An AI supplier can look like a simple subscription: create an account, connect a few files and start saving time. The real decision is bigger. The service may process customer information, influence business decisions, connect to internal systems or become part of a process that is difficult to replace.
An AI vendor risk assessment NZ small businesses can actually use does not require a large procurement department. It requires clear questions, evidence in proportion to the risk and a named person who can make a sensible decision. The aim is not to reject every new tool. It is to understand what you are buying before the tool becomes part of daily work.
The questions below form a practical starting point. They are general guidance, not legal or cybersecurity advice. A specialist review is sensible when the service will handle sensitive personal information, core financial records, safety-related work or a large customer base.
1. What problem are we trying to solve?
Start with the task rather than the feature list. Write down the work you want to improve, the current process, the people involved and what a useful result would look like. A supplier may describe a tool as an assistant, analyst or agent, but those labels do not explain the actual business risk.
Ask whether the service is drafting, summarising, classifying, recommending or taking action. A tool that suggests wording has a different risk profile from one that changes records, sends messages or approves a transaction. A precise use case also stops a trial expanding into unrelated work before anyone has reviewed the consequences.
2. What information will the supplier receive?
List the information the system will handle at the beginning, not after it is connected. Include prompts, uploaded files, retrieved documents, conversation history, output records, account details and information sent through integrations. Separate public, internal, confidential and highly sensitive material.
Use the minimum information needed for the task. If an assistant only needs a short paragraph, do not provide a complete customer file. Remove names, addresses, account numbers and other identifying details where they are not essential. If the supplier cannot clearly explain the information path, treat that uncertainty as a risk that needs resolving before approval.
3. Where is the information processed and stored?
Ask where information is stored, where it may be processed and which other service providers may handle it. Location alone does not decide whether a service is suitable, but it affects privacy, contractual and operational questions. A business should understand the arrangements that apply when information leaves New Zealand.
Ask how long prompts, files and outputs remain available, whether backups follow a different retention period and how deletion requests work. Check whether information may be used to improve a service, train a model or create a shared product. Do not rely on a general statement that data is secure; ask for the setting or contractual term that applies to your account.
4. How does the supplier protect access?
Review account security, multi-factor authentication, administrator controls, role-based access and audit records. Find out whether the supplier can show who accessed information, changed a setting, connected a data source or approved an action. Individual accounts are safer to manage than a shared login because access can be removed when a role changes.
Ask what happens if an employee leaves, an administrator account is compromised or an integration is accidentally made public. Start with the smallest permission set. A writing workflow usually does not need unrestricted access to an entire file store, customer system or mailbox.
5. What happens when the output is wrong?
AI systems can produce convincing but inaccurate answers, omit important context or follow an instruction hidden in the material they are reading. Ask how the service exposes uncertainty, supports source checking and records the inputs or actions that led to a result.
Define a human checkpoint before the trial begins. The reviewer should check facts, figures, dates, recipients, permissions and the effect on the customer or worker. For high-impact work, the person with proper authority must make the final decision. A contract cannot transfer business responsibility to a generated answer.
6. How does the service handle connected systems?
Integrations are often where risk grows. Map every connection, including storage, forms, calendars, communications, payment processes and internal databases. For each one, ask what the service can read, create, change or delete.
Separate reading from acting when possible. A system may be allowed to prepare a draft without being allowed to send it. Require a person to confirm changes involving money, legal commitments, customer records, employment decisions or safety. Build a stop button and a fallback path before enabling automation.
7. What evidence supports the supplier’s security claims?
Ask for clear, current information about security controls, incident handling, access management, backups, vulnerability reporting and independent assurance. The exact document will vary by supplier, but vague marketing language is not enough for a material business dependency.
Consider whether the supplier has a way to notify customers about incidents and important changes. Check whether the service has a history of outages that would affect your process. A small business does not need to demand every enterprise document, but it does need enough evidence to match the tool to the sensitivity and importance of the work.
8. What does the agreement actually promise?
Read the terms that govern data use, confidentiality, retention, subprocessors, intellectual property, service changes, incident notification and termination. Keep a copy of the version accepted by the business. If the supplier can change important conditions without meaningful notice, record that as part of the risk decision.
Be careful with ownership language. A business may own its input and still have uncertainty about how outputs are generated, reused or checked. Decide how the team will review generated material before it is published, sent to a customer or used in a decision. Do not promise that generated content is accurate, exclusive or free of third-party restrictions unless that has been properly assessed.
Write down what happens during an outage, account lockout, pricing change or sudden feature removal. Can the business continue manually? Can it export prompts, documents, records and configuration? How long would it take to rebuild the process elsewhere?
Keep important source records outside the AI service. Do not make a supplier the only place where a customer history, operating procedure or business decision can be reconstructed. Test an export and a manual fallback before the tool becomes essential.
10. Can we leave safely?
An exit plan is part of the initial assessment, not a sign that the business expects failure. Confirm how to stop data processing, remove integrations, delete accounts, retrieve records and revoke access. Allocate an owner and a review date.
Use a simple risk register with the decision, the information involved, the controls required, the person responsible and the conditions for stopping the trial. Revisit it when the service changes, the business starts using more sensitive data or the tool gains new abilities.
Make the review proportionate
A low-risk experiment with public information may need a short checklist and a limited account. A service connected to personal information or business-critical systems needs stronger evidence, tighter permissions, a written agreement and a tested recovery process. Proportional does not mean casual; it means spending attention where a mistake would matter most.
Involve the people who understand the work. A technical person can explain access and integration risk, while the process owner can explain customer impact, quality requirements and practical fallback. Privacy, cultural and community considerations should be included when the information or use could affect people in ways a generic checklist does not capture.
Turn the assessment into a useful decision
At the end, record one of four outcomes: approve, approve with conditions, run a limited trial or do not proceed. State what evidence would change the decision. If the tool is approved, tell staff exactly what they may use it for, what information is prohibited, what must be checked and who answers questions.
A careful AI adoption process gives a small business room to experiment without pretending that every supplier is interchangeable. The best assessment is not the longest one. It is the one that makes the information flow, human responsibility, supplier promises, fallback plan and exit route visible before the first important task depends on them.
Frequently asked questions
1. What is an AI vendor risk assessment?
It is a structured review of an AI supplier, the information it handles, its access and security controls, its business impact, its contractual terms and the plan for recovery or exit.
2. Does a small business need a formal procurement process?
Not for every low-risk experiment. A short documented review is still useful, and higher-risk tools deserve stronger evidence and specialist input.
3. What is the first question to ask an AI supplier?
Ask what information the service receives, where it is processed and stored, how long it remains available and whether it is used for any purpose beyond providing the service.
4. Should an AI tool have access to the whole business drive?
Usually no. Start with the narrowest permissions that support the approved task and expand only when there is a recorded reason and an appropriate review.
5. How can a business check AI supplier security?
Request clear information about access control, encryption, incidents, backups, vulnerability handling, independent assurance and customer notification, then match the evidence to the risk.
6. What should happen if an AI supplier has an outage?
Pause dependent work, use the documented manual fallback, protect source records and review whether the recovery plan needs improvement before resuming automation.
7. When should a business get specialist advice?
Seek help when the tool handles sensitive personal information, core financial or safety processes, large-scale customer data or decisions with significant consequences.
8. How often should an AI vendor risk assessment NZ businesses use be reviewed?
Review it before a major new use, when the supplier changes its service or terms, when the information becomes more sensitive and at a regular interval set by the business.
Leave a Reply