An explanation is not a technical dump of feature importance. A customer needs to understand what happened, which information materially influenced the result, what can be corrected and how to request a review. The ICO's guidance on explaining AI decisions is a practical reference for designing that communication.
Explain the actual decision path
Separate the model's contribution from policy rules, human judgment, data checks and external information. If a decision was partly automated, explain the roles in plain language. Do not give a generic statement that AI was used if the customer cannot understand the consequence.
Build contestability
Provide a route to correct data, submit relevant information, request human review and receive a response within a defined time. Monitor whether review changes outcomes and whether particular groups face higher referral or complaint rates.
Preserve the evidence
Keep the input snapshot, model version, reason codes, threshold, reviewer action and communication. The ICO explaining-AI guidance and AI Act should be considered alongside sector-specific rules.