Business

Proof of concept vs pilot: choose the test your decision needs

Stefan Weiss, Co-Founder & CEO

Stefan Weiss

Proof of concept vs pilot: choose the test your decision needs

A proof of concept tests whether an idea or a critical part of it can work. A pilot tests a limited implementation in the setting where people would actually use it. Choose a proof of concept when feasibility is still uncertain; choose a pilot when you need evidence about operating the proposed solution in practice.

The distinction is useful, but the names are not universal. Before approving either, agree the question, the conditions of the test and the decision its results will support. Calling something a pilot does not establish that it is ready for real users, and a successful demonstration does not establish that an entire service will work.

This guide is for product teams, managers and consultants deciding how to test a proposed change. It explains the differences, gives a worked fictional example and shows how to preserve the evidence needed for the next decision.

The difference between a proof of concept and a pilot

A proof of concept, often shortened to PoC, concentrates on an important uncertainty about feasibility. That might be a technical connection, an unfamiliar method or a component whose behaviour the team cannot yet demonstrate. Its useful output is evidence about that bounded question.

A pilot puts a limited version of the proposed solution into use. It can reveal difficulties with roles, handovers, support, training and everyday exceptions that a controlled exercise leaves out. Its useful output is evidence about operating within the pilot's stated conditions.

Nesta distinguishes early feasibility work from limited live implementation and explicitly notes that these terms carry different meanings in different disciplines. Treat the following as a working comparison to agree with your team, rather than a universal project classification. Nesta: proof of concept, prototype, pilot and MVP.

Question

Proof of concept

Pilot

What are we trying to learn?

Can the important idea or component work?

How does this limited implementation work in our setting?

What is included?

Enough of the solution to investigate the feasibility question

Enough of the operating workflow to observe the chosen use case

Who needs to be involved?

People who can build, inspect and judge the feasibility evidence

People who use, operate, support and own the affected workflow

What should we keep?

Test conditions, observed results, limitations and unresolved assumptions

Operational observations, user feedback, exceptions and results against agreed criteria

What does success allow?

A decision about the next relevant uncertainty

A decision to extend, change, stop or investigate further

What does success leave unproven?

Overall usability, adoption and performance in routine operation

Performance outside the tested group, conditions and period

A pilot need not be longer simply because it has that name. Scope the work around what you need to learn and what you can responsibly operate. Avoid setting a standard duration before you understand the question.

Start with the decision, then name the test

Write the decision that is currently blocked. “We need to explore automation” is too broad. “Decide whether to invest in a limited trial of automated supplier intake” gives you a boundary.

Then complete this sentence: We cannot make that decision because we do not yet know whether…

If the missing fact is whether a component can perform the required transformation, a PoC may be appropriate. If the missing fact is whether staff can use the workflow alongside their existing responsibilities, a pilot may be more informative. If the missing fact is what people actually need, start with discovery rather than assuming either test will answer it.

Use requirements elicitation techniques to understand an unclear need before designing a test of a proposed solution. Testing a component efficiently is not helpful if it solves the wrong problem.

When a proof of concept is the useful next step

Choose a PoC when you can isolate a consequential feasibility question and judge it from observable evidence. Keep the question narrow enough that the result can change a decision.

For example, a team may need to know whether an existing export can be read by a proposed processing method without losing required information. A useful PoC uses representative authorised samples, records what happened and identifies the cases that failed. A screen showing a successful example is insufficient if it hides a required field or excludes a relevant exception.

State what you are deliberately leaving outside the exercise. Perhaps there is no production support, no staff training and no test of real workload. Those exclusions make the finding interpretable; they should travel with the result rather than disappear from the final slide.

Plug and Play's corporate innovation guide also separates technical feasibility from performance in the organisation's operating environment. Its examples concern corporate-startup engagements, so its suggested schedules and investment patterns should not be treated as requirements for every project. Plug and Play: PoC and pilot decisions.


An orange cloth ball is suspended above a cream safety net in a dark frame.

When a pilot is the useful next step

Choose a pilot when there is enough confidence to operate a limited version responsibly, but important questions remain about its use. Specify the people, work, period and conditions included.

Include the parts of the service needed to interpret the result. If a workflow depends on a person correcting every output, that human effort belongs in the observations. If an exception is routed back to the existing process, record how often that happens and who handles it. A smooth front-end experience can conceal an unsustainable handover.

