How Many Phases Are There in the CPMAI Methodology? Complete Guide (2026)

The task of managing an AI project requires more than predetermined steps; here, project teams need to follow an iterative approach. As AI models are constantly evolving, project managers should follow a clear roadmap that will help them return to an earlier phase. This allows them to refine existing data, improve model performance, and reassess business goals when necessary.

Here, the need for a six-phase CPMAI methodology comes into the picture. Each phase of this methodology - from business understanding, data understanding, data preparation, model development, model evaluation, to model operationalization aims to minimize risks or ethical issues posed by AI. In this detailed guide, we will explore how this structured approach helps project teams define business problems, gather data, build, test, and deploy AI models.

What is PMI-CPMAI Certification?

The PMI Certified Professional in Managing AI certification is issued by the Project Management Institute (PMI). It is designed for professionals who are responsible for managing Artificial Intelligence and Machine Learning projects from start to finish. It is based on a six-phase CPMAI framework that helps teams identify a business problem, develop, monitor, and deploy an AI model.

Obtaining this AI project management certification demonstrates to employers and clients that you have the adequate amount of theoretical knowledge and practical skills necessary to:

• Identify how AI can minimize costs, improve customer satisfaction, speed up the decision-making process, or automate repetitive work
• Ensure AI project initiatives are aligned with business goals
• Evaluate whether the data available is sufficient and ready-to-use
• Bring cross-functional teams together - including technical, legal, operations, and business teams
• Document how the solution will be developed, used, monitored, and improved
• Ensure AI is used ethically and responsibly
• Minimize risks related to bias, security, compliance, and privacy

According to PMI, approx. 70-80% of AI projects fail due to poor stakeholder engagement, non-alignment of project initiatives with business goals, unclear objectives, irrelevant data, and limited governance.

All phases in CPMAI methodology are designed to equip professionals with a structured approach to identify the business problem. Next, teams can choose a suitable technology, continue to test, and refine it even after its deployment.

What Are the Six CPMAI Methodology Phases?

The CPMAI methodology breaks down an AI project into six phases. These phases are related to each other - each phase answers important questions, like why the project is important, how the final AI model will be deployed, and how it can be improved.

Phase 1: Business Understanding

During this phase, project teams should identify the business problem and understand it in detail. Teams should get started by answering questions such as why the project is required and what issues it can solve. For example, the project can be carried out to minimize costs, improve customer service, minimize business risks, or contribute to better decision-making.

This phase helps teams define long-term goals and allows them to work on a Minimum Viable Product. Teams can test the product in a real-world environment before the organisation makes a larger investment. Eventually, teams can decide if:

• Sufficient and suitable data is available
• The organization is equipped with adequate resources: people, tools, budget, and support

Phase 2: Data Understanding

Once the team has a clear understanding of the business problem, they can move to the next phase - data understanding. Teams should identify what data is required, who owns it, and where it is stored. Most importantly, they should examine whether the data is accessible and can be used legally.

Next, the team has to create a data inventory and assess it to examine the following factors:

Volume: Teams should ensure there’s sufficient data available. Availability of adequate data allows teams to test and train the model smoothly.

Variety: A thorough analysis of the data available helps teams to list different types of data formats: text, numbers, images, or documents.

Veracity: It helps teams measure the extent to which the data is reliable, accurate, and consistent.

Velocity: Teams can estimate the time taken to create and update data.

It also gives project managers insight into missing data, errors, inconsistencies in data, and bias.

Phase 3: Data Preparation

CPMAI Phase III requires teams to extract meaning from raw and unorganized data and train models based on it. This is considered one of the most time-consuming phases included in an AI project. To execute this phase completely, teams should follow the steps explained below:

Data Cleaning and Wrangling

Teams should study the available data and fix issues that can have a negative impact on the model. It includes activities such as:

• Replacing missing information
• Correcting incorrect values
• Removing duplicate records
• Merging data from multiple systems
• Removing noise or elements that cannot address the business problem

Addressing Data Debt

CPMAI encourages teams to identify issues that have accumulated due to disconnected systems, inconsistent data standards, and weak data governance. Teams should resolve these issues before moving to the next phase of model development.

Labeling and Annotation

Data labeling refers to categorizing data. For example:

• Images can be segregated as defective or non-defective
• Customer messages can be labeled as neutral, negative, or positive
• Financial transactions can be labeled as fraudulent or legal

Phase 4: Model Development

The model development phase equips teams with the approach necessary to develop an AI solution that solves the underlying business problem.

Choose the Right Approach

Teams should choose the most effective algorithm and development tool. The choice may vary based on factors such as:

• The volume and type of available data
• The business goal
• The estimated response time
• The amount of computing resources available
• Privacy, regulatory, and security needs

Buy, Build, or Adapt

Teams will be required to decide if the model should be built from scratch or if they can deploy an existing solution. Some of the available options are commercial AI platforms, cloud-based AI services, open-source models, foundational models, and Large Language Models.

