A credible quantum computing business case does not begin with a promise of quantum advantage. It begins with a measurable business problem, a fair classical baseline, controlled pilot costs and decision criteria agreed by stakeholders. This guide provides a repeatable method for assessing a quantum computing pilot, estimating potential ROI, documenting assumptions and deciding whether to proceed, pause or redirect the work.
Overview
Enterprise quantum computing is best evaluated as an investment decision under uncertainty. A proof of concept may be valuable even when it does not outperform a classical method: it can test data requirements, expose integration constraints, develop internal capability and clarify whether a problem is suitable for future quantum hardware. However, those benefits should be recorded separately from financial returns.
A strong quantum computing business case answers five questions:
- What business decision or process is being improved?
- How does the current classical approach perform?
- What will the pilot cost in staff time, infrastructure and external services?
- What evidence would justify further investment?
- What is the decision if the quantum approach does not meet the target?
Suitable candidates often involve complex optimisation, simulation, sampling or machine-learning workflows. That does not make them automatic quantum use cases. The problem must also have a clear objective, usable data, a tractable pilot scope and a classical method that can be tested fairly. For example, a supply-chain optimisation pilot should specify the routes, constraints, planning horizon and business costs before selecting an algorithm. See the quantum computing supply-chain evaluation framework for a more detailed problem-selection approach.
How to estimate quantum computing ROI
Use a two-layer model: estimate the cost and operational value of the pilot, then assess the potential value of a future production deployment. Keeping these layers separate prevents a small proof of concept from being presented as if it had already delivered production savings.
1. Define the baseline
Document the method currently used, including software, hardware, data preparation, runtime, solution quality and human review. If the organisation already uses a classical optimiser, compare against that implementation rather than against an intentionally weak alternative. Record the baseline on representative problem instances, not only on a small demonstration dataset.
Useful baseline measures include:
- time to produce a usable solution;
- solution quality against the agreed objective function;
- operational cost of the resulting decision;
- data preparation and post-processing effort;
- reliability, repeatability and constraint violations.
2. Estimate the pilot cost
Use the following simple model:
Pilot cost = internal labour + external support + cloud or hardware usage + data and integration work + governance and contingency.
Internal labour should include domain experts, software engineers, data specialists, security or procurement reviewers and the person responsible for programme decisions. Cloud usage should be estimated from the proposed number of experiments, circuit or job size, simulator requirements and expected repetitions. Platform pricing and availability change, so record the date and source of every estimate rather than treating a price as permanent. A guide such as this comparison of quantum cloud platforms can help structure the platform review, but current supplier terms should be checked before approval.
3. Estimate value scenarios
Do not use a single optimistic ROI figure. Build at least three scenarios:
- Downside: the pilot produces learning but no measurable operational improvement.
- Expected: the workflow reaches a defined improvement threshold or reduces a specific part of the process.
- Upside: the approach supports a larger deployment and creates material recurring value.
For a production scenario, a basic calculation is:
Net annual value = annual benefit - annual operating cost.
Indicative ROI = (net annual value - initial investment) / initial investment.
Label the result as an estimate. Benefits may include reduced planning time, fewer constraint violations, improved asset utilisation or better service levels, but each benefit must be translated into a business measure and validated by the process owner.
Inputs and assumptions
Before requesting funding, create an assumptions register. This is more useful than a polished headline number because it shows which inputs could change the recommendation.
