CompanyGraph
How to compare an observed business decision with a credible alternative without turning hindsight into evidence.
The missing outcome is the problem
When a company launches a product, investors see what happened after the launch. They do not see the same company, with the same customers, competitors, employees, and macroeconomic conditions, choosing not to launch. That unobserved outcome is the counterfactual.
The question is not “Could management have done something else?” Almost anything is possible in that loose sense. A useful counterfactual specifies the decision, the alternative, the starting date, the resources available, and the outcome being compared. Without those boundaries, a counterfactual becomes a story about what the analyst wishes had happened.
From thought experiment to causal method
The potential-outcomes tradition formalizes a causal effect as the difference between an outcome under treatment and the outcome for the same unit under control. Imbens' review explains the framework and its use in economics. Only one potential outcome is observed for a company at a given time, so the missing one must be approximated with a randomized experiment, a natural experiment, a matched comparison, a time-series design, or an explicit model.
| Counterfactual design | What it can contribute | Typical weakness |
|---|---|---|
| Before-and-after comparison | Whether an outcome changed after the decision | Other changes occurred at the same time |
| Peer or control comparison | A reference path without the decision | Peers differ in exposure, timing, or quality |
| Internal alternative | What another unit or product experienced | Selection and different operating conditions |
| Scenario model | Consequences of explicit assumptions | Results are only as credible as the assumptions |
Netflix and Blockbuster: an observed divergence, not a single cause
Netflix and Blockbuster are often used as a simple story in which one company embraced the future and the other refused it. The filings show something more specific. Netflix's 2007 filing describes an online subscription service with no due dates, no late fees, and a growing subscriber base. Blockbuster's 2007 filing describes more than 7,800 stores, an online program, and the elimination of late fees while it tried to integrate store and online rentals.
The relevant counterfactual is not simply “Blockbuster should have copied Netflix.” It could have invested more in online delivery, closed stores sooner, changed pricing, or pursued a different partnership. Each alternative required capital, licensing, logistics, customer migration, and a willingness to reduce the economics of the existing store network. The observed outcome cannot tell us which alternative would have produced the best result without a model or comparable evidence.
Physical and financial constraints narrow the alternatives
Counterfactuals become more credible when they follow the resources and timing of the real business. A factory cannot instantly switch to a different certified product. A lender may not fund the alternative after a covenant breach. A subscription company cannot keep both an expensive legacy channel and a low-price digital channel without changing its cash needs.
These constraints also create second-order effects. If a company diverts cash to a new product, maintenance or debt repayment may be delayed. If it preserves the old channel, it may retain cash in the short term while losing customer learning. The alternative must include the consequences that would have changed the company's future state.
Where counterfactuals go wrong
- Outcome bias. A successful result makes the decision look obvious even when the evidence available at the time was weak.
- Single-cause hindsight. A later failure is attributed to one decision while ignoring rates, demand, competitors, and execution.
- Impossible alternatives. The proposed path assumes money, talent, approvals, or technology that the company did not have.
- Changing the comparison unit. A company outcome is compared with a sector average without accounting for size, customers, or starting conditions.
- Double counting. The same revenue or cost effect is attributed to both the decision and the response it triggered.
How to use the question
State the decision and the alternative in one sentence. Record what was known at the decision date. Identify a comparison group or model, list the assumptions that make it comparable, and test a competing explanation. Then ask what evidence would have changed the conclusion.
A counterfactual is valuable because it forces the analyst to identify a causal claim rather than merely narrate sequence. It remains an inference, not a hidden observation. The stronger the claim, the more carefully the unobserved alternative, the operating constraints, and the source of comparison must be specified.