Scenario planning when policy can fork: the HarmonyOS example

On 23 November 2024, Reuters reported that Huawei was targeting 100,000 apps for its HarmonyOS operating system within the next 6 to 12 months, when it had more than 15,000 (Reuters, 2024). For the product director of a Vietnamese app studio, that news shrinks into a nearer question: should part of the engineering budget go to a second ecosystem? The answer hangs on something the studio cannot control: whether the restrictions on Huawei will ease, hold or tighten. The studio is hypothetical and every figure attached to it is mine. Huawei is only the public setting, and the post says nothing about the company's internal plans. The full case is at Huawei and HarmonyOS.

When policy can go three ways, why not best, middling and worst?

Because the branches of a policy decision differ in the structure of the market, not only in size. The scenario planning framework names this very error: turning scenarios into best/worst/middle when they should differ in structure. The post One number or several scenarios? walks through six steps and a two-axis matrix. This one takes a single axis and goes deep on it.

The driver comes from the Political box of PESTLE analysis, with the framework's warning in mind: pick one factor and show why it is decisive, using evidence from outside the company. The evidence sits in the Reuters report. Huawei unveiled Harmony in August 2019, three months after Washington placed it under trade restrictions, and the operating system came after US sanctions cut off support for Google's Android. Xu Zhijun, whom Reuters calls Huawei's chairman, said the company had been forced to speed up building its own operating system, and that any operating system is worth nothing if nobody uses it (Reuters, 2024). My reading is that the direction of policy therefore decides whether the second ecosystem needs outside developers at all. The same report notes that the US president-elect was then threatening to be tougher on China (as at November 2024).

The studio's focal question: how much should it spend over the next 18 months to support HarmonyOS? It has 400,000 monthly active users on Android and iOS, of whom 5 per cent (20,000 people) use Huawei devices.

How do the three branches tell three different stories?

Each branch changes a different thing in the studio's market. The last column is the assumed value of the HarmonyOS channel: the net revenue the studio earns through it over three years, before build costs.

The "Tightened" branch need not mean a ban, and two public cases show why. In June 2020 Singapore gave final approval to two standalone 5G networks: Singtel was in talks with Ericsson, and the StarHub and M1 joint venture chose Nokia. CNBC wrote that Huawei lost out to those two on the main networks but might stay in the broader ecosystem at a reduced capacity. Singapore's minister said no vendor had been specifically excluded, and the article adds that the US had urged allies to keep Huawei out of 5G (Choudhury, 2020). In November 2020 Xinhua reported that Huawei Technologies Thailand had announced 700 million baht (US$23 million) for a third data centre in 2021 (Xinhua, 2020). Two countries, two different outcomes for Huawei in the same year; the Thai report mentions no US pressure. Neither case is about apps or about Vietnam. I use them only as a reminder that a political branch can arrive as a shrinking role.

Which choices pay in every branch, and which are bets?

The framework's test step asks exactly this. The three labels below are my extension of that question, not terms from Wack or Schoemaker. No-regret: pays in every branch. Bet: pays in only one. Keep-open: cheap, and buys the right to choose again once more is known.

The first decision, separating shared logic from each platform's interface layer, costs VND 250 million. Assume it saves VND 400 million of maintenance over three years whichever branch arrives, so it nets VND 150 million in all three. That makes it a no-regret candidate, but the conclusion hangs on the savings assumption, which the studio must check against its own code.

For the HarmonyOS channel itself, four options exclude one another. Figures in the table are three-year net gains (VND million): channel value times the share captured, minus cost. A pilot captures 25 per cent of channel value. The pilot-then-upgrade path reaches full coverage six months late and forfeits an assumed 10 per cent of that branch's revenue. "Regret" is the best option's result in that branch minus this option's result, taking the maximum; I added this measure myself.

Checking one row: the full app in "Unchanged" is 1,400 less 900, or +500; pilot-then-upgrade is 1,400 × 0.9 = 1,260 less 900, or +360, which is 140 short of building at once. Building now is a bet on "Unchanged". Doing nothing is also a bet, that the channel is worth little. The last option keeps the choice open: it pays 50 in the worst branch and gives up 140 if the middle branch arrives.

It fits if nobody on the team can say which branch will arrive and the money is hard to recover. It does not fit if you are sure of one branch, or if the full app costs only a few weeks of work.

Which signals show which branch is arriving?

A good signal can be read from outside, carries a date, and is tied in advance to an action. The columns for action and the threshold in the last row are assumptions.

One caution: the Cairo OpenLab served enterprise customers and was not a HarmonyOS programme, so it is only a sample of how Huawei builds a partner network in a country.

When does this approach not fit, and what are its limits?

It does not fit when the decision is small or reversible: try and measure, with no need for three stories. Two limits remain.

First, scenarios only cover what the team thought of. That is my inference; the public parts I read do not discuss it. The studio's three branches turn on one US policy. Vietnamese policy, the stance of the two big app stores, or a shock nobody has named has no cell.

Second, a single driver ignores interaction with other drivers. The "Tightened" branch looks different depending on whether Huawei's ecosystem is growing fast or stalling, and the table treats both alike. When a second driver is large enough, go back to the two-axis matrix.

Going deeper: what did I read, and what not?

For Wack (1985) I read only the public introduction to the September Harvard Business Review article (the rest sits behind a paywall): forecasting errors have become more frequent since the early 1970s. For Schoemaker (1995) I read only the abstract at MIT Sloan Management Review: scenarios offset two familiar errors, overconfidence and tunnel vision. So I do not know whether either article sorts decisions by branch, and I do not credit them with the three labels above.

The library's Reuters link (via Yahoo Finance) returned an HTTP 429 error when I opened it, so I read the full text of the Reuters report as republished by The Economic Times on 23 November 2024. I read the full CNBC article (25 June 2020), the Xinhua piece on China Daily and the Cairo OpenLab release; for the 2025 annual report I read the web page, not the PDF. The sources are mostly press or Huawei's own words: they show what the company says, and do not verify it. I have read no official US or Vietnamese government text, and no source on policy after November 2024. The same case through another framework: the VRIO post.