Automate the answer you already approved. Keep the decision you have not.

Sample staff time for one normal week

Ask the people who answer messages to log each request type and roughly how long it took. A simple tally sheet is enough. The goal is to see which questions repeat, which take the longest, and which need a decision rather than information.

• Request type
• Channel
• Minutes spent
• Information or decision?

Write and approve the knowledge

For each repeated question, write one approved answer with an owner and a review date: hours, locations, services, preparation steps, delivery areas, documents needed. The workflow answers only from this approved information. If the answer is not there, it says so and hands over rather than guessing.

Define escalation before launch

List what always goes to a person and who receives it: complaints, refunds outside policy, medical, legal or safety questions, unhappy or vulnerable customers, and requests the information does not cover. WhatsApp’s policy also expects a clear route to a human for any automated conversation.

Illustrative case (not a real customer)

Illustrative: a customer asks whether a branch has parking and whether a specific card is accepted. Both answers are in the approved information, so they are answered directly. The same customer then asks for a refund on a cancelled appointment outside the policy window. That is a decision, so it is handed to staff with a summary: who, which appointment, what was asked and what was already answered.

Hand over with context

A handover should never make the customer repeat themselves. Pass the summary, the details already collected and the reason for escalation. Track how long handed-over conversations wait for a person.

Review monthly

Each month, look at unanswered questions, handovers and corrections from staff. Add approved answers where the same question keeps coming back, retire outdated ones, and repeat the staff-time sample every few months to see whether repetition actually fell.