All Growth Is Borrowed

In the last week of July 2009 the United States government ran out of money for a car promotion. Congress had set aside $1 billion for the Car Allowance Rebate System, better known as Cash for Clunkers, which offered up to $4,500 for trading in an old, thirsty vehicle. The money was meant to last for months. It lasted about a week. Congress added $2 billion. By late August roughly 680,000 vehicles had been traded in, showrooms that had been empty for a year were full, and officials were calling the programme a success.

Any marketing dashboard would have agreed. Conversions up. Cost per sale known to the dollar. Demand, apparently, created.

Then two economists did what almost no marketing department does. Atif Mian and Amir Sufi compared US cities that had many eligible clunkers with cities that had few. The clunker-rich cities bought far more cars during the programme. Over the following months they bought fewer. By about March 2010 the gap had closed. Almost every extra sale was a sale that would have happened within the year anyway. There was a loan inside that summer's figures, and the lender was the following spring.

That episode is one of the few times anyone saw the whole transaction. It had a control group and a window long enough to catch the repayment. Most growth has neither. So here is the claim this publication starts from.

All growth is borrowed. Every conversion is financed from an account no dashboard shows: attention, memory, trust, margin, permission or future demand. Reported growth that ignores those accounts is a loan booked as income. The first job of marketing is to know which account it is drawing down.

Borrowed does not mean bad. Sound companies borrow all the time. It means there is a lender, a rate and a repayment date. A management that cannot name them has a growth number, and no idea what it cost.

This is a proposal, not a finding.

The statement nobody prepares

Every serious company can say where its cash came from. The cash flow statement has a line for operations and a line for financing, because a dollar raised from a lender and a dollar earned from a customer look the same in the bank and mean opposite things.

Growth reporting has no such line. A marketing review shows conversions, cost per acquisition, return on ad spend and revenue. It does not show what financed them. Three problems follow.

Extraction looks like production. There are two ways to raise conversions per rupee. You can produce more willingness to buy, through a better product, a clearer idea or a wider memory. Or you can extract existing willingness faster, through a deeper discount, one more notification, a more aggressive checkout. A miner sees the same thing: output per shift rises with a richer seam and with a harder-working drill, and only one of those adds to what is in the ground. Both show up as efficiency. No conversion report separates them.

Interest arrives under other names. Banks send statements. Markets do not. Repayment turns up as rising acquisition costs, falling open rates, or a slide explaining that consumers have become more price sensitive. Sometimes they have, and the company taught them. Carl Mela, Sunil Gupta and Donald Lehmann studied about eight years of packaged-goods panel data and found that consumers grew more sensitive to price and promotion as promotion rose and advertising fell. Much of what a company calls a headwind is a repayment schedule it signed without reading.

The borrower and the repayer are different people. The typical CMO holds the post for roughly three to four years. Trust and memory accounts can take longer than that to run dry. The loan is taken out by one management and repaid by the next, which inherits a "challenging market".

JC Penney shows how large an unbooked debt can be. In 2012 a new chief executive, Ron Johnson, ended a regime of hundreds of promotions a year and moved to everyday prices. Sales fell by about a quarter in twelve months. The usual telling is about an arrogant outsider, and it has truth in it: he changed much at once and did not test the core change. But read it as a balance-sheet event. For decades the chain had drawn on its shoppers' sense of what a fair price was, teaching them that the list price was fiction and the coupon was the reason to visit. Johnson stopped the borrowing, and the company learned what it owed.

Why a sale has to be financed at all

A conversion looks like a single moment. It is the last step of a chain. Someone had to notice, remember, believe, accept the price, allow the contact and want the thing now rather than later. Each link draws on something scarce, and in almost every case the customer holds it. The company has a drawing facility, not an asset. The facility can be withdrawn.

