What is problem tractability?
Problem tractability is the judgment of whether a problem can realistically be solved with the data, methods, and compute that are available. In computer science the word has a strict meaning: a tractable problem can be solved exactly in reasonable computing time, while an intractable one grows so fast with size that brute force search is hopeless. In business AI the meaning is broader and rests on three conditions.
The data must exist, be accessible, and contain the signal, including labels for the outcome to be predicted. A known method must reach the accuracy the business needs in order to act. The compute, time, and skills must be within reach. A problem that fails one condition can often be reshaped into one that passes.
Problem tractability in practice
A law firm wants a model that predicts case outcomes. Its systems hold billing and document data but little about the facts that decide cases, and each judge appears in only a handful of matters. As posed, the problem is not tractable. Reframed as classifying incoming documents by matter type and predicting time to resolution from case activity, it becomes a project with data behind it.
What to watch for
- Ask for the accuracy the business needs before judging whether the data can deliver it.
- Treat missing labels as the usual blocker and plan to collect them before modeling.
- Prefer a narrower problem that is tractable now over a broad one that might be tractable later.