The minutes that didn't add up

The premise

If saved time has to pass through a channel to become value, then where no channel exists the saving shows up as neither money nor free time.

That’s a condition, not a conclusion. This piece doesn’t say AI saves nothing, or that hours should be cut.

Was anything actually saved?

Ten minutes here, ten minutes there. In an earlier piece I did a rough sum: ten-odd tasks a day, ten minutes saved on each, two hours a day, about 30 hours a month. Then the question: where did those hours go?

The saving itself is measured. In Denmark, two economists surveyed workers in 11 occupations, about 25,000 in each of two rounds. Chatbot users said the tools saved them, on average, 2.8% of their working hours. In an experiment across 66 firms and 7,137 office workers, those who used the AI tool spent two fewer hours a week on email, and worked less outside normal hours.

The other side of the same studies reads differently. When the Danish researchers matched the answers against administrative records, earnings and hours had not moved. Effects larger than 1% were ruled out. The email experiment found no sign that handing the tool to individuals changed how much work they did, or what kind.

The minutes didn’t vanish. They flowed into other tasks. What they did not do was turn into anything countable. Ten minutes on its own buys neither money nor rest.

Hence the question here: where does saved time stop before it becomes value?

The channels it has to pass through

In Belgium, two routes raise what work pays. The law shapes both.

The first is automatic indexation. Wages follow consumer prices upward. It is why Belgian households are relatively sheltered from inflation, and the central bank notes that indexation has been the main driver of wage growth for two decades. Indexation follows prices. It doesn’t follow productivity.

The second is the wage norm. On top of indexation and scale increases, the law sets a maximum for how much labour costs may rise, every two years. Article 5 of the law of 26 July 1996 says how that maximum is worked out: take the forecast labour costs of Germany, France and the Netherlands, subtract the indexation expected in Belgium and a correction term, then take 25% of what remains as a safety margin and subtract that too.

Productivity is not one of those inputs. The word does appear in the same article — employers’ labour-cost handicap against the three neighbours is to be reported corrected for the level of productivity. It is a reporting item, not part of the sum.

And for 2025-2026 the maximum is 0%. With no agreement between employers and unions, the government set it. Indexation and scale increases continue outside that limit.

In numbers:

Belgium, 2015 = 100	2025
Real labour productivity per hour	104.9
Nominal hourly compensation	135.0 (€38.0 → €51.3)
Consumer prices	135.5

Over ten years, productivity rose 4.9%. Hourly pay rose 35%, and consumer prices rose 35.5% — so in real terms, measured against what households buy, hourly pay is roughly where it was. In this table, the productivity gain leaves no trace on the wage side.

And on the other side?

You can also look at how much of the value created goes to labour: compensation of employees as a share of gross value added. From Eurostat data, I make it 57.7% in 2013 and 53.8% in 2024 for Belgium. Over the same years the EU as a whole went from 53.0% to 52.9%.

The National Bank of Belgium has published an analysis of that fall. Its main cause, the Bank writes, is rising labour productivity without a corresponding increase in real hourly wages. In the euro area, by contrast, real hourly wages rose by even more than productivity, leaving the share broadly stable.

The same report blocks the easy conclusion. Cuts to employer social security contributions held labour costs down, which pushed the share lower on their own. And firm by firm, between 2013 and 2022, more companies saw their wage share rise (52%) than fall (48%) — the aggregate is driven by the largest firms. So this evidence cannot carry the sentence “firms took it all.” Where an extra hundred euros of output actually goes was the subject of another piece.

Is the factory ready to change?

A century ago, electric motors in factories at first simply turned the old drive shafts. The productivity figures moved only once each machine had its own motor and the buildings and layouts had been rebuilt around them. It was the factory that had to change, not the motor.

Today the factory is the set of rules that fix pay and working time. Where the doors are in those rules can be checked.

Door one: the wage norm is reset every two years. For 2025-2026 it is shut at 0%. For 2027-2028 nothing is set yet. Putting productivity into the sum would mean amending article 5 of the 1996 law.

Door two: a narrow door already stands open beside the norm. Article 10 of the same law lists what is left out of the labour-cost calculation: profit participation, worker capital participation, innovation bonuses, the 2023 purchasing power premium, a €2 rise in meal vouchers for 2026. Value comes through as a one-off payment rather than as wages. It doesn’t repeat, and the firm decides who gets it and how much.

Door three: the route that pays in time is open but narrow. The “four-day week” in the law of 3 October 2022 doesn’t shorten the week; it compresses it. At 38 hours or less, the daily maximum rises to 9 hours 30 so the same hours fit into four days, the worker has to ask in writing, and the arrangement runs for up to six months at a time. Weekly hours and pay stay as they were.

Door four: the most recent move went the other way. Under the social security office’s instructions, from 1 June 2026 reference wages above €4,000 a month are indexed only up to 2%, and a special contribution falls due on the part held back. The system did move. It moved to hold labour costs down, not to carry productivity into pay.

So the answer isn’t yes or no. The doors are identifiable and the dates they reopen are known. What goes through them is not for this piece to choose.

What this piece does not claim

• That time saved with AI is of no use
• That firms captured that value
• That working hours should be cut, or that indexation should be reworked
• Any prediction about what the rules will do after 2027

What we checked, and what we couldn’t

Checked

• Denmark: reported average time savings of 2.8% of work hours; no effect on earnings or hours in administrative data (effects above 1% ruled out) — Humlum and Vestergaard (July 2025 version)
• Experiment with 7,137 workers at 66 firms: two fewer hours a week on email; no detected change in the quantity or mix of tasks from individual-level provision — Dillon et al. (2025)
• How the wage norm’s maximum is calculated — from the forecast labour costs of Germany, France and the Netherlands, less expected indexation and a correction term, less a 25% safety margin; productivity appears only as a reporting item — law of 26 July 1996, article 5
• Maximum of 0% for 2025-2026, with indexation and scale increases guaranteed separately — Federal Public Service Employment
• What is excluded from the labour-cost calculation (profit participation, capital participation, innovation bonuses, purchasing power premium, meal voucher rise) — same law, article 10
• The four-day week fits the same hours into four days, daily maximum of 9 hours 30 at 38 hours or less, on the worker’s written request — law of 3 October 2022
• From 1 June 2026, indexation of reference wages above €4,000 is capped at 2% and a special contribution applies — National Social Security Office instructions
• Belgian real hourly productivity 2015 = 100 → 104.9 in 2025; nominal hourly compensation €38.0 → €51.3; consumer prices 100 → 135.5 — our own calculation from Eurostat raw data
• Compensation of employees as a share of gross value added: Belgium 57.7% (2013) → 53.8% (2024); EU 53.0% → 52.9% — same method
• The fall is mainly productivity rising without matching real hourly wages, with the euro area the other way round; employer contribution cuts also played a part; firm by firm, more companies saw their wage share rise than fall — National Bank of Belgium, Economic Review 2025 No 7

Couldn’t check

• My 30 hours is a rough personal estimate
• What workers actually did with the time they saved (the Danish question asked what they expected to do)
• How much the one-off payment routes are used in practice
• The article numbers of the law behind the June 2026 indexation cap
• The wage norm for 2027-2028
• How other countries connect productivity to pay, if they do