Attention. People have a tolerance for interruption, and it is spent by use. The cleanest evidence comes from one of the largest extractors in history. In a long-running set of experiments, Google researchers Henning Hohnhold, Deirdre O'Brien and Diane Tang measured how users learned to ignore ads as ad load rose, a form of ads blindness that short tests missed. The company reportedly cut mobile search ad load by about half as a result. A firm with every incentive to sell more ad slots measured the account and chose to draw less.

Memory. This account behaves differently. It decays whether or not you draw on it, and it is usually filled by someone other than the person claiming the sale. A retargeting ad converts on a memory that a television campaign paid for three years earlier. Stop the refill and sales hold up for a while, which is exactly why the refill looks optional.

Trust. Jamie Luguri and Lior Strahilevitz tested dark patterns on large online samples and found that mild ones more than doubled acceptance of a dubious offer. Aggressive ones raised acceptance further and also provoked anger. That is a precise picture of the account: a measurable conversion lift, and a cost that lands elsewhere. Wells Fargo is the extreme case. Its cross-sell count, long praised by analysts, turned out in 2016 to include about two million accounts opened without authorisation, a figure later raised to about 3.5 million. The bank paid $185 million in initial fines and $3 billion in a 2020 settlement, and it lived under a Federal Reserve cap on its asset growth from 2018 until 2025. Few growth metrics have carried a higher rate of interest.

Margin. The discount is the most visible draw and the most misread, because it appears to be paid only once. The deeper cost is the change in what the buyer thinks the product is worth, and the change in when they are willing to buy.

Permission. A company's right to reach someone feels like property. It is a licence. Advertisers learned this in 2021, when Apple's App Tracking Transparency asked iPhone users whether they wanted to be tracked. In early 2022 Meta put the cost at about $10 billion in revenue for that year. The account had been drawn for a decade by firms that never held it, and a third party called it in with one dialogue box.

Future demand. Cash for Clunkers is the clear case. Igal Hendel and Aviv Nevo showed the supermarket version: shoppers stockpile on deal, so short-run price response overstates long-run demand. Every Singles' Day, White Friday and Big Billion Days carries the same question: how much of the peak is new, and how much was taken from the weeks on either side?

The accounts differ in two ways that matter. They refill at different speeds. Attention can recover in weeks, trust in years. And they fail differently. Margin and future demand decline smoothly. Trust and permission hold for a long time and then go over a cliff, which is why a steady reading from those accounts is less comforting than it looks.

The framework

We call it the Six-Account Ledger. It is a management statement, not an accounting standard.

Account	Held by	Typical draw	Interest disguised as	Refill speed	Failure shape
Attention	Audience	Ad load, notifications, interruption	Falling response, ad blindness	Weeks to months	Gradual
Memory	Category buyers	Living off past brand spend	Rising acquisition cost, weaker search demand	Quarters to years	Slow, then visible
Trust	Customers, regulators, staff	Dark patterns, overclaiming, mis-selling	Complaints, churn, legal cost	Years	Cliff
Margin	Buyers' price beliefs, partners	Discounts, subsidies, free delivery	Promotion dependence, weak full-price sales	Quarters to years	Gradual
Permission	Users, platforms, law	Over-mailing, tracking, consent tricks	Opt-outs, deliverability loss, policy shocks	Slow, sometimes never	Cliff
Future demand	Next period's buyers	Pull-forward offers, stockpiling, launch spikes	Post-peak dips, soft next quarter	Immediate	Sharp dip

Every material growth plan then answers four questions.

1. Which account pays? Name it. "The market" is not an account.
2. What is the rate? How much future capacity does one conversion consume today?
3. Who can recall the loan? Customers can leave slowly. Platforms and regulators can leave overnight.
4. What goes back in? Does the conversion return anything to the account it drew from?

The fourth question matters most, and the arithmetic shows why.

The arithmetic of a drawdown

Take a deliberately crude, hypothetical model. A company has a stock of 100 units of latent willingness to buy. Each year it converts a share of that stock, which we will call the extraction rate. Marketing and word of mouth refill the stock by 10 units a year. At an extraction rate of 10%, it sells 10 units and the stock stays at 100. Steady state.

