AI Stakeholders: Complete List for Artificial Intelligence Projects

AI stakeholders are the people, groups and organisations that can influence an artificial intelligence system, contribute to it, use it, govern it, or be affected by its outputs and decisions.

An AI project usually involves more stakeholders than the team building the technology. The system may depend on data collected by other teams, models supplied by external companies, people who review or act on AI outputs, and people who may never use the system but are still affected by what it recommends or decides.

AI stakeholder list

Use this list as a prompt rather than a fixed project structure. A small internal AI assistant may involve only a few of these stakeholders. An AI system used for recruitment, healthcare, finance, education or public services could involve many more.

The quick list gives you a starting point. The sections below explain why each group matters and help you find stakeholders who are easy to overlook.

1. Decide: business and product stakeholders

These stakeholders decide why AI is being used, what outcome it should improve, what decisions it may influence and what level of risk the organisation is prepared to accept.

Executive sponsor

The executive sponsor provides authority, funding and senior support. For higher-risk AI, the sponsor may also need to make sure there is clear accountability for how the system is used.

Business owner

The business owner owns the process or outcome the AI is intended to improve. They should be able to explain the problem before the team decides that AI is the right solution.

AI product owner and product manager

Product roles turn the business need into priorities for the AI team. They help define what the system should do, who it is for, how success will be measured and what behaviour is unacceptable.

Project or programme manager

The project or programme manager coordinates the people, dependencies, risks, decisions and change required to deliver the AI capability.

Business analysts and process owners

These stakeholders explain how work happens now and where the AI will fit. They can identify points where an AI recommendation, prediction or automated action changes an existing process.

Subject matter experts

AI outputs need to make sense in the real domain. Clinicians, lecturers, accountants, engineers, customer-service teams or other specialists may be needed to define requirements and judge whether outputs are useful and safe.

2. Data: people who provide and manage AI data

Data stakeholders are particularly important in AI because the behaviour of the system may depend heavily on the information used to train, fine-tune, retrieve for or evaluate it.

Data owners

Data owners are accountable for important datasets. They may decide whether the project can use the data and under what conditions.

Data stewards and data quality specialists

These stakeholders understand how data is defined, collected, maintained and checked. They can identify missing, inconsistent, outdated or misleading data before it becomes an AI problem.

Data engineers and data architects

Data engineers build the pipelines that move and prepare data. Data architects help define how information is structured, connected and governed across systems.

People who collect or create source data

A dataset may look like a technical asset, but it often comes from human activity. Staff entering case notes, customers filling in forms, researchers labelling examples or operational teams recording outcomes can all affect the quality of AI inputs.

People represented in the data

People whose behaviour, records, writing, images or other information appears in the data can be stakeholders even when they never interact with the finished AI system.

3. Design and build: AI and technical teams

These stakeholders turn the proposed use case into a working AI system and connect it to the organisation's existing technology and processes.

Data scientists

Data scientists explore data, develop models, compare approaches and evaluate how well a model performs against the intended task.

Machine learning engineers

Machine learning engineers turn models into reliable systems that can be deployed, scaled, monitored and maintained.

AI engineers

AI engineers may build applications around foundation models or other AI services, design prompts and workflows, connect models to organisational data and add controls around model outputs.

AI and solution architects

Architects decide how the AI capability fits with other systems, data sources, security controls, platforms and suppliers.

Software developers and integration developers

Most AI projects still require conventional software. Developers create the interfaces, workflows, APIs and integrations through which people and systems use the AI.

UX, user research and conversation design

AI systems can fail because people misunderstand what the system can do, place too much trust in an answer, or cannot tell when human review is needed. Designers and researchers help make the interaction understandable and usable.

4. Test and assure: people who challenge the AI

AI testing needs to go beyond checking whether the software runs. The project may also need to test whether outputs are accurate enough for the use case, whether the system behaves consistently, and how it responds to unusual or adversarial inputs.

Test and QA teams

Test teams check the application around the AI as well as the workflows, integrations, permissions and expected behaviour.

Model evaluation specialists

Evaluation specialists define test cases and measures for the model itself. They may compare model versions and look for unacceptable failure patterns.

