Creative Desperation: The Insights That Only Show Up When You’re Cornered

A corner changes the price of your assumptions. The option you refused to look at last quarter becomes cheap to look at, because whatever that option threatened has already been taken from you.

Gary Klein built a triple path model of insight from 120 case studies. Two of the paths run on new input: you spot a connection, or you spot a contradiction. The third runs on subtraction, and he calls it creative desperation. You get trapped, you discard the assumption holding the trap shut, and the way out turns out to have been there the whole time.¹

So the useful question is mechanical. Constraint does not sharpen the judgement you apply to your options. It changes which options get written down at all.

Why these three models

The Wire Model scores the features of a decision, routes them to a small ensemble of formal models, then forces the ensemble to produce dated actions. Four features carried this one:

• Path dependence in the option set (0.85). What you generate is a function of where you already stand, and you have been standing there a while.
• Constrained allocation across multiple fronts (0.9). Same customers, segments and channels as last year. Less cash to hold them with.
• An absorbing barrier (0.8). Runway ends, a licence lapses, the anchor customer leaves. Some states have no return path.
• Attentional distortion under scarcity (0.7). Scarcity pulls people deep into one problem while other problems go unwatched.²

Those route to rugged-landscape search (where options come from), Colonel Blotto (which options can win on a smaller budget), and bold play under an absorbing barrier (what shape the bet should take). Complex, equilibrium and random outcome types, so their errors point in three different directions.

The behavioural layer is folded into the cards and the levers instead of running as a fourth model. Scarcity, fixation and deadline pressure all move parameters inside these three. They point at no separate action of their own, and a fourth card would cost the argument a hundred words for nothing.

The framework: constraint edits the menu before it edits the choice

1. The option set: you have been searching your own neighbourhood

Levinthal modelled organisational adaptation as search on a fitness landscape whose height is performance. Where attributes interact, the landscape is rugged, and a rugged landscape carries many local optima. Organisations that search their immediate neighbourhood walk uphill until they hit one, then stop.³

The model runs a second process alongside that one, the long jump: a move to a distant form rather than an adjacent one. The timing is the part worth stealing. Long jumps happen often early, while current fitness is modest, and become rare as fitness rises.³

Read that as an option-generation rule and the corner stops being mystical. A business performing acceptably generates adjacent options, because adjacent is all it needs and distant risks a peak it already holds. Lose the anchor customer and the height of your current peak collapses, which is the one condition under which a jump pays. The option was available last quarter. What changed is the price of considering it.

Klein’s creative desperation is that same move described from inside the head. Something gets treated as fixed. The exit requires it to be dropped. Pressure makes the dropping affordable.¹

Your anchor is almost always a sentence about distribution. “We sell through the distributor.” “We need the warehouse.” “Orders come through the app.” Each one has a neighbourhood around it and a set of moves outside it: forty shops ordering on WhatsApp with mobile money on delivery, fulfilment handed to a boda network paid per drop, a signed LPO discounted at the bank instead of a raise.

Rugged-landscape search, the generation lens

Assumes: performance is a landscape with many peaks, and search is local by default.

Fits because: path dependence in the option set scored 0.85.

Breaks when: the landscape is smooth. If your problem is execution on a known form, a jump destroys the peak you hold and buys nothing.

Counteracts: the belief that the answer is a harder version of what you already do.

May reinforce: serial pivoting, where every hard quarter gets treated as evidence that the form is wrong.

2. The fronts: a smaller budget buys a different shape, not a smaller copy

Colonel Blotto is the formal version of your allocation problem. Two players spread fixed budgets across several fronts simultaneously, each trying to win the most fronts. What the weaker player should do depends entirely on how a single front gets won.

