value them at the total cost of employment, not at the gross salary.
3. Deduct the annual running cost, not just the one-off investment.
4. Assume that part of the work stays with people. If the model comes out at a hundred percent elimination, the error is in the model.

And one condition. The saving is only real if the freed capacity moves across to sales, quality or client service. If you roll out the system and hire the person anyway, the return is zero.

When an in-house system makes no economic sense

This is the part software vendors tend to skip. An in-house system is the wrong answer more often than it might seem.

• Low frequency. If a process runs twenty times a year, no calculation will work out. The annual loss is duration multiplied by the number of repetitions, so at low frequency it stays small.
• Unstable rules. If the process changes every quarter, you will pay more for rewriting the system than for the work it replaced.
• An off-the-shelf solution exists. If a standard tool covers 80 percent of the need for a monthly fee, buy it and handle the remainder through the process. I say that as someone who builds bespoke systems for a living.
• It is mainly judgment or relationship work. In that case you do not need a system, you need a person, and hiring is the right answer.
• There is no process owner. A system with nobody accountable for it stops fitting the process at the first change of rules, and the team quietly drifts back to spreadsheets.

A worked example: a ten-person firm considering a hire

The example is hypothetical and the workload figures are modelled assumptions rather than measurements taken at a client. The rates and salaries come from public sources.

The firm in the model has ten people and is weighing up an administrator on a gross salary of EUR 1,400. At a coefficient of 1.362 that is EUR 1,907 a month, roughly EUR 22,900 a year before recruitment and equipment. First, though, it works out its cost of friction: time multiplied by frequency, multiplied by the number of people, multiplied by the hourly cost.

• Re-keying orders between email, a spreadsheet and the accounting system: 45 minutes a day, three people, 220 working days, so 495 hours a year.
• Assembling monthly reports by hand: 6 hours a month, two people, so 144 hours a year.
• Chasing the status of tasks and duplicate re-keying: 20 minutes a day, four people, so 293 hours a year.

That comes to 932 hours a year. The average monthly wage in the Slovak economy in 2025 stood at EUR 1,620 according to the Statistical Office of the Slovak Republic, and with the coefficient of 1.362 the total cost of labour is roughly EUR 26,500 a year. Across 1,760 working hours that works out at about EUR 15 an hour, deliberately a low figure with no overheads included. The cost of friction of this process is therefore around EUR 14,000 a year.

The firm is weighing up two routes: a redesign built on off-the-shelf tools at EUR 8,000 as a one-off plus EUR 1,800 a year, or a bespoke system at EUR 18,000 plus EUR 2,400 a year. Conservative assumption: the solution removes 70 percent of the quantified burden and the remainder stays with people. The gross annual saving is EUR 9,800.

• Off-the-shelf tools. 9,800 minus 1,800 gives a net saving of EUR 8,000 a year, and a payback period of roughly twelve months.
• Bespoke system. 9,800 minus 2,400 gives EUR 7,400, and a payback period of roughly twenty-nine months.

The conclusion is an uncomfortable one for a vendor. The cheaper route has both the better return and the lower risk, and I would not recommend a bespoke system to a ten-person firm unless off-the-shelf tools hit a ceiling. The calculation deliberately leaves out the deferred hire at EUR 22,900 a year, because it is not a certainty. How I would build a company that barely needs administration of this kind, I have set out separately in the piece on building a company without an admin army.

How to measure the return after rollout

A calculation made before the decision is a hypothesis. Without measurement it turns into an anecdote. So capture the baseline before you change anything.

• Time per transaction. How many minutes it takes to process an order from receipt through to posting.
• Volume per person. How many cases the team handles with headcount unchanged. This is the direct substitute for a hire.
• Error rate and the cost of putting things right. Complaints and credit notes. Errors from re-keying are booked nowhere, yet they are paid for.
• Lead time from enquiry to invoice. An invoice sent a week earlier tends to be paid a week earlier too.
• Share of time spent on judgment work. If people are doing the same things, only faster, you have missed the main benefit.

Review the numbers after three months and after twelve. The first tells you whether the change took hold, the second whether it survived changes in people, volumes and rules.

Hiring is not a bad decision, it simply should not be the first one. An hour spent over a breakdown of the work and a single calculation will tell you whether you are buying capacity or paying for a problem you could remove.

Key takeaways

• Before hiring for admin work, calculate the total cost of employment, not the gross salary. Under the rates from January 2026, every euro of gross salary costs the employer EUR 1.362, and that is still before recruitment and equipment.
• Break the remit of the role into five types of work. If judgment work and relationship work account for well under half, hiring is probably not the cheapest answer.
• Work out the annual cost of friction and set it against the investment. With 70 percent of the burden eliminated and annual running costs of around a fifth of the investment, a simple rule applies: if the investment is higher than the annual cost of friction, payback will take more than two years.
• Calculate the payback period conservatively: only the hours that genuinely disappear, valued at the total cost of labour, less the annual running cost.
• Capture the baseline before the change. Without a figure from before the rollout, you will never know whether the investment paid off or whether it merely looks that way.