Now a clever team raises the extraction rate to 15%, with better targeting, sharper offers and tighter funnels.

Year	Stock at start	Sales	Change in sales	Extra sales to date
Before	100.0	10.0
1	100.0	15.0	+50.0%	5.0
2	95.0	14.3	-5.0%	9.3
3	90.8	13.6	-4.5%	12.9
4	87.1	13.1	-4.0%	15.9
5	84.1	12.6	-3.5%	18.5

Year one is a career-making result. Sales are up by half with no extra refill. Then come years of mild decline that any executive would blame on the market, and which most would answer by raising the extraction rate again.

Look at the last column. After five years the extra sales total 18.5 units, and the stock has fallen from 100 to 81.5 at the start of year six. The difference is 18.5. This is not a coincidence. The extra sales are the drop in the account, unit for unit. Over time sales drift back to 10 and the stock settles at 66.7, so the whole triumph adds up to 33.3 units of sales and a business a third smaller underneath. Nothing was created. It was moved from the balance sheet nobody keeps to the income statement everybody reads.

Now add the term that separates good companies from clever ones. Suppose each sale returns something to the account: a referral, a satisfied user who buys again without prompting, a habit, a reason to trust. Call the share returned the deposit ratio, d. A little algebra gives the steady state.

Long-run sales = refill ÷ (1 − d)

With a refill of 10, a deposit ratio of zero gives long-run sales of 10. A ratio of 0.5 gives 20. A ratio of 0.8 gives 50. At 1 or above, the account stops being a constraint and sales compound, which is the logic of the best consumer products and of network effects.

Read the formula again for what is missing. The extraction rate does not appear. In this model, funnel efficiency decides when you collect. It has no effect on how much you collect in the long run. Only two things do: what you put into the account, and what each sale puts back. Conversion optimisation is timing. Refill and deposit are value.

That is too strong for the real world, where stocks do not behave so tidily and better funnels can remove genuine friction. But most marketing organisations put nearly all their measurement effort into the one variable that drops out of the equation.

The strongest objection

The best objection has two parts.

First, this is old news. Robert Blattberg and John Deighton, and Roland Rust, Katherine Lemon and Valarie Zeithaml, taught marketers to treat customers as an asset. David Aaker and Kevin Lane Keller did the same for brands. Rajendra Srivastava, Tasadduk Shervani and Liam Fahey argued that market-based assets belong in any account of shareholder value. Agreed: all of this stands on them. The ledger's proposal is narrower. It treats six accounts as one system with drawdown rates, insists that the customer rather than the firm holds most of the balance, and asks which accounts can be exchanged for one another at what rate. Whether that is useful beyond those frameworks is an empirical question.

Second, the word "all" is doing too much. A great product is a sale that makes the customer better off. What is being borrowed when someone loves their purchase? The objection is half right, and the deposit ratio is our answer. Even a delighted buyer had to notice, remember and pay. But a conversion can return more than it took. Such a loan is self-liquidating, like a bank facility that finances stock which is then sold at a profit. The claim is that all growth is financed, not that all growth is reckless. A company with a deposit ratio near one can borrow heavily and should. The danger is the company that only behaves as though it has one.

Who borrowed well, and who did not

Japan. In December 2018 PayPay launched a campaign that gave 20% back on purchases, with 10 billion yen set aside. It reportedly ran out in about ten days. On the Clunkers reading that looks like a burnt subsidy. But PayPay was spending its own margin to buy something durable: users and merchants on a two-sided network, where each new user made the service more useful to the next merchant. That is a high deposit ratio. It went on to become a leading QR payment service in Japan.