Under an auction rule, where whoever commits more to a front takes it outright, equilibrium sends the disadvantaged player to guerrilla warfare: allocate zero to a random subset of fronts and fight properly on the rest. The advantaged player does the reverse, covering every front with random positive amounts. Under a lottery rule, where your chance of taking a front is your share of what both sides commit there, both players divide their budgets evenly across every front. An experiment with asymmetric budgets reproduced both patterns and priced the gap between the two rules: the disadvantaged player’s equilibrium payoff was 0.3 under the auction rule and 0.375 under the lottery rule.⁴

Classify your fronts before you cut any of them. Auction-shaped fronts settle winner-take-all at the moment of decision: a tender, a distributor exclusivity, a shelf category, an enterprise RFP with one winner. Lottery-shaped fronts pay in proportion to presence: share of wallet in an open market, an agent network, reorders over WhatsApp, a thousand small buyers.

On an auction front, a cornered founder who keeps covering everything loses everything, slowly and politely. Abandon fronts, stay unpredictable about which ones, and put real weight where you remain. On a lottery front the instruction inverts. Thin even coverage is the equilibrium, and cutting fronts to look decisive throws away the only shape that pays.

The payoff gap carries a second instruction. The weak player does measurably better in proportional contests, so when you hold the smaller budget, move revenue toward the contest shape that pays for presence. Volume orders ahead of tenders. Repeat purchase ahead of annual bids.

Colonel Blotto, the allocation lens

Assumes: budgets are fixed, fronts are contested simultaneously, and the win rule per front is known.

Fits because: constrained allocation across fronts scored 0.9.

Breaks when: the number of fronts is yours to choose, or a front’s win rule changes mid-quarter, as when a tender turns into a framework agreement.

Counteracts: proportional across-the-board cuts applied to every line at once.

May reinforce: premature abandonment of markets that were paying in proportion all along.

3. The bet shape: below the barrier you stop maximising the average

Take a gambler holding a fortune below a goal, playing a game with the odds against him, who has to reach that goal. The established result is that he should play boldly, staking as much as the goal allows on each attempt rather than grinding out small bets.⁵ Small repeated bets in a negative-expectation game are a slow certain loss. Concentration is the only thing that puts probability mass on the far side of the target.

The same paper adds a qualification worth more than the theorem. Introduce a discount factor, so that elapsed time itself costs something, and bold play stops being guaranteed optimal.⁵ Your discount factor is the engineer who resigns in month three and the licence that lapses in month six.

Then the empirical half, which is unkind to the myth. Across professional tournaments, risk taking traces an inverted U against interim rank. Competitors trailing just behind the leaders take the most risk. Those far off the pace take less.⁶ Being cornered does not make you bold. Past a point it makes you careful, at the exact moment caution is the losing move.

Boldness under pressure is a design decision rather than a mood. Which is why the levers below get written down instead of felt.

Bold play under an absorbing barrier, the variance lens

Assumes: a fixed target, a real absorbing state, and odds currently against you.

Fits because: the absorbing barrier scored 0.8 and expected value stops being the right objective below it.

Breaks when: the game is favourable. With a compounding positive-expectation channel, small steady bets win and concentration is vandalism.

Counteracts: grinding on small reversible experiments while the cash runs out.

May reinforce: single-bet thinking, and treating any large irreversible move as courage.

GEER: the levers, ordered by what they cost to undo

Four channels carry this outcome: the anchor you hold fixed, the number of fronts you fund, the shape of the bet, and the clock. Take the free ones first.

1. Write the anchor sentence. “This business only works if …” One line, on paper. You cannot drop an assumption you have never named. Free.
2. Date the barrier. Cash on hand over net monthly burn, resolved to a calendar day. A vague crisis produces vague options. Free.
3. Build the front ledger. Every segment, channel, product and geography, each with the cash it consumed last month. One hour.
4. Label every front auction or lottery. The label decides whether cutting helps or hurts. One hour.
5. Generate three options that break the anchor. Three, costed, rather than thirty. In a longitudinal study of option generation, athletes across expertise levels produced the same number of options on average, and what separated them was the quality of the options, including the first one they named.⁷ Volume is not the lever.
6. Buy the gap properly. Setting a problem aside raises solution rates, and filling that gap with a cognitively demanding task shrinks the benefit.⁸ So the block is low-demand work. Deliveries, collections, a long drive. Not the investor update.
7. Hand the deadline to somebody else. People do impose costly deadlines on themselves, and those deadlines do improve performance, and they are set worse than externally imposed ones.⁹ Give a board member or a customer the date.
8. Supply the cover that pressure needs. Creative time pressure tracks creativity in an inverted U, and in the measured case the shape appeared only among employees high in openness who also had support for creativity around them.¹⁰ Pressure without cover buys compliance. Costs a standing block and explicit permission to propose the disloyal option.

