What If the Numbers Are All We Have?

The artificial intelligence boom is building an astonishing amount of infrastructure for customers who are still working out what to do with it. Chipmakers are selling processors, cloud providers are adding capacity and data centers are consuming enough power to attract the attention of central banks. The spending may prove farsighted, but its eventual return depends on customers finding enough value in AI to keep paying for everything being built on their behalf.

That makes AI a useful illustration of a problem found in much smaller investment cases. A company proposes a new product, estimates how many customers will buy it and produces a revenue forecast in which the next three years unfold quarter by quarter. The model shows when revenue is expected to arrive, but not what will make customers appear.

Not all forecasts are the same

An operating sales forecast estimates the performance of a commercial system that already exists. It can draw on a real pipeline, conversion history, renewal behavior and known customers. The forecast may be wrong, but it is anchored in a business that has already demonstrated how it produces revenue.

A pre-investment revenue projection is different. It may describe a product that has not been built, customers who have not been acquired and a sales motion that has never been repeated. Rather than extrapolating from a business, it is making a claim about the business that capital is expected to create.

Confusing the two gives speculative projections more authority than they have earned. A mature company forecasting next quarter is asking what its existing commercial engine is likely to produce. A new initiative forecasting $50 million in its third year may still be asking whether the engine can be built.

Follow the first dollar backward

Most pre-investment models begin somewhere around price and volume. From there, they introduce conversion, retention and growth until the spreadsheet produces the desired destination. The arithmetic may be impeccable even when the commercial argument is not.

Take the first dollar of projected revenue and ask what produces it. A customer must abandon or reduce whatever they do today, accept the disruption of adopting something new and believe the promised improvement justifies the price. The company must reach that customer without spending the dollar to acquire it, then deliver enough value for the relationship to continue.

Each step changes the economics of the one that follows. An initiative addressing an urgent problem can tolerate more adoption friction than one offering a modest improvement. A product requiring extensive customization may support a high price while undermining the volume assumed in the model. A sales process built on executive relationships may win early customers but never achieve the acquisition rate in the forecast.

These are not implementation details to resolve after the investment is approved. Together, they determine whether the forecast describes a business that can exist. If the customer behavior, commercial motion and delivery economics are incompatible, adjusting the growth rate will not solve the problem.

Product-market fit belongs in this discussion, although it is only part of it. A company cannot prove fit before its product meets the market, and genuine innovation rarely comes from asking customers to dictate what should be built. It can still investigate whether the underlying need is important enough to support a business before committing itself to a solution.

The important distinction is between discovering a need customers have not articulated and hoping they will develop a need for something the company has already decided to build. Both can look like innovation from inside the organization. Only one begins with demand.

Once a product exists, that distinction becomes harder to see. Research starts serving positioning, feedback starts shaping features and sales begins searching for receptive buyers. Useful work is happening, but it can gradually turn into an effort to manufacture a market for an asset the company is now emotionally and financially committed to defending.

Sales can postpone the reckoning

A talented salesperson can make a weak proposition look healthier than it is. Relationships open doors, customization removes objections and persistence creates deals that would never have happened through ordinary market pull. The revenue is real, but its existence does not prove that the method can be repeated economically.

This matters because forecasts tend to treat early customers as the beginning of a curve. Five wins become the basis for fifty, even if each one required a founder, a bespoke implementation and the commercial equivalent of hand-to-hand combat. The spreadsheet preserves the revenue while quietly forgetting what it took to produce it.

A company may still choose that model, and some excellent businesses depend on highly skilled selling. The mistake is forecasting a repeatable product business when the early experience describes a labor-intensive commercial one. Heroic sales performance should be celebrated, but it should not be mistaken for a substitute for demand.

Financing changes who carries the uncertainty

A business may fund a new initiative with cash flow, equity, debt or some combination of them. Each can be appropriate. Cash from an established business can support exploration, equity can absorb uncertain outcomes in exchange for upside, and debt can make sense when the borrower has sufficient assets or cash flows to support fixed obligations.

The financing mix does not make the commercial proposition more or less true. It determines who carries the uncertainty, when the consequences become binding and how much time the initiative has to answer its unresolved questions. As more risk moves outside the operating business, clarity about the assumptions beneath the projection becomes more consequential.

The AI infrastructure market makes this visible at enormous scale. The Bank of England’s July 2026 Financial Stability Report noted that expected AI earnings depend on adoption spreading through the economy and producing benefits customers will pay to obtain. It also discussed circular financing arrangements in which technology companies invest in AI businesses that then purchase their investors’ infrastructure or services.

Those transactions are not imaginary, nor do they prove that the boom is unsustainable. They complicate the meaning of revenue appearing inside the ecosystem because capital moving through a supply chain can resemble durable demand before end customers have demonstrated that the economics work. The question is not whether AI deserves investment, but what must happen downstream for today’s investment to earn its expected return.

A projection should expose the bet

The purpose of upstream diligence is not to eliminate uncertainty. If certainty were the standard, most worthwhile innovation would be rejected before it began. The purpose is to stop a company from confusing a calculated risk with a calculated number.

A credible investment case should reveal where belief is doing the work. Leadership may decide that customers will change their behavior even though they have not done so yet, or that an expensive sales process will become more efficient as the product matures. Those can be reasonable bets when the people supplying the capital understand what must happen for the return to appear.

The problem begins when the projection conceals that dependence. A revenue figure gains authority because it appears in a financial model, while the fragile idea supporting it remains buried several assumptions away. Precision then becomes less a measure of knowledge than a convincing costume for optimism.

Before debating whether a projection should be higher or lower, leadership should ask what must become true for any version of it to happen. If the answer describes a credible change in customer behavior, the numbers have something underneath them. If the answer is that the product is excellent and the sales team will find a way, the company may not have a forecast yet.

It has a number in search of a business.