Product Portfolio Theory

Learning goal: Treat products as bets, compute Expected Value and variance, and design a portfolio for 160 hours a month using small bets, the barbell, Stage-Gate, kill criteria set in advance and WIP limits.

As of: 2026-09-29. All probabilities and amounts are illustrative assumptions.

Key Concepts

Principles

Expected Value and Variance

Expected Value = Σ (probability of an outcome × value of the outcome)

Suppose the same 480 hours (160 hours a month × 3 months) are spent in two ways. The value is monthly net revenue on success (all assumptions).

How it is computed (in units of KRW 10,000):

• Big bet variance = 0.1 × (300 − 30)² + 0.9 × (0 − 30)² = 7,290 + 810 = 8,100 → standard deviation 90
• Variance of one small bet = 0.15 × (50 − 7.5)² + 0.85 × (0 − 7.5)² = 270.94 + 47.81 = 318.75; six together 1,912.5 → standard deviation about 43.7
• At least one success = 1 − 0.85⁶ = 1 − 0.3771 = 0.6229

How to read it: under these assumptions small bets have both a higher expected value and less scatter. But the key point is that the probabilities and payoff sizes are assumptions. If each small bet had a 5% success probability, the expected value would be 6 × 0.05 × 50 = 15 (KRW 150,000), lower than the big bet. The purpose of this table is not the conclusion but the habit of writing numbers down and comparing them.

Small Bets and the Barbell

Daniel Vassallo said "Don't build a product. Build a portfolio of small bets," recommending that you only bet more once you see a signal (2020). Taleb's barbell strategy keeps one side very safe and takes large risks only on the other side, avoiding the middling middle.

Translated to a solo studio:

Stage-Gate and Kill Criteria Set in Advance

Cooper's Stage-Gate is a model for new-product development in large companies, but its core idea, "decide Go/Kill at each gate against criteria set in advance", carries over directly to a solo studio. The threshold numbers below are examples (assumptions); 03 and 09 set them again per product.

flowchart LR
    I[Idea] --> G1{G1 problem check: at least 3 of 5 interviewees faced it recently}
    G1 -->|Go| V[Demand validation, max 16 hours]
    G1 -->|Kill| K[Archive]
    V --> G2{G2 demand signal: 10% sign-ups from 300 visitors or 5 pre-sales}
    G2 -->|Go| B[Build, max 80 hours]
    G2 -->|Kill| K
    B --> G3{G3 launch readiness: core flow and payments connected}
    G3 -->|Go| L[Launch]
    G3 -->|Recycle| B
    L --> G4{G4 at 60 days: monthly net revenue KRW 100k or 100 weekly actives}
    G4 -->|grow| S[Keep in operating slot and invest]
    G4 -->|Kill| K

Criteria are written before starting and not changed after seeing the results. Criteria changed after the fact are rationalization, not criteria.

The Sunk-Cost Fallacy

Arkes and Blumer (1985) showed, among other studies, in a field experiment that people who paid more for a theater season ticket attended more plays over the following months than people who received discounts: money already spent drags later decisions along. In products it looks like "I've already spent 200 hours, just a little more."

• Bad example: "I've already spent 200 hours on this app, I can't throw it away. Let's add three more features."
• Good example: "What changes if I spend 40 more hours (KRW 875,000 worth)? Is there evidence it can reach the G4 criteria? If not, move it to the archive and write down what I learned."

WIP Limits and Little's Law

Little's Law states that in a stable flow, average lead time = average WIP ÷ throughput. Kanban uses this relationship to limit the number of items in progress.

Example studio: WIP 12, throughput 0.5 launches per month (assumption) → lead time = 12 ÷ 0.5 = 24 months Cut WIP to 2 with the same throughput → 2 ÷ 0.5 = 4 months

In practice, reducing WIP often raises throughput too, because there is less context switching. Suggested starting point (assumption): at most 2 in build, at most 2 in validation, and products in operation managed by an hour budget rather than a count.

Applied: The Example Studio

Hour allocation (160 hours a month, assumption)

This table is the track's default allocation. The 90-day plan in 10 starts from it and lists every deviation with its reason.

Sorting the 12 prototypes (assumption)

The archive is not deletion. Tidy the repository, write one line on "the condition for taking it back out", and leave it alone. The launched P1 sits in the operating slot. P1 shipped before the G4 gate existed and so had no criteria set in advance; document 10 writes its first date and state, using the same numbers as G4.

Going Deeper

Correlation within the portfolio. As with financial portfolios, products that move together lose the benefit of diversification. If four products all depend on Google search traffic and AdSense, one change in the search algorithm shakes all of them at once. The axes to diversify are acquisition channel (search, stores, communities), revenue model (ads, sales, subscriptions) and platform (web, iOS, Android, Steam). Look at the fit table in 04 again from this angle, and use the checklist in 04's Going Deeper to score the example studio's three-path combination.

Validation buys information. If 16 hours of validation can avoid 80 hours of building, those 16 hours (KRW 350,000 worth) are insurance that can save up to 64 hours (KRW 1,400,000 worth).

The Portfolio's Shared Asset: One Audience

Products can be archived or killed, but the people who found them can stay with the studio. To avoid gathering the first 100 users (08) from scratch for every bet, design the audience as an asset of the studio rather than of one product.

• Consent: collect consent for the waitlist and consent to receive promotional messages (newsletter, new-product notices) separately. See the Business section's "Terms · Privacy · E-Commerce".
• Correlation: a shared audience makes products share an acquisition channel. It is an asset and a common risk at once, so measure how much with the checklist in 04's Going Deeper.

What the 16 hours a month of "shared distribution" in 10 contain (assumption)

Common Misconceptions

• "Small bets always beat big bets" — It flips depending on probabilities and payoff sizes. Write the numbers down and compare.
• "The barbell means splitting half and half" — The point is not the ratio but a structure that avoids middling risk.
• "Kill criteria can be set after seeing results" — Criteria set after the results are rationalization.
• "Running many things at once finishes them sooner" — By Little's Law, the more WIP, the longer each product's lead time.
• "Moving something to the archive is failure" — An explicit stop is a decision that wins back time.

Self-Check Questions

1. If each of the six small bets had an 8% success probability, what would the monthly Expected Value and the "chance at least one succeeds" be?
2. Sort your current products into the barbell's safe side, aggressive side and middle to avoid.
3. For one product in progress, what should its G4 criteria have been if you had written them before starting?
4. Compute your lead time from your current WIP and monthly launch throughput.
5. Which acquisition channel do your products share, and how many would be hit at once if that channel closed?

References

• Daniel Vassallo, "Don't build a product. Build a portfolio of small bets." — X (2020-05, accessed 2026-09-28)
• Daniel Vassallo — dvassallo.com (accessed 2026-09-28)
• Antifragile (book) — Wikipedia (Taleb, 2012, accessed 2026-09-28)
• The Stage-Gate Model: An Overview — Stage-Gate International (accessed 2026-09-28)
• The psychology of sunk cost — Arkes & Blumer, Organizational Behavior and Human Decision Processes 35(1) (1985, accessed 2026-09-28)
• Stable Systems: Little's Law and Kanban — Nave (accessed 2026-09-28)