Go/No-Go Decisions on Data Readiness

Overview
In the PMI Cognitive Project Management for AI (PMI-CPMAI) methodology, AI projects follow an iterative, data-centric lifecycle built on CRISP-DM-style phases. These include Business Understanding, Data Understanding, Data Preparation, Model Development, Model Evaluation and Model Operationalization. A defining feature of the method is the use of formal Go/No-Go decision points at the end of each phase. The Go/No-Go decision on data readiness sits inside the Identify Data Needs and Data Understanding work. It is the checkpoint where the team and stakeholders decide whether the available data is good enough, in both quality and quantity, to justify moving forward with AI development.

Why It Is Important
Data is the foundation of every AI system. Industry studies keep finding that most AI projects fail because of data problems, not algorithm problems. Common examples are:
- insufficient data
- poor-quality or biased data
- inaccessible data
- data that lacks labels
- data that cannot legally be used

The data readiness Go/No-Go decision matters because:
- It prevents wasted investment. Stopping or pivoting early is far cheaper than discovering data problems after months of model building.
- It reduces project risk. It surfaces gaps in data access, quality, volume, privacy and governance before commitments escalate.
- It aligns stakeholders. It forces an explicit, documented decision rather than optimistic assumptions.
- It supports trustworthy AI. It checks for bias, representativeness, consent and compliance, such as GDPR or HIPAA, before data is used.
- It reflects CPMAI's core principle. AI projects are data projects first. If you do not have the data, you do not have an AI project.

What It Is
A Go/No-Go decision on data readiness is a structured evaluation of whether the data needed to solve the defined business problem is:
- available
- accessible
- sufficient
- of adequate quality
- appropriately labeled (if supervised learning is needed)
- representative
- compliant with legal, ethical and organizational requirements

The decision has several possible outcomes:
- Go: Data is ready, or can feasibly be made ready. Proceed to Data Preparation.
- No-Go: Data needs cannot be met. Stop the project, or return to Business Understanding to rescope the problem.
- Conditional Go / Iterate: Proceed with specific remediation actions. Examples include acquiring more data, labeling, data augmentation, synthetic data, third-party data purchase, or reducing scope to a smaller pilot.

How It Works
1. Start from the business problem.
Business Understanding defines the problem, the AI pattern (for example recognition, predictive analytics, conversational, or hyper-personalization), and the success criteria. Data needs are derived from these, not the other way around.

2. Identify data needs.
Determine which data is required, its sources (internal or external), its type (structured or unstructured), the required volume, the required history, the labeling needs and the update frequency.

3. Assess data against key readiness questions.
These are commonly framed around the 'V's of data (volume, variety, velocity, veracity) and practical criteria:
- Quantity: Is there enough data to train, validate and test?
- Quality: Is the data accurate, complete, consistent, timely and free of significant noise?
- Access: Can the team actually obtain the data, considering permissions, silos, APIs and cost?
- Labeling: For supervised learning, is ground truth available, or can it be created affordably?
- Representativeness and bias: Does the data reflect the real-world population and conditions the model will face?
- Compliance and ethics: Is use of the data permitted, considering privacy, consent, intellectual property, regulation and data sovereignty?
- Infrastructure: Are storage, pipelines and tools in place?
- Cost and time: Can data gaps be closed within budget and schedule?

4. Document findings.
Record data inventories, data quality reports, gap analyses and risks.

5. Convene the decision.
The project manager facilitates. Business owners, data owners and stewards, data scientists, legal/compliance and sponsors review the evidence against predefined criteria.

6. Decide and act.
Record a Go, No-Go or conditional decision with clear rationale and next steps. CPMAI is iterative, so a No-Go often means looping back to rescope the problem rather than abandoning AI entirely.

Key Principles to Remember
- Go/No-Go decisions happen at every phase, not just once.
- Decisions should be based on predefined, objective criteria, not enthusiasm or sunk cost.
- A No-Go is a valid, positive outcome. It saves resources and is a success of governance, not a failure.
- Iteration is expected. Start small, prove value and expand. Think big, start small, iterate often.
- The business problem drives data needs. Data availability may in turn force changes to the problem scope.

Exam Tips: Answering Questions on Go/No-Go Decisions on Data Readiness
1. Choose data-first answers. If a scenario describes pressure to start modeling before data is validated, the correct answer usually involves assessing data readiness first.

2. Recognize No-Go triggers. Look for these signals:
- data that does not exist
- data that cannot be legally used
- unlabeled data with no affordable labeling path
- heavily biased or unrepresentative data
- data locked in inaccessible silos

The best answer is often to pause, rescope or return to Business Understanding.

3. Prefer iteration over abandonment. When options include 'cancel the project' versus 'reduce scope or address the gap and re-evaluate', CPMAI generally favors iterating. Typical moves are rescoping, acquiring data or running a smaller pilot, unless the gap is clearly unresolvable.

4. Avoid sunk-cost reasoning. Answers suggesting 'we already invested a lot, so proceed' are almost always wrong.

5. Involve the right stakeholders. Correct answers include data owners, compliance/legal and business sponsors in the decision. The project manager facilitates; the data scientist does not decide alone.

6. Remember compliance and ethics. A dataset may be technically perfect and still be a No-Go if privacy, consent or regulatory requirements are not met.

7. Watch the phase ordering. Data readiness is decided before Data Preparation and Model Development. If a question asks when data issues should be discovered, the answer is 'as early as possible, at the Data Understanding Go/No-Go'.

8. Know the remediation options. These include:
- data augmentation
- synthetic data
- third-party or open datasets
- crowdsourced or human-in-the-loop labeling
- transfer learning or pretrained models to reduce data needs
- narrowing the problem scope

9. Use objective criteria. Pick answers that reference predefined readiness criteria, documented assessments and measurable thresholds over gut feel.

10. Eliminate extreme answers. Discard options like 'proceed and fix data later', 'let the model handle data quality' or 'skip the checkpoint to meet the deadline'. These contradict CPMAI's disciplined, risk-aware approach.

Sample Question Pattern
Scenario: A team discovers that only 20% of the required historical records are labeled, and labeling the rest would exceed the budget. The sponsor wants to begin model training immediately. What should the project manager do?
Best answer: Hold a Go/No-Go review. Present the data gap and evaluate options such as rescoping, a smaller pilot, alternative labeling approaches or additional funding. Then make a documented decision before proceeding.

Summary
The Go/No-Go decision on data readiness is CPMAI's safeguard against the most common cause of AI failure: inadequate data. It is an objective, stakeholder-driven checkpoint that weighs quantity, quality, access, labeling, representativeness and compliance against the business objective. It results in Go, No-Go or iterate. In the exam, favor answers that are data-first, evidence-based, iterative, compliant and free of sunk-cost bias.