India. Zomato Gold offered diners a two-for-one deal, and for a time it drove strong membership growth. The account being drawn was largely not Zomato's. It was the margin of restaurants. In August 2019 restaurants, backed by their trade body, logged out of the programme in large numbers, with reports of well over a thousand leaving. A growth programme can run on someone else's balance for a while. The account holder gets to see the statement even if you do not.

China. Luckin Coffee is the uncomfortable case. It grew on heavy coupons, admitted in 2020 to about RMB 2.2 billion of fabricated sales, was delisted from Nasdaq, and then came back to pass Starbucks in China on store count. One reading: low prices were the product, so the coupons turned into a durable habit. Another: Chinese coffee demand grew fast enough to forgive almost anything. The ledger should be honest about this. A rising market can refinance bad debt.

Europe. Ryanair spent years running down goodwill with fees and open hostility, and it grew regardless, because low fares were a very large deposit. Then a 2013 profit warning led to a public turn towards treating customers better, and Michael O'Leary remarked that he would have started years earlier had he known how good it would be for business. It repaid by choice, while it could afford to.

What the CMO and CFO should do differently

Add a "financed by" column to the growth review. Every material initiative names its account in writing. A team that writes "future demand" next to a sale event will read the following month's numbers differently.

Pick one proxy per account and track it beside revenue. They will be imperfect: share of sales at full price for margin, post-promotion dip for future demand, unprompted recall among category buyers for memory, unsubscribe rate per send for permission, refund and complaint rates for trust, response decay per unit of ad load for attention. The test of a proxy is whether it moves before revenue does.

Treat rising acquisition cost as a question about the past. Before anyone blames the auction or the algorithm, ask what the company was drawing on three years ago, and whether it has stopped refilling.

Keep permanent holdouts. Mian and Sufi could see the repayment because some cities had few clunkers. A small group that never receives the offer is the only way to see what the offer cost after it ended.

Set drawdown limits. Banks set exposure limits because a loan officer's incentive is to lend. A CFO and CMO can agree ceilings in advance: share of revenue sold on promotion, messages per user, ad load per session. Like covenants, they bind before the quarter gets difficult.

Pay for deposits. If bonuses reward conversions alone, the company has paid its people to raise the one variable that drops out of the long-run equation.

How this could be wrong

The accounts may not be measurable. Randall Lewis and Justin Rao showed that even the direct sales effect of advertising is extremely hard to detect statistically. If hidden accounts cannot be estimated with useful precision, the ledger stays a metaphor, and a metaphor does not change a budget.

The claim may be empty. If every sale is financed by definition, the statement can never fail. We accept that the identity alone proves nothing. The testable content is narrower: that proxy readings on these accounts lead revenue, that deposit ratios differ widely between firms in the same category, and that companies systematically under-count their drawdowns.

Repayment may be smaller than we imply. Cars are durable goods with a replacement cycle, which makes pull-forward unusually clean. In packaged goods, Koen Pauwels, Dominique Hanssens and S. Siddarth found that promotion effects on sales were essentially temporary, with almost no permanent effect in either direction. For many categories the drawdown may be a rounding error against the refill.

The six may be two. Attention, memory and trust may be one thing, usually called brand. Margin and permission may be expressions of it. Fewer accounts would be a better theory.

The Ledger

1. By 2030: at least three of the world's 50 largest advertisers will publish a recurring measure of one hidden account with a disclosed method, such as share of sales at full price or experimentally tested marketing return. Confidence: 45%. Falsifier: fewer than three have done so by the end of 2030.
2. By 2031: a published study using firm-level data will show that drawdown proxies predict revenue decline at least four quarters ahead, beyond what sales trends alone predict. Confidence: 55%. Falsifier: credible public tests find no lead, or no such study exists by the end of 2031.
3. By 2030: at least one Fortune Global 500 consumer company will tie part of its top marketing executive's pay to a disclosed non-revenue measure of brand or customer asset health. Confidence: 40%. Falsifier: no such disclosure appears in a public compensation report by the end of 2030.