Training the Model

The team is responsible for training a model, evaluating the outcomes, making necessary changes when required, and repeating the steps.

Avoid Overfitting and Underfitting

Teams often come across two major issues during the model development phase: overfitting and underfitting.

Overfitting issues occur when models memorize noise or complex examples instead of learning general patterns.

Underfitting problems occur when the model is unable to learn anything significant from the data or it is too simple.

Read our blog on PMI-CPMAI vs PMP: Which Project Management Certification Is Right for You?

Phase 5: Model Evaluation

CPMAI phase V includes the list of quality checks a team must perform to determine if the AI model can meet real-world business requirements and includes all the technical features. This phase is carried out to ensure a model shouldn’t generate wrong business outcomes or bring forth unexpected risks.

Testing with a unique dataset

Teams should use a different dataset to evaluate the model. It’s important to ensure this dataset wasn’t used at the time of tuning or training. It offers insights into how a model will behave once it is deployed.

Evaluating Technical Performance

Teams can choose a evaluation metric based on the business problem and the nature of model that is being developed. The key metrics are as follows:

Accuracy:

It helps to measure the average percentage of predictions that are accurate.

Precision:

Out of the total number of cases that were categorized as positive, how many cases were genuinely positive.

Recall:

Out of all the positive cases, how many cases can the model actually recall?

Measuring Business Value

The final model should be compared against the business goal or objective decided in Phase 1. Teams should be able to answer the following questions:

• Does the model reduce operating costs?
• Does it speed up decision-making process?
• Does it increase revenue or conversion?
• Does it reduce risk?
• Does it improve customer experience?
• Does it save employees time?
• Does it perform better than the current process?
• Is the value greater than the cost of operating the model?

Fairness and Ethical Use

Teams should examine if a model is performing evenly across diverse geographic locations, demographic segments, customer groups, and so on. They should ask questions like:

• Are certain groups receiving less accurate predictions?
• Could the model reinforce historical discrimination?
• Are sensitive variables influencing decisions unfairly?
• Can users understand or challenge the model’s output?
• Is human review required for high-impact decisions?
• Are the model’s limitations clearly communicated?

Testing and failure issues

An AI model should be tested under challenging conditions. Teams should find out what happens when testing is done under the following conditions:

• Data is incomplete
• Input values fall outside expected ranges
• Data formats change
• The system receives unusually high traffic
• A connected service becomes unavailable
• The model has low confidence
• Users provide unexpected inputs
• An attacker tries to manipulate the system

Phase 6: Model Operationalization

During the model operationalization phase, teams should put the AI solution to use within the real world and refine it continuously. It ensures the model can deliver exceptional value.

Choose the Deployment Environment

Teams should decide how and where the model will be deployed. These models can be deployed across cloud, edge devices, or on-premises environments.

Implement MLOps

Machine Learning Operations streamlines the task of implementing, monitoring, updating, and managing models.

Continuous Monitoring

Teams should monitor the model to examine its data quality, prediction accuracy, data drift, model drift, error rates, response time, error rates, user adoption, security events, and so on.

Version Control

Organizations should have access to the version history of each model. This gives them and overview of:

• Which model is currently active
• Which dataset was used to train it
• Which code and features were used
• When it was deployed
• Who approved it
• How it differs from earlier versions
• Whether it can be rolled back safely

Automate Testing

Before a new model is deployed, it should undergo technical, security, governance, and performance checks. Through automated pipelines, teams can:

• Validate incoming data
• Run model tests
• Compare new and current model performance
• Confirm compatibility with production systems
• Deploy approved models

Retraining Pipelines

It’s essential to retrain models when new data is generated or there’s a change in real-world conditions. Make sure to retrain a model under the following circumstances:

• On a fixed schedule
• When performance falls below an agreed level
• When new labeled data becomes available
• When the business process changes
• When a new model or feature set becomes available

Incorporate Human Feedback

Operational AI systems should be refined by incorporating human feedback. Users should identify errors, unusual cases, and changing conditions. Humans should be included in the decision-making process.

Human opinions play an important role when:

• Predictions have serious consequences
• The model has low confidence
• The situation is unusual
• Regulations require human review
• The model’s reasoning is difficult to explain

Conclusion

Traditional project management practices like waterfall or predictive approach isn’t enough - when it comes to managing AI projects. AI projects are vulnerable to continuous model improvement, changing data, or governance risks. The six-phase CPMAI methodology equips project managers with a flexible yet structured framework necessary to handle these complexities - right from defining business goals to deploying, monitoring, and refining AI models. Enrolling in a PMI - CPMAI certification training helps you develop the skills necessary to plan, execute, manage AI initiatives, and deliver AI solutions in modern organizations.