Double down or venture: an art for decision makers only

Every operator I know has sat in the meeting where the revenue line goes flat and the room splits. Half the table says do more of what works, and do it better. The other half says the model is exhausted and we need something new.

The question this piece tries to answer: when the numbers flatten, is the fix more of the same or something new, and who gets to decide? I've been on both sides of the wrong answer. At Eatigo the plateau was arithmetic and the leadership chose more of the same. At JD.ID the leadership chose everything at once. At ZALORA the mission left no room to choose. Two other companies, anonymized here, showed the other two outcomes. In one, doubling down was right and tripled revenue. In the other, nothing was wrong and the cap arrived anyway.

The route: the mechanics of the plateau, with my Eatigo numbers and the churn evidence behind them. Then why the same shape pays 30% EBITDA in one city and nothing in the next. Then the two ways to get the call wrong, and the models that explain why it's a judgment and why it needs power. Then where my argument breaks, and the takeaways.

Eatigo sold restaurant reservations with time-based discounts across six markets, funded by TripAdvisor. I ran Singapore and Indonesia. The public story is yield management for empty tables. The commercial engine underneath was a policy. Two 50% slots a day and 10% all the time, enforced by the country managers, with the 50% slots right before and right after the meal times. It met a lot of resistance from restaurants, and it worked. My experience, not a published figure: the app went from 0 to 60% market share in every market it operated in, in less than nine months.

Then the line went flat, and it stayed flat. Two churns did it. The first is the nature of the business: restaurants close. The second was our own success: restaurants that filled up with walk-ins thanks to us stopped wanting to give 50% at 6 pm. New restaurants came in as a function of sales headcount, churn went out as a percentage of the base, and the two met at a number. It was so arithmetic that I had to call it the Eatigo curse.

The formula is standard in any business that signs accounts and loses a share of them every month. Take an illustrative team of eight salespeople signing ten restaurants a month each, so 80 new restaurants, against 5% monthly churn. The base converges to 80 divided by 0.05, or 1,600 restaurants. At 800 you add 80 and lose 40, and the line looks like growth. At 1,600 you add 80 and lose 80, and every new salesperson is replacing what churned. Double the team and the ceiling doubles, if there are 3,200 restaurants left that fit the model. In our markets there weren't. The fit restaurants ran out long before the formula's ceiling, which is why the plateau arrived at month nine and not at month forty.

The curse, drawn. The sales team signs the same 80 restaurants every month; at 5% monthly churn the active base can never pass 1,600, and half of every month's signings is replacement work well before that. Limits: illustrative numbers built from the ceiling formula, chosen to show the shape. They are not Eatigo's actuals, which I can't publish. Churn isn't constant in real life, and the real base flattened earlier than the formula because the fit restaurants ran out first.

Fit Conditional No fit. The addressable set is the second ceiling: once the fit rows are signed, more salespeople don't raise the plateau.

The shape was identical everywhere. The money was not. In developed cities like Singapore and Hong Kong the plateau meant 7 to 10 million in sales a year at 30% EBITDA. In developing cities like Bangkok, Manila and Mumbai the same plateau meant 1 to 2 million in sales at a 15% negative EBITDA. The marketing and sales the model needed cost more than the plateau paid. Same policy, same slots, same nine months, same flat line. If you plotted the six markets without a Y axis you couldn't tell them apart. Add the Y axis and everything changed. Same system, different financials.

ZALORA taught me the same lesson from the other side. The mission was clear: fashion brands, online, in developing countries. It couldn't do cheap clothes, because the Chinese multi-category marketplaces would have killed it. It couldn't do Shein, because that needs a level of skill in fashion we didn't have and couldn't build. So it executed the Zalando of developing countries, brand by brand, and it executed it well. The Iconic is the same version of the same company, the same playbook, run in Australia, a country completely different from South East Asia and much closer to Europe. Not coincidentally, it's the cash cow. In Global Fashion Group's 2025 accounts The Iconic produced €25.6 million of adjusted EBITDA against €3.0 million from ZALORA and €2.8 million from Dafiti. The group turned its first full-year profit only after a €22 million corporate overhead. Everything identical to Zalando or The Iconic, but when you add the Y axis, the numbers are completely different. Selling international brands to the almost non-existent fashionable, professional ladies of Indonesia gives you the same curve at a fraction of the height.

This decides whether doubling down is even worth the effort. A plateau at 30% EBITDA is a business. A plateau at a loss is a countdown. The machine doesn't know the difference, and I found no way to change the market from inside the machine.

One curve, two Y axes. The same nine-month run to a plateau; only the scale differs. Limits: the curve is a stylized logistic, not plotted from Eatigo data. The sales ranges are my own recollection of the six markets, rounded to the ranges I'm sure of.

Add the cost line and the profit appears, or doesn't. Sales and marketing cost roughly the same to run in both kinds of city, so the dashed line sits at a similar absolute level. The developed-city plateau clears it with 30% to spare; the developing-city plateau stops 15% short of it and stays there. Green is profit, red is loss. Limits: same stylized curve; the cost line is placed to reproduce the margins I remember, not from Eatigo's books.