No-lever flag: if the barrier sits inside 60 days and no anchor-breaking option has a paying counterparty, this has stopped being an option-generation problem. It is a cash problem, and the only live lever is the fastest revenue you can invoice.

RADAR: the portfolio, with dates on it

DO NOW, by T+3. Every item here is reversible and pays under any version of the next month.

1. Anchor sentence, barrier date, front ledger, auction or lottery label per front. Four artefacts, one page.
2. Three anchor-breaking options, each costed to first cash, each with a named counterparty who could say yes this month.
3. One external deadline holder, given the date in writing.

HEDGE, by T+14. Cheap insurance against the jump failing.

1. Open one non-dilutive line before you need it. Debt made up almost half of total capital raised by African tech companies in 2025, and access to it runs on cash flow visibility and governance you have to build in advance.¹¹
2. Keep one lottery front alive at minimum spend. It is your standing evidence that demand exists outside the anchor.
3. Book one 48-hour low-demand block between generating the options and choosing one.

DEFER AND TRIGGER. Irreversible, so pre-commit the observable now.

1. Defer: shutting a product line, releasing the team, announcing the pivot publicly, signing a bridge at a punitive price.
2. Trigger to jump: two of the three anchor-breaking options produce a signed test by T+21, meaning a paid pilot, a deposit or an LPO. Execute at T+28, and cut the abandoned auction fronts in a single move rather than three.
3. Counter-trigger: zero signed tests by T+28 with the barrier inside 90 days. Stop generating. Take the fastest invoiceable revenue in reach and rebuild runway before touching the anchor again.

Holding a cornered company in a portfolio. DO NOW: ask for the anchor sentence and the barrier date in writing, because the absence of either is the real signal. HEDGE: fund the three costed options at test size rather than funding the pivot at conviction size. DEFER: the recapitalisation conversation until one signed test exists, since a founder below the barrier accepts terms they will resent for years, and resentment prices into the next round.

CHAIN: the base rate for people in a corner

The precedents that matter share a skeleton: an actor below a threshold, a deadline it does not control, and a shrinking budget spread across fronts it cannot all hold. Tournament competitors trailing the leader sit in that class. So do organisations walking a rugged landscape, and gamblers who need a number before closing time. Sector tells you nothing here.

The base rate across that class is the unromantic one. Most actors below the threshold narrow rather than open. Risk taking falls at the back of the field,⁶ and long jumps grow rarer the longer a form has been performing.³ The insight-under-pressure story survives mostly because the cases that produced it got written up.

Second order: cutting auction fronts releases cash and, worth more, releases the attention that scarcity had already tunnelled.² Third order: the fronts you abandon are cheap for a better-funded competitor to occupy, and the market reads your retreat as a verdict on the segment rather than on your budget. Price that in before you announce it.

Before crediting the corner with the outcome, remove the cases where the option had already been proposed by somebody junior and refused for reasons that no longer apply. In most companies that is the majority of them. The corner generated nothing. It removed the objection.

Matrix-break flag. The barrier itself is moving. Non-dilutive capital now runs close to half of what African tech raises in a year,¹¹ which means a founder with visible cash flow can push the absorbing date out instead of betting the company against it. Where that rail is open, bold play stops dominating and the patient strategy becomes rational again. Check whether the rail is open for you before you accept the premise that you have to jump.

What this ensemble cannot see

Three of these gaps you can price. The fourth you cannot.

Whether the anchor is wrong or merely uncomfortable. The models tell you a jump is available and what it should cost. None of them tells you the distant peak is higher until you are standing on it.

Where you sit on the rank curve. Behaviour diverges sharply between trailing just behind and being far off the pace,⁶ and from inside a business the two states feel identical. Ask your external deadline holder which one they see.

