Customer assurance

AI supplier assurance is the information and evidence a supplier provides so a buyer can assess the AI in its product. It explains what the system does, who is responsible, how risks are managed and what supports the supplier’s claims.

The request may arrive through a procurement questionnaire, a tender or a customer review. HumanWox connects the records used to prepare those responses, so the work can support future requests too.

Enterprise buyers commonly request information about the AI system’s purpose, models, providers, data use, testing, human oversight and incident handling. They also want to know who is accountable and what evidence supports the answers.

A useful starting point is to prepare:

• A description of the AI system and its intended use.
• Details of the models, providers and data involved.
• Named responsibilities.
• Risk assessments and the controls addressing those risks.
• Relevant test results, reviews and decisions.
• Procedures for changes and incidents.

The depth of the review depends on the buyer, the proposed use and its potential consequences.

Start with the AI system the buyer will use. Answer each question directly, describe what is in place today and identify the evidence supporting the response. Explain any limitations, outstanding work or questions that do not apply.

Have the relevant owners check the responses before submission. A planned control should be described as planned, with an owner and target date where available.

HumanWox lets you bring in a buyer’s questionnaire, prepare and approve responses from the governance record, and export supported buyer files with approved answers.

The evidence should support the specific claim being made. A testing claim may need dated evaluation results. An oversight claim may need a documented procedure and records showing that reviews take place. A risk acceptance claim may need the decision, rationale and approver.

Check that the evidence relates to the relevant system, version and use. A result from before a significant model change may need reassessment.

HumanWox links evidence to the governance records it supports, helping reviewers understand the basis of a response.

Your model provider’s documentation is useful, but it does not describe your complete AI system. Buyers may also need to understand your configuration, connected data, tools, safeguards and oversight.

For example, a provider’s model evaluation does not show whether your application prevents users from accessing another customer’s information.

Use provider documentation alongside evidence about your own implementation. HumanWox records the responsibilities, risks, controls and decisions associated with your AI system.

Yes, provided each answer remains accurate and addresses the new buyer’s question. Check the system, deployment, contractual context and supporting evidence before reuse.

Similar wording can conceal different requirements. A question about your own use of customer data may differ from one about your model provider’s use of it.

HumanWox offers approved answers for reuse and identifies the earlier buyer. A person reviews the proposed response before approving it for the new request.

Keep a record of the information supporting each answer and review it when the AI system changes. Relevant changes include a new model, provider, data source, feature, control or operating environment.

Assign responsibility for reviewing affected responses. Reusing a previously approved answer should include checking whether its supporting information remains valid.

HumanWox records the items supporting an approved answer. If their substance changes, the answer is flagged before export.