Despite the mathematical precision, technology decisions made purely on discounted cash flow analysis and net present value tend to be absolutely wrong. The false precision gives committees comfort to make the call. The methodology ignores the dynamism of both micro and macro factors. Worse, it ignores something the committee will spend the next five years doing anyway: making choices.
Here is the part nobody puts in the business case. When you approve a $40M cloud migration, the cash flows are one part of what you're buying. The other part is a portfolio of future decisions — to expand, pause, switch vendors, walk away — and each of those decisions is worth real money. NPV assumes you've already made them. You haven't. You can't. And pretending otherwise is what makes the spreadsheet lie.
This is what real options pricing exists to fix. It's been around for forty years. Most enterprise technology committees still don't use it.
A framework that finance abandoned to academia
Stewart Myers coined the term "real options" in 1977. The insight was simple. A company's investment opportunities look a lot like financial call options — rights, not obligations, to commit capital when conditions are right. Lenos Trigeorgis turned the insight into a taxonomy that practitioners still use: options to defer, stage, expand, contract, abandon, switch, and grow.
Energy companies use it. Pharma uses it. Mining uses it. They have to — their bets are big, slow, and irreversible. Technology bets are big, fast, and also irreversible, but somehow the framework never crossed the moat. Walk into a typical capital approval meeting and you'll see a five-year DCF, a payback period, an IRR, and a sponsor performing certainty. You will not see the word "option" used in its technical sense even once.
Why the standard model is a confidence trick
A DCF model treats the investment as now-or-never. One number for revenue. One number for cost. One discount rate. A green light or a red light. The model has no place to record that you might pause at month nine, accelerate at month twelve, or pull the plug at month eighteen when the vendor doubles its pricing. Those choices don't appear, so the value they create — or destroy — doesn't appear either.
This isn't a rounding error. It systematically punishes the projects technology leaders make most: platform bets, capability investments, things whose payoff depends on what you'll do next. Two projects with the same expected NPV are not equally valuable if one locks you into a single path and the other lets you adapt. The spreadsheet scores them the same. Reality doesn't.
You could do the full Black-Scholes math. Most CIOs won't, and that's fine. The discipline of naming the options is worth more than the precision of pricing them.
The six options sitting in your portfolio right now
Deferral — the option to wait. Postponing a vendor commitment by one quarter, while a competitor's roadmap matures, often beats locking in early to "save 10%." Waiting has a price. So does pretending you have all the information you need.
Staging — the option to commit in tranches. Netflix's move from data centers to AWS took roughly seven years and was deliberately incremental. Slower, more expensive, and far more valuable than a big-bang cutover — because at every stage they retained the right to stop, change course, or push harder. Most enterprise modernizations are scoped as one multi-year commitment because that's how budgets are written, not because that's how good decisions are made.
Expansion — the option to scale up. A data platform sized for ten times current load costs more than one sized for today. The increment buys the right to grow without re-platforming. The cost shows up in the budget. The value shows up nowhere. Guess which one wins the argument.
Contraction — the option to scale down. Reserved cloud instances are cheaper per hour. They're also a quiet sale of your contraction option to the cloud provider. Whether that trade is smart depends on the volatility of your demand, which exactly zero TCO models actually model.
Abandonment — the option to walk away. This is the option enterprises destroy most casually. A custom integration is harder to abandon than a configuration. A five-year minimum is harder to walk away from than month-to-month. In fast-moving categories like AI tooling, paying more for shorter terms is often the rational choice, even when the unit price looks worse. The volatility is the point.
Switching — the option to change providers. Architectural choices that preserve switching — abstraction layers, open formats, portable data — have option value that lives entirely outside the TCO model. Choices that destroy switching have a corresponding negative value. Also not in the model. Vendors know this. You should too.
The $20M cautionary tale
Imagine the case that comes to your committee next quarter. A three-year, $20M build of a customer data platform. Annual benefit: $9M. Discount rate: 12%. NPV: about $12.4M. Beautiful. Approved.
Eighteen months in: $13M spent, 60% done. A new entrant launches a comparable product at a fraction of the build cost. The original sponsor takes another job. The remaining $7M will deliver something the team could now buy as a $4M annual subscription.
What does the original model say to do? Nothing. It was silent on this. It assumed completion as planned.
A real-options framing would have priced the project differently from day one: the first $5M as a true commitment, the next $8M as conditional on milestone evidence, the final $7M as an option to be exercised only if no superior alternative had emerged. Expected value is lower than the headline NPV. Risk-adjusted value is higher. And the decision at month eighteen is obvious instead of political.
The original estimate wasn't wrong. The model was wrong about which question to ask.
The real reason nobody does this
It isn't the math. Any competent finance team can build a binomial lattice in Excel. The reason is political.
Real options pricing exposes decisions that committees would rather not make explicit. Approve a project as a sequence of revocable commitments and you've invited the question: who decides to stop? On what evidence? With what career consequences for the sponsor who has been telling the board for two years that this thing is on track? Approve it as a single five-year commitment and the hard question is in the past tense. Off the agenda. Someone else's problem.
There's a signaling problem too. Capital approval processes reward apparent certainty. The sponsor who walks in with a fully-costed multi-year plan looks more credible than the one who walks in with $2M for stage one and a request to reconvene at month nine. The first sponsor is performing confidence. The second is doing the actual work. Guess who gets funded.
This is the gap between what looks like good governance and what is good governance. Most organizations are running on the first.
What to do Monday morning
You don't need to retrain your finance team. You need two changes.
First, require every technology investment proposal to name its embedded options and the triggers that would exercise them. What would cause you to expand? What would cause you to stop? What would cause you to switch? If the sponsor can't answer these in plain English, the proposal isn't ready. Forget the spreadsheet — the spreadsheet was never the bottleneck.
Second, fund in stages, with continuation tied to evidence rather than the calendar. Stage gates are not a new idea. Most organizations have softened them into reviews where cancellation is theoretical and continuation is the default. A real gate has the opposite default — stop unless the evidence warrants going on — and the evidence is specified in advance, not negotiated in the room.
Both changes do the same thing: they move the center of gravity from the moment of approval, when nobody knows anything, to the moments when information actually arrives. That is where the option value lives. That is where most enterprises throw it away.
The CIO's portfolio is full of flexibility nobody has been asked to price. The cost of asking is one uncomfortable conversation. The cost of not asking is currently buried in every multi-year program that's still running because nobody can quite remember how to stop it.
Start asking.