Profitable plateau Thin Loss-making. GFG figures are segment adjusted EBITDA before €22.1M corporate costs; they show the height of each platform's curve, not its growth.

At Eatigo I saw the plateau early, and so did several colleagues. The founders' team took longer to see it, and I understand why. They had built a model that went from 0 to 60% in nine months. The natural answer to a flat line is to do more of what got you there and do it better. A near-natural law of flattening revenue across six countries wasn't enough to show the way. We tried tasting menus for fine dining, delivery, and vouchers for quick-service chains. But the leadership was divided and, understandably, afraid of cannibalizing a business that had been a miracle nine months earlier. The fear was concrete. A mediocre mid-size restaurant would have asked why it had to give 50% at 6 pm when McDonald's gave 10% at midnight. Good luck explaining "but you are not McDonald's!" It was the time to venture and to accept some failures while looking for the next thing. But that acceptance was not accepted. Keep doing more of what you're doing, keep fighting churn, keep signing sellers against the churn. Then the pandemic arrived while we were still working on the plateau. Lockdowns and a failed fundraise proved hard to overcome, and the old glory of 0 to 60% in nine months never came back.

The research says this is the common failure, not the rare one. Olson, van Bever and Verry at the Corporate Executive Board tracked Fortune-100-size companies over fifty years. 87% had stalled at least once, and management rarely recovered top-line growth once a stall set in. Their book puts 87% of the stalls down to strategy or organization rather than the economy, and only one company in ten ever recaptured sustained growth. Christensen and Bower gave the mechanism from the disk-drive industry: incumbents keep allocating resources to what their best current customers ask for, and that is never the new thing. Kodak is the folk version. Steve Sasson built the first digital camera there in 1975, and management told him "that's cute, but don't tell anyone about it". The story is oversimplified, since the product wasn't ready, but the cash cow set the pace.

JD.ID was the opposite case, and I ran three business units there. The bias for action was enormous and contagious: sourcing from factories, group buying, logistics, warehouses on remote islands of the archipelago. Everyone believed and everyone moved, myself included. The problem wasn't the venturing. Each venture was too small, in size and in potential, next to the goal of 10 to 20% market share against Shopee and Tokopedia. At least the company was venturing exuberantly: it was a young executive dream and an experienced CFO nightmare.

The base rate on this side is Chris Zook's work at Bain. His team found that three-quarters of them fail. A later sample of nearly 200 companies put the odds at about 20%, mostly because companies moved too far from the core too fast. Lovallo and Kahneman explain why leaders take those odds anyway: the inside view anchors forecasts on the plan and neglects competitors. Flyvbjerg added the part Kahneman underrated: forecasts get cooked on purpose to get ventures started.

Double down Venture Hold Stop. Five rows, not four, because the fifth taught me that the framework has a floor: sometimes there is no move.

The base rates on each side of the table. Venturing succeeds one time in four or five; not venturing is how most stalls happen; and the merchant churn behind the Eatigo curse is a documented pattern, not a local accident. Limits: these figures come from different samples, decades and definitions (large US corporations, Bain client studies, US merchants, one Ohio restaurant panel) and aren't directly comparable. The 85% core-focus figure describes the winners, not the odds of winning by focusing. Sources in the bibliography.

I'm a tourist in ecology and psychology, so here I attribute rather than pronounce. The oldest formal version of this dilemma isn't in strategy at all. James March framed it in 1991 as exploiting what you know versus exploring what you don't. Organizations drift toward exploiting, he found, because it pays sooner and more visibly.

The sharpest tool is Eric Charnov's marginal value theorem from 1976. A forager should leave a patch when its intake rate drops to the average of the whole habitat. Around month nine at Eatigo the return per new restaurant fell to what the next patch would have paid, and we stayed. The catch is that the theorem assumes you know the habitat. A company doesn't know what delivery or tasting menus pay until it has tried them. Clausewitz said the same about an attack that has spent its force: everything depends on discovering the culminating point by the fine tact of judgment. The plateau you can compute. Whether the answer is more or new, you can't. That's the art.

When to leave a patch. The gain curve flattens; the dashed line from the moving cost is the habitat average; the tangent marks the optimal exit. Stay past it and you earn less per month than the next patch would pay. Limits: a textbook rendering of Charnov's theorem with an arbitrary curve, not fitted data. The theorem assumes a known habitat; the whole difficulty in a company is that the habitat average is unknown until you move. The placement of Eatigo and JD.ID on the curve is my judgment, not a measurement.

Ally Both Opposition. My concept sits between the growth research (venture rarely) and the stall research (but do venture), with the portfolio argument deciding who gets to.

