What it means
A proof of concept is a small, time-limited experiment to test whether an idea is technically possible and promising, usually with sample data and outside normal operations. It answers "Can this work?", not "Does this work in practice?"
Real-world example
A state library tests whether AI can generate useful subject keywords for digitised pamphlets. Over four weeks, a small team runs a model on 500 sample pamphlets and librarians rate the results. The PoC shows the keywords are useful for 70% of pamphlets but poor for those in older German typefaces, which shapes the next step.
Why it matters for AI adoption
PoCs reduce risk by testing assumptions cheaply. But a successful PoC with clean sample data does not prove the solution will work with real data, users, integrations and governance. Many organisations get stuck in "PoC purgatory" because they never planned the path to real use.
Often confused with
A PoC tests technical feasibility in a controlled setting. A pilot tests the solution with real users in real work for a limited time, including processes, training and support.
Questions to ask
- What exactly is this PoC trying to prove, and what result would make us stop?
- What would need to happen to move from PoC to a pilot?