| Input | What to record | Why it matters |
|---|---|---|
| Business objective | Decision, process, owner and measurable outcome | Prevents a technical demonstration becoming the project goal |
| Problem size | Variables, constraints, data volume and expected growth | Determines whether the pilot is representative |
| Classical baseline | Algorithm, configuration, runtime and solution quality | Creates a fair comparison |
| Quantum approach | Algorithm, encoding, circuit design, simulator or hardware path | Identifies technical dependencies |
| People | Roles, estimated days and opportunity cost | Captures the largest practical cost in many pilots |
| Platform | Provider, access model, job assumptions and price-check date | Allows cloud estimates to be refreshed |
| Success metrics | Quality, runtime, reliability, integration and user acceptance thresholds | Defines the decision at the end of the pilot |
| Risk and governance | Data sensitivity, security review, procurement and operational controls | Tests whether deployment is feasible, not merely interesting |
Technical readiness should be assessed alongside commercial value. A promising algorithm may depend on data encoding, error mitigation, repeated sampling or a hardware capability that is not available within the required operating conditions. For an introduction to encoding choices, read Quantum Data Encoding Methods Compared. For development workflow decisions, compare Qiskit, PennyLane and Cirq before committing to a toolchain.
Reusable pilot scorecard
Score each category from 0 to 3, using written evidence rather than enthusiasm:
- Business value: no quantified benefit to a clearly valued outcome.
- Baseline quality: unclear comparison to a documented and reproducible benchmark.
- Data readiness: inaccessible or poorly defined data to a validated dataset.
- Technical feasibility: speculative approach to a tested workflow.
- Integration readiness: isolated experiment to an identified production interface.
- Stakeholder commitment: informal interest to an accountable business owner.
Set a minimum threshold before beginning. A low score does not always mean “stop”; it may indicate that the next investment should be data preparation, baseline engineering or staff education rather than quantum execution.
Worked examples
Example A: optimisation pilot
An operations team wants to test a quantum approach for a constrained scheduling problem. The pilot scope is one planning cycle and a fixed sample of historical instances. The team estimates internal labour, platform usage, data preparation and governance review. Its primary metrics are solution quality, time to solution and the number of constraint violations.
The comparison table should look like this:
| Measure | Classical baseline | Quantum pilot | Decision question |
|---|---|---|---|
| Solution quality | Recorded baseline result | Measured result on identical instances | Does the difference matter operationally? |
| Runtime | End-to-end elapsed time | Includes preparation, execution and post-processing | Is the workflow practical? |
| Reliability | Repeatability across runs | Variation across repeated executions | Can users trust the output? |
| Total cost | Current operating estimate | Pilot and projected production estimate | Is any benefit larger than its cost? |
A positive result is not simply “the quantum result was better”. The result should show whether the improvement exceeds an agreed materiality threshold and remains useful after integration overhead is included. If it does not, the pilot may still recommend further research, a different formulation or a return to classical methods.
Example B: quantum machine learning exploration
A data team proposes a hybrid quantum-classical model for classification. The business case should compare it with an appropriate classical model using the same training, validation and test design. Measures might include predictive performance, training time, inference cost, model stability and explainability requirements. A quantum model should not be approved solely because it is novel. The business owner must identify which improvement would change a decision or create value.
A practical starting point is a small reproducible experiment, such as the workflow described in this PennyLane hybrid model tutorial, followed by a review of whether the data and evaluation method reflect the intended enterprise use.
When to recalculate
Recalculate the business case at each formal stage: before funding, after problem formulation, at the pilot midpoint and before any production recommendation. Refresh it sooner when the dataset, problem size, algorithm, platform, staffing model or success threshold changes.
At minimum, review pricing and access assumptions twice a year and before procurement. Recheck benchmark results when new problem instances become available, when the classical baseline is upgraded or when the quantum workflow moves from simulation to hardware. Platform capabilities, execution models and supplier terms can change, so a previous estimate should be treated as dated evidence rather than a standing commitment.
Use this closing checklist for an investment review:
- Confirm the business owner and the decision the pilot supports.
- Attach the latest classical and quantum results on identical instances.
- Separate demonstrated benefits from future assumptions.
- Update labour, platform, integration and governance costs.
- Record downside, expected and upside scenarios.
- Choose one decision: expand, redesign, monitor or stop.
- Set the next review date and name the person responsible for updating the assumptions.
This approach keeps an enterprise quantum computing strategy evidence-led. It also makes a pilot useful when the immediate conclusion is that the organisation is not yet ready: the team leaves with a stronger baseline, clearer data requirements and a defensible reason for its next decision.