In 1993 Kahneman and Lovallo described a decision maker who treats each problem as unique and isolates it from future opportunities. He turns timid because he never sees the aggregation across many bets. That describes a manager proposing one venture: one bet with a 20 to 25% base rate, defended alone. The person who holds the portfolio sees the same bet as one of ten and can afford to lose seven. Bezos said it in his 2015 letter: big winners pay for so many experiments. True, and it's the portfolio holder saying it. The employee who pitched the Fire Phone didn't write the shareholder letter after the $170 million write-down.

Grove faced the same choice when Intel's memory business was dying in 1985. He asked Gordon Moore what a new CEO would do if the board fired them both. Moore said get out of memories. Grove proposed they walk out the door, come back in and do it themselves.

The two famous cases of venturing from below prove the rule rather than break it. Burgelman's Intel study found that middle managers shifted scarce manufacturing capacity from memory to microprocessors before corporate strategy officially changed. They could because Intel's allocation rule rewarded margin per wafer, and the CEOs had left the rule in place. Ken Kutaragi built the PlayStation against strong opposition from most of Sony's board in June 1992. It happened because Norio Ohga, the president, kept the project alive and moved the team to Sony Music. In both cases a person with the portfolio absorbed the risk. No Ohga, no PlayStation. No wafer rule, no Intel pivot.

So this is the line I'd write on the wall of every growth team. Do not venture if you are not the decision maker. Venturing is too hard, there's too many mistakes part of the game. It's an art of the decision maker to choose whether something needs just more or needs a venture into other areas, and it belongs to nobody else. Beware that to venture you need power, you can't sell venturing to your boss or at least I have never seen it working. I'm sure some people did it, but the odds are just slim.

The two anonymized cases show the rule from the other side. A service company I worked in had found a few sweet spots in the market, unknown to most people, and had bumped into them almost by chance. It was exceptionally poorly run and still working well. Doing more of those sweet spots tripled revenue. The instruction was simple. Fix the basics, tick the boxes of good execution, let it grow and wait for the plateau to come, which could take 6 months, one year or 10 years. The plateau came two years after I started working on it, after a small change in positioning, found by chance, changed how buyers responded and changed the whole math. Very rare, but it happens. An ecommerce marketplace app I spent time inside was the opposite. Leadership acknowledged the limits and removed them, things went well for a while, and a hard cap appeared again. No strategic mistake, perfect execution, and still meager results. Nobody's fault.

The base rate for the core is stronger than my story. Bain's ten-year study found that only about one company in ten sustained profitable growth, and 85% of those did it by concentrating on a well-defined core. Read that way, the Eatigo founders had the numbers and I had a hunch across six countries. I still think the plateau was structural, but "just do more" is the base-rate winner and I can't dismiss it as inertia.

Venturing from below is real. Burgelman showed that strategy at Intel was made bottom-up before it was declared top-down. Kutaragi sold a games console to a company that thought games were toys. My rule is about odds, not impossibility, and I have no dataset on the odds. Both cases had a sponsor with the portfolio, which is a pattern in the cases I know and nothing more.

The counterfactual is unknowable. Eatigo didn't die. It merged into FunNow in September 2023 with 4,000 restaurants and 5 million users, under its own brand. The pandemic and a failed fundraise sit between the plateau and that outcome, and I can't prove that tasting menus or delivery would have paid. Zook's 20 to 25% says they probably wouldn't have, on the first try.

• The plateau is arithmetic. Any business that signs accounts and loses a share of them every period has a ceiling: new accounts per period divided by the churn rate, times revenue per account. It holds for a restaurant marketplace, a SaaS product, a gym, a telco, an agency, a B2B service, a subscription box, the seller side of any marketplace. The inputs sit in every dashboard, so the ceiling can be computed months before the revenue line flattens against it. There is an inner ceiling too: the number of accounts that fit the model. It arrives first and is harder to see, because the sales team keeps hitting quota while the base stops moving.
• The machine doesn't set the Y axis. The same model, the same policy and the same nine months to a plateau produce a business in one market and a countdown in another. What differs is the height of the curve, and the market sets it: what customers earn, what they'll pay, how many of them fit, and what it costs to reach them. Running costs are roughly flat across markets, so the same plateau clears them in one place and doesn't in the next. Nothing inside the machine changes that. Before choosing between more and new, decide whether the plateau in front of you is a business or a countdown.
• Double down or venture is a common dilemma. Every business model I've run into hits it, most within a few years, and the research says the same: most large companies stall at least once, and the stall is a decision point whether the leadership treats it as one or not.
• Making the right choice is an art, and most people get it wrong. The base rates are brutal on both sides. Most adjacency moves fail; most stalls come from not moving. There's no formula that picks between them, because no company knows what the next patch pays until it moves. Some people have the judgment for this. Most don't.
• Venturing is for the decision maker only. Venturing needs power to absorb the mistakes that are part of the game, and the odds of selling it upward are slim. If you're not the real decision maker, skip the dilemma. Your job is to make the numbers speak: show the plateau if it's there, or show that it isn't and keep optimizing for growth. Either way, the call belongs to the person who holds the portfolio.