Learning Matrix

Weight total asset control, operational speed and clean exits above a 5% price fight, and size the cost of delay before conceding on any of them.

"Saving basis points while losing 6–12 months is a losing trade."

Run every deal through four passes; reshape the cash-flow structure in the financial pass and infer the counterparty's real wants from behaviour, not their stated position.

"how can the cash-flow structure be reshaped to extract more value?"

Require a short leadership sign-off before any term sheet leaves the internal desk; a walkback after it is external costs both the deal and reputation.

"một bước sign-off ngắn từ lãnh đạo trong 12 giờ trước send"

To back a third party without taking on liability, bind them with a one-way indemnifying commitment; a two-way contract turns support into a partnership.

"cam kết một chiều thì không" [tạo quan hệ hợp tác]

Expect a midday intervention to cut protective terms first and keep structural ones; pre-empt by staging floors and guarantees into a later legal-review step.

"8 of 9 protective terms" were the ones that died

When transfer-at-cost is floated against a lease, set your floor at the NPV-zero price including opportunity cost, then bridge the gap with an earn-out.

"244 bn" just to reach NPV = 0

Read a new counterparty principal introduced mid-session, plus a leadership meeting slotted between your morning send and afternoon reply, as a walkback warning.

"Ms. Ngân introduced as new Director of Investment"

When a counterparty bundles several assets into one thread, use their interest as leverage but keep each asset on its own separate negotiating track.

"use the hook as leverage but keep Hưng Yên separate from the TM-01 MOU"

Screen a business as an acquirer would: predictable cash flow, a fragmented industry, a founder wanting out, and a consolidation or tech-scaling angle at exit.

"A good LBO keeps buyers awake at night thinking of ways to make money"

Score every multi-year deal on NPV at a 12% WACC across three scenarios, and restructure a sub-WACC deal rather than trying to negotiate the headline price up.

"NPV at WACC 12% = -7.3 bn"

Bring a one-line NPV cost for each protective term into the meeting, so cutting one becomes a priced, conscious tradeoff instead of an instinct.

"-25.8 bn in NPV" was the cost of one afternoon

Split a program budget into pillars by financial nature, each with its own disbursement gate, so one pillar can be closed without collapsing the rest.

"705,5 triệu" ring-fenced in its own pillar

Once CAPEX is signed, the operating-fee load is the only IRR lever left; even day-one full occupancy still leaves payback near ten years.

"9.12 years" to payback at full occupancy

An unresolved operational dispute on an entity under valuation becomes a brand-impairment hit or a buyer's forced-discount lever; settle it cheaply first.

"Settling the voucher dispute cheaply protects the headline valuation."

A business is fundable only if it owns something proprietary, defensible and scalable; doing it better than rivals is a service, not a venture.

"own something unique, defensible, and scalable"

Wall off each operating business in its own limited-liability entity, and keep personal trading and estate separate from operating risk.

"If your business is your stuff and your stuff is your business, one bad day could wipe you out."

Never sell appreciating land. Lease it, develop it, or borrow against it, but keep the title and let it compound.

"Never sell the land. Lease it. Develop it. Borrow against it. But never sell it."

Lock measurable fail thresholds with dates and a named triggered action before launch, then re-read them on a fixed cadence so discretion cannot creep back.

"Below 35% on three consecutive cohorts"

For a binary legal risk, disburse a small tranche first to buy a written external Go/No-Go; a No-Go then halts the program with the rest untouched.

"89%" of the spend held behind the gate

Screen any tenant-support proposal against three linkage tests before the meeting; the more it trips, the more you read as a co-operator, not a landlord.

"người cùng kinh doanh" is the line you must not cross

Run the hard legal eligibility conditions against the real asset list before locking KPIs; one condition can silently shrink the whole program's scope.

"hợp đồng mua bán ghi mục đích 'để ở'" quietly excludes units

Verify the signed contract body against the negotiation checklist; the MVillage deal shipped without a money-form performance guarantee and without the setup-fee waiver.

"No money-form Performance Guarantee"

Track the units sliding backward through the funnel, not the fill rate; backward flow caught a one-week reversal that occupancy completely hid.

"21.2% of the entire negotiating pipeline" moved backward

Policy breaches enter at the sales handoff, not the leasing desk; put the control on the file before the tenant signs, because countersignature has no cheap fix.

"The control point is the handoff, not the countersignature."

A kill criterion with a threshold but no named data source cannot fire; name the source that produces each number when you write the rule.

"a threshold with no data feed cannot fire"

Test every risk owner on day one: is the named owner a real person who has agreed, or a role that is still hypothetical?

"is the named owner a person who has agreed, or a role that is still hypothetical"

Inspecting what is easy to see is not diligence; a small physical sample beats a full paper review every time.

"hope wearing inspection's clothes"

A report that counts only footfall and headcount hides a rising cost base against falling demand; add a revenue KPI before you scale staff.