The counterparty’s private clock. The customer’s budget cycle, the regulator’s queue, the distributor’s own cash problem. Your barrier is rarely the only barrier in the deal.

The option nobody wrote down. Every study cited here counts options that were generated. The ones that never reached a page leave no trace, and they are the majority. More thinking does not close that gap, so close it structurally: put one person in the room whose only job for that hour is to propose the move that costs somebody their remit.

Then decide on the artefacts rather than the adrenaline. Anchor sentence, barrier date, front ledger, three costed options and one external deadline holder by T+3. Signed tests or no signed tests by T+28. Jump on the evidence.

Sources and notes

1. Klein, G. “Triple Path Model of Insight.” Author’s own site, describing the model introduced in his 2013 book “Seeing What Others Don’t” after examining 120 case studies of human insight, with the three paths given as contradictions, connections and creative desperation. Model page.
2. Shah, A. K., Mullainathan, S., and Shafir, E. “Some Consequences of Having Too Little.” Science 338(6107), 2012, 682-685. Scarcity shifts attention, leading people to engage more deeply in some problems while neglecting others. Record and abstract.
3. Levinthal, D. A. “Adaptation on Rugged Landscapes.” Management Science 43(7), 1997, 934-950. Local search terminates at a local optimum, the number of local optima rises with ruggedness, and long jumps are frequent early and rare once fitness is high. Full text on a university course mirror. The link is http only because the publisher page refuses automated access, so treat it as the readable copy rather than the canonical one.
4. Chowdhury, S. M., Kovenock, D., and Sheremeta, R. M. “An Experimental Investigation of Colonel Blotto Games.” Economic Theory 52(3), 833-861. Working paper dated 16 September 2011. Guerrilla warfare by the disadvantaged player under the auction rule, stochastic complete coverage by the advantaged player, equal division by both under the lottery rule, and equilibrium payoffs of 0.7 and 0.3 (auction) against 0.625 and 0.375 (lottery). Full text.
5. Chen, R. “Subfair Primitive Casino with a Discount Factor.” Zeitschrift fur Wahrscheinlichkeitstheorie und verwandte Gebiete 39, 1977, 167-174. The paper opens by restating the known result that a gambler below a goal in a subfair primitive casino should play boldly, then shows the bold strategy need not stay optimal once a discount factor prices elapsed time. Full text.
6. Genakos, C., and Pagliero, M. “Interim Rank, Risk Taking and Performance in Dynamic Tournaments.” Centre for Economic Performance Discussion Paper 928, 2009, revised July 2012. Risk taking follows an inverted U against interim rank, with competitors trailing just behind the leaders taking the greatest risks. Full text.
7. Raab, M., and Johnson, J. G. “Expertise-based differences in search and option-generation strategies.” Journal of Experimental Psychology: Applied 13(3), 2007, 158-170. Two-year longitudinal study: athletes of varying expertise generated the same number of options on average, and the options differed in quality for both initial and final choices. Record and abstract.
8. Sio, U. N., and Ormerod, T. C. “Does incubation enhance problem solving? A meta-analytic review.” Psychological Bulletin 135(1), 2009, 94-120. Positive incubation effect overall, larger with longer preparation, smaller when the incubation period is filled with high cognitive demand tasks. Record and abstract.
9. Ariely, D., and Wertenbroch, K. “Procrastination, Deadlines, and Performance: Self-Control by Precommitment.” Psychological Science 13(3), 2002, 219-224. People self-impose costly deadlines and those deadlines improve performance, but they are set less effectively than externally imposed ones. Record and abstract.
10. Baer, M., and Oldham, G. R. “The curvilinear relation between experienced creative time pressure and creativity: moderating effects of openness to experience and support for creativity.” Journal of Applied Psychology 91(4), 2006, 963-970. Data from 170 employees and 10 supervisors; the inverted U appeared for employees scoring high on openness while also receiving support for creativity. Record and abstract.
11. Partech, 2025 Africa Tech Venture Capital Report. Debt financing grew sharply past the 2021 peak and made up almost half of total capital raised by African tech companies, while the report flags persistent pressure at the pre-Seed and Seed stages. Report page.