Should we use AI to write a difficult message to a customer after we have made a mistake?
Use AI to help organize a draft, not to decide what your company owes the customer or send the message without review. The person responsible for the relationship should establish the facts, decide what can be offered, and choose whether a call is more appropriate than an email.
Give the tool only the information needed for the draft. Ask it to state what happened, acknowledge the impact, explain the next step, and flag details that still need confirmation. Avoid asking it to sound empathetic while leaving out facts that would help the customer understand the situation.
Before sending, check every factual claim, date, commitment, and proposed remedy. Remove language that sounds evasive or more certain than the evidence allows. If you do not yet know the cause or resolution, say so and explain when the customer can expect an update.
A forward-deployed engineer could build a Lovable or Replit workflow that prepares drafts and routes them to the accountable team member. It should make review and editing straightforward rather than treating approval as a formality.
This approach follows the EOS emphasis on clarity and accountability while preserving the human relationship. AI can reduce the effort of composing a message, but a real person should own the response and be available for the conversation that follows.
Category: AI that stays human