Four times we chose the smaller number we could defend

Asking Japanese clients to share the yen's fall

By the autumn of 2022 the yen had lost more than a fifth of its value against the renminbi since the start of 2021, and nobody could say where it would settle. Japanese clients provided 61 per cent of our revenue. Our contracts were priced in yen per person-month; our engineers were paid in renminbi and expected annual rises of about 6 per cent. Japan gross margin fell from 32 per cent in 2021 to 23 per cent in the second half of 2022.

Two proposals reached the executive committee. Hidenori Kasamatsu, president of Yanxiao Japan, argued that asking long-standing clients to pay more because of their own currency would damage relationships built over twenty years; he proposed freezing pay in Dalian and moving junior work to Shenyang. Our chief financial officer, Pei Wenjing, wanted to extend forward hedging. I accepted that hedging could smooth a year but not change the level, and that a pay freeze would cost us the engineers with the best Japanese.

I proposed a clause to our eighteen largest Japanese clients, who provided 74 per cent of Japan revenue. For the first year the base was the 2021 average exchange rate, so that clients shared part of the fall that had already happened; after that the base resets at each renewal. Yanxiao absorbs the first 5 per cent of any movement from the base, and the rest is shared equally, in either direction if the yen recovers. Kasamatsu's memo went to the board with mine. The chairman supported my proposal on condition that I visit all eighteen clients myself.

Fourteen accepted the clause during 2023. Two preferred a one-time rate increase of 6 per cent. Two, a Nagoya machinery group's IT company and an Osaka regional bank's systems subsidiary, moved their work to other vendors, taking RMB 146 million of annual revenue. We still had to hold Dalian pay rises to 2 per cent in 2023. Japan gross margin recovered to 26 per cent in 2024 and 28 per cent in 2025, and every new Japanese contract now carries the clause.

I would do two things differently. I would have proposed the clause in 2020, when the yen was steady and a symmetric clause looked like prudence rather than a demand. And I would have visited the two lost clients before their renewal dates, not after them.

Refusing a billing cut-over the database could not yet carry

In early 2023 Liaoxin Port Group, a fictional provincial state-owned port operator, awarded us a RMB 186 million contract to move 41 applications, including container billing and settlement, from a foreign database and operating system to domestic products. The group had committed to finish its core migrations by December 2023, and that date appeared in its managers' annual assessments.

Our load test, run on a copy of the port's busiest month-end, was discouraging. The domestic database version then available took 6 hours 40 minutes to complete a monthly billing batch that the old system finished in 1 hour 50 minutes. Rounding and collation differences changed 0.3 per cent of tariff calculations: small amounts, each of which would have become a dispute with a shipping line.

Liaoxin's head of IT, Zou Ruiping, asked us to cut over all 41 applications by the deadline and tune afterwards. Our own domestic sales head agreed with him, since the whole contract was at stake. I decided that we would migrate the 40 peripheral applications on time but would not accept the billing cut-over before three months of parallel running, and I sent that position to the client in writing with the test results attached.

The group re-tendered the work. A competitor took the 40 applications, worth RMB 112 million, on the original date. We kept billing and settlement, worth RMB 74 million, because no other bidder would accept parallel running. We worked with the database supplier's engineers on partitioning and batch design, and their next release brought the batch to 2 hours 10 minutes. Parallel running found 1,900 invoice differences, each traced to a rule and corrected. Billing moved in September 2024, and its first month produced no disputed invoices. In 2025 Liaoxin gave us RMB 138 million of work on its bulk-cargo terminals.

Every domestic migration bid we make now starts with a written readiness report that lists what will not yet run at equal performance.

What would I change? I would bring the database supplier into the load test in the first month rather than the fourth. And I would take the results to the group's IT committee, which had set the deadline, rather than only to the project owner who had to meet it.

Moving a German client's controlled scope out of China

Wendelstrom Antriebstechnik, a fictional German maker of drivetrain components, had been a client since 2011. By 2023 it was our largest European account, at RMB 168 million a year, with 410 Yanxiao engineers in Dalian testing embedded software and maintaining engineering-data systems. In January 2024 we found that one of our teams had copied client technical files into a Dalian project share without the classification check our contract required; my Board Readiness page describes how we reported it. In March, Wendelstrom's export-compliance review concluded that part of its toolchain and technical data, including items its lawyers treated as controlled under US and European rules, could not be accessed from China. It asked us to move that scope out of China within nine months, or it would transfer the work to a vendor in India.

Two people I respect argued against moving. Ren Shuo, who ran the Dalian unit, said the move would tell 7,000 Dalian engineers that their employer did not trust them. The director nominated by the Beiganghu Innovation Fund, our municipal shareholder, opposed taking work out of the city. I judged that the client was not asking for an opinion of our engineers but for a position its lawyers could defend, and that refusing would lose all 410 roles rather than some of them.

We moved about 150 roles: 110 to our Ho Chi Minh City centre, opened in 2023 for Japanese clients, and 40 to Munich, where our office grew from 20 to 90 people. The other 260 Dalian engineers continued on work outside the controlled scope, in a separate network domain whose access logs the client can inspect. The move cost RMB 21 million, and the relocated scope costs about 22 per cent more to deliver. Of the 150 Dalian engineers whose work moved, 112 were redeployed to other accounts and 38 left. Ren Shuo left in 2025.

The client stayed. Wendelstrom revenue reached RMB 181 million in 2025, and defects escaping to the client on the relocated scope were no higher after six months. Two other European clients have since asked for the same arrangement.

I would have offered the separation in 2023, before the client had to ask for it. And I would have explained the decision to the Dalian unit in person on the day it was made, not three weeks later.

Listing in Hong Kong rather than relabelling delivery as research

By 2023 Yanxiao had been preparing a listing for three years. Changxu Capital, a fictional private equity firm that had invested in 2020 and then held 11 per cent, wanted the STAR Market of the Shanghai Stock Exchange, where software companies traded at far higher multiples. Our sponsor's preliminary review was frank. The exchange's guidance on science-and-technology attributes looks at research spending as a share of revenue, research staff, invention patents and growth. Our research spending was 3.8 per cent of revenue, because most of our engineers deliver client work.

The sponsor suggested that the time delivery engineers spent building reusable test-automation and code-conversion tools could be reclassified as research, lifting the ratio above 6 per cent. Some of that work had produced tools we used across clients. Most of it had been billed to clients as delivery. I told the board I would not sign a prospectus built on that reclassification. It would mislead investors about what kind of company we were, and it would have to survive the exchange's written inquiries, where I did not think it deserved to.

Changxu's partner, Mo Jiayi, argued that Hong Kong would value us at perhaps half of what STAR might, and that I was choosing a lower price for every shareholder, including employees holding shares through our incentive platforms. She was right about the price. Changxu's nominee director abstained when the board chose Hong Kong in February 2024. Mainland approvals for new listings had also slowed sharply, which made the decision easier but did not make it for us.

We completed the filing with the China Securities Regulatory Commission that domestic companies now make before listing abroad, and our H shares began trading in November 2025. We raised HK$1.08 billion at about sixteen times 2024 earnings. Changxu sold part of its stake in the offering and now holds 8 per cent. The prospectus stated research spending at 3.8 per cent, with a separate note on tool development.

What would I do differently? I would have separated genuine product development from delivery in our accounts in 2020, three years before any sponsor asked, so that the honest ratio was also a well-evidenced one. And I would have told Changxu my position before the sponsor's review rather than after it.