"An operation measured only on motion will optimize motion."

The biggest recoverable loss is signed-but-dark space, not vacancy; occupancy conceals it and the fix needs no new tenant and no price change.

"It is leased space that never opened."

Settle a well-formed amenity-at-sale grievance cheaply before its deadline; public escalation contaminates the next project you market with the same amenity.

"Cost is small; press exposure cost is not."

Deploy capital, IP and relationships so a venture runs without you operating it; if it needs more than a few hours a week, replace the operator rather than compensate.

"the operator is the wrong operator; replace, don't compensate"

Killing a venture does not destroy its IP; carve the reusable asset out of the kill and force a finish, shelve, or abandon decision on it.

"killing the venture does not destroy the IP"

In fragmented Vietnamese service markets the deepest moat is the customer's trusted regular, which price and advertising cannot break.

"60% hộ gia đình ở Việt Nam có ít nhất một 'thợ ruột'"

Pick retail sites by daily foot traffic, not prestige destination pull; frequency and lower rent beat special-trip demand.

"focus on daily foot traffic, not destination visits."

Compete on what big chains cannot copy and let real customer experience replace marketing; a small format can out-earn scale.

"Trải nghiệm khách hàng có thể thay thế marketing."

If income stops when you stop selling, you own a job; enterprise value comes from distribution systems and narrative control, not effort.

"can the founder leave for 60 days without revenue dropping more than 10%?"

Sort your hours into labor, capital, code and media; convert rented time into owned assets that keep paying while you are away.

"stopping it for 6 months doesn't stop the income"

In a slow-to-localize market, first to ship a proven foreign model at scale wins; treat rejection as proof-of-demand data, not a verdict.

"Distribution beats invention in slow markets."

Before committing 6+ months or 100M+ VND, run four premortem passes; the operator-specific passes three and four surface the identity risks you instinctively defend.

"100M+ VND of capital"

Build an explicit assumption book into every board memo, so the approver has to mark which numbers are real and which are guesses before signing.

"self-declared assumption book"

Put both figures on the same basis and baseline before flagging a cost gap; the 9.4bn "unexplained" gap was just VAT plus a stale pre-restaurant estimate.

"internal v4 × 1.10"

If a plan engineered to need almost nothing from you still never opens its first conversation, the bandwidth answer is already no; kill it.

"the plan still did not clear a bar that low"

Sort risks into evidence-backed and anxiety-driven; name the fake ones in the room and refuse to spend meeting time debating them.

"anxious-team risk vs evidence-backed risk"

Cut scope hard across every axis at once; a decision-cut executes for free, while every build needs a person you may not have.

"cuts land, builds do not"

Kill on judgment before the trigger date arrives; a hard deadline makes inaction legible, and a clean early kill beats a slow bleed.

"a clean kill instead of a slow bleed"

Before a high-stakes move, assume it failed 90 days from now, list five concrete causes, fix each one, then score your readiness.

"Assume it failed 90 days from now."

The CRV model's 60M VND/sqm GMV runs about 2.9× the Vinh comparable, so model a 20–40–60M band and anchor to the conservative end.

"800 USD/sqm/year (~20.4M VND)"

For tenant, team and community messages, open with acknowledgement before the ask; the position stays firm while the wrapper stays warm.

"Open with acknowledgement, not the ask."

For board and investor audiences, lead with the number, take a side, and put the math inside the recommendation rather than in an appendix.

"IRR drops from 19% to 14% if we accept the 36-month free rent"

The limit on AI output is prompt specificity, not model power; make the model interview you and draft the prompt before it answers.

"Be my prompt engineer. Clarify my goal, restate it, ask questions, then create the optimized prompt."

For an AI product the moat is the proprietary data and feedback loop it captures, not the model; sell the finished result, not the tool.

"The best AI startup moat is not the model."

When you commission code you cannot read, run an AI engineering audit before and after handoff to catch a cosmetic refactor passed off as real work.

"run #8 to surface whether the refactor was real or cosmetic"

When commissioning software you cannot build, give the developer a named default stack and make them justify every deviation from it.

"hand the developer this stack as the default and make them justify any deviation"

Set context once in a project, start a fresh chat per topic, and specify the output's length and shape before the model starts writing.

"set context once, keep each session narrow, and tell the model the shape of the answer before it starts"

Health inputs follow a dose-response curve with an optimum; ask where the curve peaks before assuming more is better.

"more exercise is not always better"

Visceral fat is an active organ driving disease; for East-Asian men the waist warning line is 90cm, not the US 40 inches.

"waist circumference ≥90 cm (35.5 in) for men is the population-adjusted visceral-fat warning line"

Chronic stress ages you biologically; a daily parasympathetic practice like breathwork or gratitude is the reversing input, and HRV tracks it.

"You don't age from time, you age from your state."