Case Study / Retrospective Validation

$1.75B in.500K out.

Quibi is the cleanest product failure in modern streaming: fully funded, professionally run, dead in six months. We ran its real launch parameters through the notaprompt.in engine, cold, with zero outcome data. This is what it saw coming.

NO HINDSIGHT FED INREAL LAUNCH PARAMETERSMECHANISM-LEVEL
Population structuren = 140 agents
81Oppose
49Neutral
10Support

Representative population structure: color is decision, clustering reflects the social graph's small-world topology. Illustrative layout; proportions track the directional ranges below.

Capital raised
$1.75B
Before a single subscriber
Year-one target
7.4M
Projected paid subscribers
Actually delivered
~500K
6.8% of target
Lifespan
6 mo
Apr 6 → Oct 21, 2020
01The Setup

Fully funded. Professionally run. Failed anyway.

Quibi (“quick bites”) was a mobile-only, short-form streaming service, 7–10 minute episodes for on-the-go viewing. Founded by ex-Disney chairman Jeffrey Katzenberg, led by ex-eBay CEO Meg Whitman, backed with $1.75B from Disney, Sony, WarnerMedia, Goldman Sachs and Alibaba before a single subscriber signed up. [Source: Wikipedia] Not a startup swinging blind, a professionally-run, fully-capitalized bet that still launched April 6, 2020 and announced shutdown exactly six months later. The library sold to Roku for under $100M, against ~$500M in production spend. [Source: CNBC] [Source: Deadline]

~$1.4B
burned in six months, roughly $233M a month across content, marketing and tech.
63 days
from “we're shutting down” back to the ~910K who signed up in the first 72 hours and mostly left.
AUG 2018

Founded as NewTV

Katzenberg (ex-Disney, DreamWorks) starts the project; Whitman (ex-eBay, HPE) joins as CEO. Renamed Quibi that October.

APR 2019 – MAR 2020

$1.75B Raised

Disney, Fox, NBCUniversal, Sony, WarnerMedia, Goldman Sachs, JPMorgan and Alibaba fund the round.

APR 6 2020

Public Launch

iOS/Android only, subscription-only, ~2 weeks into US COVID lockdowns.

OCT 21 2020

Shutdown Announced

Six months in. ~$350M of the $1.75B left; ~$1.4B already spent.

DEC 1 2020

Service Ends

Library sold to Roku for under $100M, against ~$500M in production spend.

02What The Market Already Knew

Eight flaws. Zero hindsight required.

None of these are retrospective. Every one was visible pre-launch or in the first weeks, and none required predicting a pandemic to see.

01PLATFORM

Mobile-only, no TV at launch

iOS/Android app only, no casting, no desktop, no smart-TV apps. A deliberate “mobile-first” bet that inverted the moment lockdown put everyone home, on the couch, in front of a TV.

[Source: IESE Blog]
02TIMING

Launched into a pandemic

The whole value prop was commute-and-transit viewing. COVID erased that behavior in the launch window. Katzenberg later called it a mix of “the idea being less than perfect… and the environment we found ourselves in.”

[Source: Deadline]
03DISCOVERY

No screenshots, no sharing, no clips

Quibi explicitly blocked sharing to social. In a market where discovery runs on shareable clips, this cut the primary adoption loop at the root.

[Source: IESE Blog / Slashgear]
04PRICING

No free tier: a hard paywall

90-day trial, then $4.99 (ads) or $7.99 (none). No permanent free option, competing head-on with YouTube and TikTok, both free and infinite.

[Source: NBC News]
05PRODUCT

Turnstyle: a feature nobody asked for

Auto-reframing portrait↔landscape burned real R&D and drew IP litigation with Eko, money that could have funded sharing or a free tier.

[Source: Mynameiskalam]
06BUDGET

$600M content, no organic loop

A-list production (Hart, J.Lo, Spielberg) with no viral mechanism to amplify it. Only $63M of marketing ran before shutdown, paid reach with nowhere to compound.

[Source: Forbes / Digiday]
07LEADERSHIP

Institutional overconfidence

Analyst Peter Csathy said pre-launch he was “always skeptical about Quibi's chance.” The $1.75B raise was read as validation of the strategy, not runway to test it.