AI red teams

Red teams deliberately challenge an AI system to discover unsafe, misleading, insecure or easily manipulated behaviour before it causes problems in real use.

Bias and fairness reviewers

Where AI affects people, reviewers may need to examine whether outcomes differ unfairly between groups or whether the underlying data creates systematic disadvantages.

Domain experts

Subject matter experts are often needed again during testing. A technically plausible answer may still be wrong or unsafe in the real domain.

User acceptance testers

Real users should test whether the AI helps them complete the work safely and effectively, including what happens when the AI is uncertain or wrong.

5. Use: people who use AI outputs

AI users may interact directly with a chatbot, recommendation system or prediction tool, or they may simply receive an AI-generated score, summary, alert or report inside another system.

Primary and occasional users

Frequent users understand the everyday strengths and weaknesses of the system. Occasional users may need clearer guidance because they have less opportunity to learn its limitations.

Managers and decision-makers

Managers may use AI-generated dashboards, forecasts, risk scores or recommendations. Their interpretation of the output can be as important as the model itself.

Employees working alongside AI

AI may change how people research, write, assess, plan, diagnose, approve, prioritise or communicate. Those employees need to be treated as stakeholders in the change, not simply as end users of a new tool.

6. People affected by AI

Some of the most important AI stakeholders may never see the system. They are affected because somebody else uses an AI prediction, classification, recommendation or generated output when making a decision about them.

Possible stakeholders include:

• job applicants assessed with AI support;
• customers receiving AI-generated decisions or recommendations;
• students affected by AI-supported assessment or support decisions;
• patients affected by AI-supported clinical decisions;
• employees whose performance or work is assessed using AI;
• people whose access to services is influenced by an AI score;
• people whose data is used to train or operate the system;
• people whose jobs or responsibilities change because of automation;
• employee representatives or trade unions; and
• communities that may experience wider effects from the system.

Who could be helped, disadvantaged, excluded or treated differently because this AI system exists?

7. Run and monitor: people who operate the AI

AI systems can change after launch because data changes, suppliers update models, user behaviour changes or the system encounters situations that were not present during testing.

MLOps engineers

MLOps engineers manage deployment, model versions, pipelines, monitoring and the operational processes needed to keep machine-learning systems reliable.

DevOps, platform and cloud teams

These teams provide the infrastructure, environments, access controls and shared services on which the AI application depends.

Model monitoring teams

Monitoring may look for changes in performance, data, costs, response quality, misuse, unexpected outputs or other signals that the AI needs attention.

Application support and service desk

Support teams deal with incidents and user problems. They need a route for escalating AI-specific failures rather than treating every issue as an ordinary software fault.

Human reviewers and escalation teams

Some AI processes need a person who can review an output, override it, investigate uncertainty or handle cases that should not be left to automation.

• human reviewers;
• exception-handling teams;
• incident response teams;
• model owners;
• service owners; and
• business continuity teams.

8. Govern: risk, legal, ethics and compliance stakeholders

Governance stakeholders help decide whether the proposed AI use is acceptable, what controls are required and who is accountable when something goes wrong.

Cybersecurity team

Cybersecurity stakeholders review threats to models, data, prompts, interfaces, identities and connected systems.

Data Protection Officer and privacy team

Where personal data is involved, privacy specialists may need to review what data is used, why it is needed, how long it is kept and how people's rights are protected.

Legal team

The legal team may advise on issues such as:

• contracts;
• intellectual property;
• licensing;
• use of personal or confidential information;
• responsibility for AI-generated content; and
• legal duties that apply to the use case.

Risk management and internal audit

Risk and audit teams can challenge assumptions, review controls, test accountability and consider what happens if the AI behaves unexpectedly or the supplier becomes unavailable.

AI ethics or responsible AI group

Some organisations have a responsible AI committee, ethics group or AI governance board that reviews higher-risk uses and sets organisation-wide principles or controls.

Senior governance and regulators

Boards, steering committees and regulators may become important stakeholders where an AI system has significant effects on customers, employees, patients, students, citizens or regulated decisions.

External AI suppliers and dependencies

Many AI systems depend on technology, models and data supplied by other organisations. Those suppliers should appear on the stakeholder map because changes to their services can directly change the project's risk, cost and behaviour.

