All insights Strategy May 6, 2026 9 min read

Translated from the Vietnamese original

LTV must exceed CAC. Don’t scale until those two numbers work. The advice is right. But it can be completely right and completely irrelevant at the same time, depending on what kind of business you run.

Three games, not three stages

There are three ways to play unit economics. Not three stages of evolution, but three different games, with three sets of decision criteria that cannot be swapped.

Level 1: transaction

Each transaction: profit or loss on every order.

Level 2: customer

The customer lifetime: LTV versus CAC, the cost of keeping a customer versus the cost of acquiring one.

Level 3: system

The whole system: value is created not in each unit, but in how units interact over time, and, more importantly, whether the system has a self-reinforcing mechanism.

Most people stop at Level 2 and call it strategy. That isn’t wrong. But it skips something more important: what kind of business you are running, and how that game lets you read the numbers.

These levels don’t depend on the industry. They depend on the strategic goal and the structure the owner chooses to build, or drifts into.

F&B chains, fashion retail and private clinics are mostly Level 1: they must win on each transaction, because their structure has no mechanism that improves itself as it grows. Grab, Shopee and logistics platforms are designed to play Level 3 from the start, accepting losses at Levels 1 and 2 because the winning mechanism sits in the whole system. Level 2 is the game for businesses with long enough customer loops: management software (SaaS), insurance, education, any model where keeping a customer costs much less than acquiring a new one.

The core difference between the games: Level 1 must have positive unit economics from the start. Levels 2 and 3 allow negative unit economics today, but only with a clear calculation of when they will turn: for Level 2, when LTV overtakes CAC after a long enough period; for Level 3, when the whole ecosystem reaches its self-reinforcing threshold. This isn’t “accepting losses” in the usual sense. It is a calculated bet on an economic structure that will form in the future. The difference between a smart bet and losing money for no reason lies right there: can you calculate the turning point, and do you have the resources to survive until then?

The problem isn’t which level you are at. It is whether you are playing that level’s game correctly.

Retail: no mechanism that improves with size

Retail has to get unit economics right from the start, not because of a rule, but because its structure has no mechanism that improves as it grows. No network effects. No data advantage that compounds with scale. No natural switching cost strong enough to keep customers without continued investment.

That doesn’t mean every retail chain fails when it scales. It means advantage has to be built deliberately: bargaining power with suppliers, standardized operations, store density high enough to create a logistics advantage in a specific area. Without these, store number 50 is not cheaper than store number 5 by any natural mechanism.

Scaling in retail isn’t growth. It amplifies whatever is already there, mistakes included.

Platforms: read the other way round

The platform game reads the other way round. Shopee or Grab in their early days: high CAC, heavy losses, poor unit economics at both Level 1 and Level 2. But this is often simplified into “accepting losses to buy market share,” and that is the wrong reading.

The real logic: these systems are designed to play at Level 3 from the start. And two conditions must both hold for early losses to make economic sense.

1. First: are the network effects real?Does each new user make the system better for existing users, through a specific mechanism rather than a hope?
2. Second: is the multi-homing cost high enough?Multi-homing cost is the cost for users of using several similar platforms at once. If it is low, network effects don’t create a lasting advantage, however large the scale. Users simply use all of them. That is why GrabRewards exists: not only to raise frequency, but to raise switching costs through behavioral lock-in.

If these two can’t be proven, the losses are not a strategic investment. They are the wrong structure with a better story.

Vingroup and the logic of cross-subsidy

Last week was Vingroup’s shareholder meeting. VIC hit its daily limit-up for several sessions in a row while VinFast was still losing money. Plenty of people asked: why?

The question is not a paradox. It shows the market reading Vingroup according to the game it is actually playing, Level 3, not Level 1 or Level 2.

Vinhomes has high margins and stable cash flow. That cash is allocated to VinFast to hold a position in a market that offers no second chance if you miss it. This is structured cross-subsidy, not disorganized internal support. Vingroup tried something similar with Vinmart, Vinpro and a series of other retail concepts. Some were closed, some were sold. That isn’t failure in the usual sense; it shows they know how to cut when a mechanism doesn’t form, and to move capital elsewhere. VinFast is being kept for a different logic, not because they don’t know how to cut.

But cross-subsidy only makes economic sense when three conditions hold at once:

1. 1. The market is still taking shape.
2. 2. The subsidized unit has a real path to economies of scale once it is large enough.
3. 3. The subsidizing unit is stable enough to carry the waiting period.

VinFast is gradually proving the second condition: gross margin is improving with clear signals as volume rises, though the road to breakeven is still long. The third depends on the whole group, not VinFast. The first, whether Southeast Asia’s EV market is at a stage where an early position creates an asymmetric advantage, is still an open question. And that is the real risk of this blueprint: not that VinFast loses money, but whether that window for shaping the market is real.

You can’t play one game with the mindset of another

Three games. Three sets of criteria. They can’t be swapped, and you can’t play one game with the mindset of another.

An F&B chain that expands quickly while individual outlets haven’t proven their unit economics usually isn’t run by a founder who doesn’t understand the numbers. They are applying platform logic to a business with no platform mechanism: no network effects, no natural switching costs, no compounding data advantage. Store number ten doesn’t make store number one stronger or cheaper by any natural mechanism. Scaling then isn’t growth; it amplifies the existing flaws at a higher cost of capital.

And here is the part people mention less: each game demands not only different criteria, but different psychological tolerance. Playing the platform game without being able to bear today’s losses for a bet on the future isn’t playing platform. It is drifting between two games without belonging to either.

The real question of unit economics isn’t whether LTV exceeds CAC. It is: which game are you playing, and what decisions do that game’s criteria allow you to make, and forbid you to make?

The businesses caught between two games

Some businesses sit right on the line between two games: chains building CRM, loyalty programs and a data stack, but not yet large enough for a data advantage to kick in, and not yet able to prove switching costs high enough to keep users without continually buying them back with promotions. The easiest sign: the loyalty program has run for six months, but customers still buy out of old habit, not because the system is better. They buy because it is convenient, familiar, or because this week’s voucher beats last week’s. Switching costs haven’t formed; there is only discounting under another name.

They are investing with Level 3 logic but operating with a Level 1 structure. Not because they chose wrong, but because Level 3 mechanisms take time to form, and the market doesn’t yet have enough evidence to know whether they will form here. The question isn’t whether to continue or stop. It is: what evidence shows the mechanism is forming, and do they have the resources to survive until they know?

I don’t know. And there is no sign yet that the answer is forming.

If you aren’t clear which game you are playing, every number you optimize is a local optimum in a problem you haven’t yet defined. And optimizing locally in the wrong problem isn’t progress. It is just going the wrong way with more confidence.