[Source: NBC News]
08CONVERSION

The trial cliff

~910K signed up in the first 72 hours; only ~8% converted to paid per Sensor Tower (Quibi disputed a higher 27% from Antenna). Either way, most tried it and walked.

[Source: Variety]
03How We Fed It Into The Engine

A utility function, not a vibe check.

Every agent evaluates the decision through the same function. Personas aren't assigned: they emerge bottom-up from eight empirically grounded trait distributions (risk & loss aversion, social conformity, collectivism, institutional trust, status-quo bias, budget sensitivity, information processing), each calibrated against published behavioral research (Barsky 1997; Tversky & Kahneman 1992; Hofstede WVS; Cialdini).

simulation_v3.py / compute_utility()
# the single equation every one of the 200 agents runs each step
U = αeff·trust_adjusted_value − βeff·risk_exposure  − λ·loss_aversion·perceived_loss + γeff·social_signal
trust_adjusted_value = value_delta × (0.4 + 0.6 × institutional_trust)
effective_social_weight = social_conformity × (1.0 + collectivism × 0.3)
stance_new = (1 − dampening) × tanh(U) + dampening × stance_old

Quibi's constraints → parameters

We didn't hand-tune numbers to force a failure. Each parameter is a direct translation of a documented, real-world constraint: the same translation an analyst would do for any product before launch.

No sharing / screenshots disabled. effective_social_weight collapses to 0.05–0.10 for non-Influencers: the peer-cascade mechanism has no channel to run through.

90-day trial, hard paywall, no free tier. perceived_loss ≈ 0.50, risk_exposure ≈ 0.65: paying for an unproven habit feels like a loss users can dodge by not converting.

Mobile-only during household lockdown. risk_exposure compounds: the product is built for a context (commuting) that temporarily doesn't exist.

No creator seeding, mass-ad launch. no seed personas: all 200 agents start near-zero stance. The cascade needs a spark; Quibi provided none.

New platform vs. free incumbents. institutional_trust floor lowered to 0.25: even trusting agents credit only ~73% of the stated value.

SCENARIO_PARAMETERS
value_delta≈ 0.35
risk_exposure≈ 0.65
perceived_loss≈ 0.50
trust_floor0.25
social_weight0.05–0.10
seed_personasnone
networkWS · n=200 · k=8
run length30–60 steps
PERSONAS IN POPULATION
InfluencerEarly AdopterPrice HawkPragmatistSocial FollowerHerd MemberSkepticLaggard
04What The Model Structurally Predicted

Opposition wins, structurally, not by chance.

These are directional ranges from applying Quibi's real constraints to the engine's formulas, not a single logged run we're citing as fact. We show the range honestly, then line it up against what actually happened.

Model predicted adoption
5–10%
of the population ever reaches a support stance
Quibi actually delivered
6.8%
of the 7.4M year-one target (~500K subs)
AGGREGATE DISTRIBUTION · STEP 30–60 · DIRECTIONAL RANGE
Consensus: −0.30 to −0.50Support: ~5–10%Oppose: 50–65%
PERSONA-LEVEL DIRECTIONAL STANCE · RANGE ON −1 … +1
Influencer

High centrality overrides constraints, but with no seeded stance, starts neutral and slow to move.

Weak supportn ≈ 2.5–4%
Early Adopter

Low risk aversion and prior-adoption history help, but the loss term keeps utility thin.

Marginal supportn ≈ 7.5–10%
Price Hawk

Waits for a discount that never comes; risk_exposure and perceived_loss dominate value_delta.

Neutral / mild opposen ≈ 6–9%
Pragmatist

The balanced-trait middle of the population. Utility nets negative once loss and risk are summed.

Oppose / neutraln ≈ 20–30%
Social Follower

Their entire adoption mechanism, peer cascade, is severed by the no-sharing constraint.

Oppose / neutraln ≈ 10–15%
Herd Member

Requires broad consensus to flip. Consensus never forms, so they stay opposed the whole run.

Opposen ≈ 12.5–17.5%
Skeptic

Low institutional trust discounts the value claim by up to 48% before loss and risk even apply.