• foundation model providers;
• AI platform providers;
• cloud providers;
• model hosting providers;
• vector database or retrieval service providers;
• external data providers;
• data labelling suppliers;
• AI consultants;
• software vendors embedding AI features;
• cybersecurity suppliers;
• specialist assurance providers;
• research partners;
• open-source model and software communities;
• professional advisers; and
• regulators and standards bodies.

Example: stakeholders for an AI customer-service assistant

Imagine an organisation introduces a generative AI assistant to help customer-service staff answer questions. The assistant retrieves information from internal guidance and drafts responses for staff to review before they are sent.

Notice that the customer is a stakeholder even though the employee, not the customer, operates the AI tool. The customer experiences the result.

AI stakeholders that are often missed

The technical team and sponsor usually make the first stakeholder list. Before you finish, check specifically for:

• people represented in training, evaluation or operational data;
• people affected by an AI-assisted decision who never use the AI;
• people responsible for collecting the source data;
• data owners and data stewards;
• human reviewers and escalation teams;
• people who need to challenge or override an AI output;
• staff whose jobs or processes will change;
• employee representatives or trade unions;
• accessibility specialists;
• model monitoring and operational support teams;
• owners of systems connected to the AI;
• foundation model and external AI providers;
• privacy, legal, risk and internal audit teams;
• responsible AI or ethics governance;
• customers or members of the public who may be affected; and
• regulators where the AI is used in a regulated activity.

How to identify AI stakeholders

Work through the AI lifecycle and the consequences of the system rather than trying to remember stakeholder job titles from a blank page.

Why are we using AI?

• Who asked for the AI capability?
• Who owns the problem it is meant to solve?
• Who is paying for it?
• Who decides whether it has succeeded?
• Who decides what the AI is allowed to do?

What data does it depend on?

• Who owns the data?
• Who collects, creates or maintains it?
• Who understands its weaknesses?
• Whose information is represented in it?
• Who approves its use?

Who will design and build it?

• Who chooses the model or AI service?
• Who builds prompts, agents, workflows or models?
• Who designs the user experience?
• Who integrates the AI with existing systems?
• Who owns the technical architecture?

Who will test and challenge it?

• Who knows what a good answer or prediction looks like?
• Who tests unusual and high-risk cases?
• Who looks for security weaknesses?
• Who tests accessibility?
• Who looks for unfair or systematically poor outcomes?

Who will use the output?

• Who types prompts or supplies inputs?
• Who reads AI-generated outputs?
• Who acts on AI recommendations?
• Who is expected to review or override the AI?

Who will be affected?

• Who could gain or lose something because of the AI output?
• Who may be classified, prioritised, recommended, rejected or investigated?
• Whose work changes?
• Whose data is used?
• Who might find it difficult to challenge an AI-supported decision?

Who will run and monitor it?

• Who deploys it?
• Who monitors quality and performance?
• Who investigates incidents?
• Who supports users?
• Who decides when the model or system needs to be changed or withdrawn?

Who governs it?

• Who is accountable for the use of AI?
• Who is responsible for privacy and security?
• Who checks legal and compliance requirements?
• Who reviews ethical or responsible-AI concerns?
• Does a regulator or standards body have an interest?

Quick AI stakeholder checklist

Use this shorter list as a final check before completing your stakeholder identification.

What to do next

Identifying AI stakeholders is only the first step. Once you have your list, you can:

• add them to a stakeholder register;
• carry out a stakeholder analysis;
• assess their influence, interest, needs and concerns;
• create a stakeholder map;
• identify people who are highly affected even if they have little formal power; and
• plan how you will engage your stakeholders.

For closely related checklists, see the software development stakeholder list, Big Data stakeholder list and the 105 stakeholder checklist. You can also browse all sector and project-specific pages in the Stakeholder Lists hub.

Summary

A complete AI stakeholder list needs to go beyond the sponsor, data scientists and users. It should include the people who decide, provide data, design, build, test, use, operate and govern the AI, plus people who may be affected by its outputs without ever using the system themselves.

Use the selectable checklist on this page to create a tailored list for your own project, then use stakeholder analysis to decide who needs the closest attention and engagement.