Starting an AI project is not simply a technology decision.
Before committing, test more than technical possibility: will people use and trust the output; is the data appropriate for the purpose; do costs work at expected usage; and are supplier, policy, and regulatory assumptions understood?
The AI Value Canvas helps teams surface those questions before they become expensive commitments.
It is a practical framework for deciding whether an AI project is worth pursuing and, if it is, how to design it responsibly. It brings customer, business, technical, risk, and operational perspectives into one conversation before code is written.
Its purpose is not to make every idea look viable. It is not a promise that an AI project will deliver ROI.
Its purpose is to improve the quality of the decision before commitment: is this a real user problem, is AI the right response, can the organization operate the solution responsibly, and are the expected benefits and costs credible for the current stage of work?
Why AI projects need a different kind of discipline
AI initiatives require teams to make several connected decisions at once: what task is being addressed, what evidence supports the approach, how people retain appropriate control, and how the system will be operated after launch.
Model availability is a supplier dependency. Providers may retire models, requiring dependent software to change; check the lifecycle terms that apply to the selected provider and keep a migration path proportionate to the risk. See, for example, OpenAI’s Deprecations. Evaluate and monitor the system in its intended context of use, including after deployment, as recommended by the National Institute of Standards and Technology’s Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (2023).
The human side is equally important. Define when people should use the system, how they can correct or challenge an output, and where a qualified person remains accountable. Adoption and interaction design belong in the solution concept from the start, not in a communications plan after launch.
Identify the jurisdictions and legal, privacy, sector, procurement, and transparency requirements that apply; record the assumptions and their owners; and define what would cause the project to change course.
This does not mean that AI projects should be avoided. It means they should be approached with enough structure to learn early, limit unnecessary exposure, and make deliberate choices rather than treating a promising demo as proof of a sustainable solution.
The problem the Canvas solves
An AI project can begin with a solution: “We should build a chatbot,” “We need an agent,” or “Can we use AI to automate this?”
Those questions are understandable, but they arrive too late in the decision sequence. They assume that AI is necessary before the team has agreed on the user’s job, the current pain, the non-AI alternatives, the acceptable level of autonomy, or the evidence needed to justify investment.
The AI Value Canvas changes the starting question:
What problem does AI solve better than the alternatives already available to the user?
That question makes room for a better outcome than building an AI system. The right answer may be a rule, a simpler model, a process change, better access to information, or additional human support. It may also be AI—but with a narrower scope, more human approval, stronger data controls, or a different delivery approach.
The Canvas makes the assumptions behind these choices visible. It helps a team identify what it knows, what it is estimating, what evidence it still needs, and what risks or dependencies could invalidate the plan.
What the Canvas helps a team decide
The Canvas examines ten elements across four domains:
- Customer Understanding (Desirability) asks whether the identified users have a meaningful problem and whether they are likely to adopt the proposed change.
- AI Opportunity & Solution (Feasibility) asks whether AI is appropriate and how people should interact with the system.
- Data, Model & Risk (Sustainability) asks whether the data, model approach, safeguards, and operating conditions are responsible and sustainable.
- Value & Cost (Viability) asks what success means, what it will cost to build and run the solution, and whether to build, buy, or partner.
It also records the conditions around the project: its Strategic Intent, accountable Project Owner, Horizon, and Organizational Readiness.
A completed Canvas should support one of four legitimate next moves:
- Proceed when the evidence is sufficient for the current Horizon.
- Test an assumption when a specific uncertainty needs an Experiment.
- Rescope when a smaller, simpler, or safer version of the opportunity is more credible.
- Stop or defer when the case is not currently justified.
Stopping a weak project early is not failure. It is a better decision than committing resources to a problem, solution, or operating model that has not earned confidence.
Return on investment (ROI) matters—but it is not the promise
AI projects should have an honest value case. Teams need a baseline, a target, a unit of measurement, and a realistic view of both build costs and ongoing run costs.
But a return-on-investment estimate is not a guarantee, especially when user behaviour, model performance, usage volume, vendor conditions, and operating constraints are still changing.
The Canvas therefore treats ROI as a Viability question to test, not as the framework’s promised output. A credible project can show what value it expects, what assumptions that expectation depends on, how costs may change, and what evidence would cause the team to revise its plan.
When to use the Canvas
Use the Canvas when an AI idea is still open to influence. That includes during ideation, before a proposal is funded, when assessing a vendor, after an Experiment, before expanding a Pilot, or when deciding whether to scale.
It is especially useful when the proposed solution affects important decisions, sensitive data, customer or employee trust, regulated processes, or recurring operating costs.
The Canvas may be unnecessary for a genuinely low-risk, reversible experiment with a narrow scope and no meaningful impact on users, data, or operations. Even then, its central question remains useful: what evidence would show that this idea deserves further investment?
How to use this guide
Work from the customer outward, not from the model backward.
Start by understanding who the user is, what they are trying to accomplish, and what gets in their way today. Then test whether AI is needed, design an interaction that gives people appropriate control, assess the data and risks, and finally test the value and cost case.
Treat the Canvas as a living document. Revisit it after user research, a data assessment, an Experiment, a Pilot, a vendor or model change, a regulatory change, or a production incident.
The goal is clear thinking before commitment: to pursue the right AI project with appropriate safeguards—or to avoid spending on the wrong one.