Decide who can stop the pilot, how ordinary work will continue if it stops and who handles problems while it runs. The level of control should reflect the consequences of failure. Use the organisation's existing approval and specialist review processes where they apply.

A pilot can support learning about a defined setting. It does not automatically establish that a larger rollout will behave the same way. Different users, volumes, integrations or support arrangements create new questions.

A worked example: supplier intake

Consider a fictional operations team that wants to reduce manual re-entry of supplier information. The following tests and thresholds are illustrative planning choices, not measured results or industry benchmarks.

First question: can the information be transferred correctly?

The team proposes a PoC using a small set of authorised sample files with ordinary and difficult cases. It defines the fields that must survive the transfer and checks the resulting records against the source.

The result should distinguish successful transfers, incorrect fields, rejected inputs and cases the sample did not cover. A reviewer checks the outputs. “The method produced records” is a weaker conclusion than “these required fields matched the source in these test cases”.

If a necessary input cannot be handled, the team may change the method or stop. If the result is encouraging, it can decide whether a limited operational pilot is justified. It has not yet established that staff can manage the workflow.

Second question: can the team operate the workflow?

The team then proposes a two-week pilot with five staff members handling an agreed subset of intake work. These numbers only describe this fictional example; another question could require a different design.

Before starting, it defines how to observe successful completion, corrections, exception handling and support effort. It records the starting process so that later comparisons have a clear basis. It also decides which cases remain in the existing workflow.

If the pilot finishes, the review asks what happened within those boundaries. Did the expected work actually pass through it? Were difficult cases excluded? Did the support person quietly absorb extra work? A team could decide to revise the process even if participants liked the interface.

Keep the conclusions separate

A passing component test does not prove a workable service. A favourable pilot does not prove performance at every future scale. Keep both findings, each with its own conditions and limitations, when preparing a business case for the next commitment.

Agree success criteria before the work starts

For either approach, write down the question, the observations that would answer it and the possible next decisions. A criterion should be clear enough that people can recognise the same outcome when reviewing the evidence.

For the fictional intake PoC, “works well” is not useful. A more reviewable criterion identifies the required fields, the sample cases and the person who checks the transfer. For the pilot, a completion target alone may miss correction work, so include the operational burden that matters to the decision.

Define how unknown results will be handled. An unavailable measurement is not a failed measurement, and an untested exception is not a successful case. If the test conditions change, record the change and explain which conclusions remain supported.

Keep criteria proportional. A small feasibility exercise does not need an elaborate governance system, but it still needs a question someone can answer. A live operational trial needs owners and boundaries as well as a technical result.


Weeve source reader showing a synthetic Pilot planning transcript with speaker names, timestamps and proposed next steps.

Genuine Weeve interface with demonstration content. These passages propose a pilot and a later review; they do not report a completed test.

Use planning conversations without mistaking them for results

Recorded discussions can help preserve why a test was proposed, which conditions people agreed and what evidence they promised to bring to the review. They are also useful when an assumption changes between the planning meeting and the final recommendation.

With permission to record, Weeve can retain the conversation and transcript on your Mac. Chat with recordings can help you revisit selected planning and review discussions and follow source references back to the relevant passages.

A useful request is: “Across these planning and review recordings, identify the original test question, agreed boundaries, changes and reported results. Cite the passages and keep unconfirmed claims separate.” Check the returned sources and the coverage of the answer.

A statement that someone will run a test is a commitment. A later statement that it passed is a report. Neither replaces inspection of the actual test results when that evidence is required. Weeve helps review what the conversations contain; it does not independently execute or validate the experiment.

Local processing is available with the downloaded models. If you choose an optional external AI provider for chat, relevant text is processed by that provider. Check the selected mode and your permissions for the material before using it.

Where prototypes and MVPs fit

A prototype helps explore a proposed form, interaction or experience. It may be a sketch, role play or working model, depending on the question. A minimum viable product, or MVP, puts a minimal useful offering in front of users to learn about value and demand.

These approaches can sit alongside PoCs and pilots. Nesta notes that there is no mandatory sequence through all four. Choose the method that answers the next uncertainty, and agree what the label means in your project. A prototype may help you learn what to build before a technical feasibility exercise is useful.

Make the next decision explicit

Finish the review with a bounded conclusion: what the evidence supports, what remains uncertain and what should happen next. Name an owner for the next action and state whether it is a proposal or an agreed commitment.

The useful question is not whether the project has earned a more advanced label. It is whether the evidence justifies the next commitment under the conditions you expect to face.