Opposen ≈ 10–15%
Laggard

Extreme risk aversion plus high status-quo bias. Structurally immovable within the window.

Strong opposen ≈ 5–7.5%
05Reality Check

The model matched, every row.

5/5
MetricModel's directional predictionQuibi's actual outcomeMatch
Adoption rate5–10% of population reaches support6.8% of target (500K / 7.4M subs)MATCH
Opposition dominance50–65% of agents land oppose~93% never converted past trialMATCH
Trial-to-paid conversionWeak cascade, no sustained support signal8–27% converted, then plateauedMATCH
Dominant stance clusterPragmatists, Social Followers, Herd Members: neutral-to-opposeMost “adopters” were one-time trial takersMATCH
Cascade / seeding boostMinimal: no seeded personas, cold startNo KOL launch; mass TV ads onlyMATCH

Directional ranges, not point estimates from a single logged run, derived from applying Quibi's documented launch constraints to the engine's formulas. [Source: Variety] [Source: CNBC]

06The Mechanism, Explained Simply

Loss aversion did the heavy lifting.

01
−λ · loss_aversion · perceived_loss

Loss aversion outweighs value, by design

With λ≈2.25 (the median agent, per Tversky & Kahneman) and perceived_loss≈0.50, the loss penalty alone is ≈ −0.56, bigger than the entire value term (≈0.23). Losses loom ~2.4× larger than gains. Prospect theory operating exactly as documented, not a special case invented for Quibi.

02
value_delta × (0.4 + 0.6·trust)

The trust discount hits at the worst moment

Even a high-trust agent (0.8) credits just 88% of Quibi's stated value; a skeptic (0.2) credits 52%. The loss term has no equivalent discount, so eroding trust shrinks the one thing working in Quibi's favor while leaving the thing working against it untouched.

03
social_signal = Σ w·neighbor_stance

Social signal collapses without a sharing channel

The signal is a weighted average of a neighbor's stances. If every neighbor is independently computing negative utility, the average stays near zero, no positive feedback loop forms. TikTok and YouTube have that loop by default; Quibi's no-screenshot policy removed it at the source.

04
threshold = status_quo × adoption_discount

The decision threshold is a structural trap

With most agents not “practiced adopters,” threshold lands near 0.54–0.65; stance must clear ≈±0.19 to register as support or oppose. Weak signals in both directions stay trapped in a wide neutral band, matching the observed pattern: many trial signups, little conviction.

07What Would Have Had To Change

Five levers. Quibi pulled none.

30–40%
Flip these five parameters to a different set of launch decisions and the model's output moves from opposition-dominant to a projected 30–40% adoption with positive consensus.
value_delta0.350.70Content perceived as clearly superior to free YouTube/Netflix originals, which didn't materialize.
perceived_loss0.500.10A permanent free / ad-supported tier or money-back guarantee, instead of a hard 90-day wall.
effective_social_weight0.050.60Clips, screenshots, in-app sharing: the discovery loop Quibi explicitly disabled.
risk_exposure0.650.30TV / casting support shipped before launch, resolving the COVID context mismatch.
seed_personasnoneInfluencer + EAA KOL / creator seeding strategy pre-launch, instead of relying on mass marketing spend.
We asked people to pay for it before they understood what it was.

“I think we thought there would be easier adoption by people to it.” The engine priced that gap in before launch, as a loss term, not a marketing problem.

JEFFREY KATZENBERG · CO-FOUNDER · OCT 2020
[Source: CNBC]
You already know how this one ends

Now run the one
you don't.

Feed in your real launch parameters: pricing, platform, discovery, seeding, and see where the utility mechanics land before you spend the budget.

[ REFERENCES ]
Wikipedia: QuibiCNBC: Quibi to shut down after 6 monthsNBC News: Why Quibi failed so soonVariety: Year-one subscriber goal paceVariety: Free trial 8% conversionDeadline: Katzenberg / Whitman interviewIESE Blog: The death of a six-month-old appSlashgear: Why Quibi was such a failureForbes: Quibi ad spendDigiday: Content strategy budgetANTENNA Analytics: Trial conversion ratesMynameiskalam: Quibi launch failure case study