When the yen fell, the person-month stopped working

Dalian's software industry was built on one unit of sale. A Japanese integrator needed a system built or maintained. It estimated the effort in person-months, agreed a rate in yen for each person-month with an offshore vendor, and the vendor's engineers, paid in renminbi, did the work. The model rested on Japanese-speaking engineers, on bridge engineers who turned Japanese design documents into tasks, and on a level of review that made Dalian code acceptable to clients who will return work over a misnamed variable. It made Yanxiao possible, and much of the city's software industry with it.

The unit has two halves: a currency and a quantity of human time. Both are now failing, for different reasons, and I think the industry's usual response to each is mistaken.

What the yen exposed

From early 2021 to the end of 2023 the yen lost about a fifth of its value against the renminbi. For a vendor paid in yen and paying in renminbi, that is not a fluctuation to be hedged; it is a change in the price of everything. Over the same period Dalian engineers expected pay rises of around 6 per cent a year, because domestic employers were competing for them. Our Japan gross margin fell from 32 per cent in 2021 to 23 per cent in the second half of 2022.

The industry's responses were predictable, and I considered all of them. Forward contracts can fix the rate for a year, but they only postpone the new level. Pay freezes protect margin until the best engineers, usually the ones with the best Japanese, leave for domestic employers. Moving work to cheaper Chinese cities or to Vietnam helps, but slowly, and it does nothing for contracts already signed. And the response most vendors chose in practice was silence: absorb the loss, hope the yen recovers, and avoid an uncomfortable conversation with clients who value stable relationships above almost everything.

The silence is what the yen really exposed. Our person-month contracts had always placed the entire currency risk on the vendor, but nobody had said so, because for years the risk had been small. Japanese clients had never been asked to carry any of it, and when the risk became large, raising it felt like breaking an understanding that had never been written down.

A band rather than a bet

In December 2022 I proposed a different allocation to our eighteen largest Japanese clients. For the first year the base was the 2021 average rate, so that clients shared part of the fall already behind us; after that the parties reset the base at each renewal. Yanxiao absorbs the first 5 per cent of any movement in the yen against the base. Beyond 5 per cent the movement is shared equally, in either direction: if the yen recovers, clients pay less.

The symmetry mattered more than the arithmetic. Japanese purchasing departments are rightly wary of suppliers who ask for help only when conditions turn against them. A clause that would have reduced their costs in a year of yen strength was easier to carry through internal approval than a request for a price rise, because it could be described as a rule rather than a concession. It also obliged us to explain, in writing and in Japanese, exactly how our costs behaved, which was uncomfortable for a company that had never shown its margins to a client.

Fourteen of the eighteen accepted during 2023. Two preferred a one-time increase of 6 per cent. Two moved their work elsewhere, taking RMB 146 million of annual revenue; our Japan president had warned of exactly that, and he was right to. Even with the clause we held Dalian pay rises to 2 per cent in 2023, and our engineers bore that for a year. By 2025 Japan gross margin was back at 28 per cent, and every new Japanese contract carries the clause.

I would make one point to other vendors. Propose such a clause when currencies are calm. In 2022 ours looked like a demand. In 2020 it would have looked like prudence, and clients would have signed it more easily.

Pricing the output when the tools change the hours

The second half of the unit is failing more quietly. AI-assisted coding tools have changed how many hours certain tasks take. In our own measurements across Japanese maintenance work, code conversion and unit-test writing now take between 20 and 35 per cent fewer hours than in 2023, while design, review and acceptance take about as long as before.

Under person-month pricing, every hour saved is revenue lost. That gives vendors a reason to use the tools without saying so and to bill the old hours, and clients have noticed. Several Japanese clients now ask in their questionnaires whether AI tools are used, partly for data-protection reasons and partly, I suspect, because they wonder what they are paying for. A unit that rewards the vendor for concealing productivity is no longer an honest unit.

We have been moving Japanese work to fixed prices for defined scopes and to unit prices for repeatable outputs: a converted program, a test case written and executed, a maintenance request closed within its service level. These contracts made up 9 per cent of Japan revenue in 2021 and 31 per cent in 2025. The productivity saving is now shared openly. Clients see a lower price per output, and we keep a margin that does not depend on concealment.

This has costs. Fixed prices transfer estimating risk to us, and two early fixed-price projects lost RMB 23 million before our estimates improved. Unit prices need definitions that both sides accept, and agreeing what counts as one converted program took months with one client. Some clients' procurement rules still assume person-months, and their approval processes have not caught up.

The person-month will not disappear quickly from Japanese offshore work, because it is built into how integrators estimate and how their own clients budget. But a vendor that wants to remain in that market for another twenty years should stop treating it as natural. It was a convenient unit for a period when the yen was stable and hours were the main input. Neither condition holds now, and the honest response is to write down who carries the currency and who keeps the saving.