Once service messages are billable, the cost of a conversation is a function of how many messages it takes. Design decisions that used to be only about experience now have a cost attached.
1. Fewer turns
This is the largest lever, and it's mostly conversation design.
Collecting a delivery address conversationally (street, then number, then city, then postcode, then a confirmation) is five or six outbound messages. Collecting the same address in a single structured form is one interaction.
WhatsApp Flows exist for this: a form the customer fills in inside WhatsApp, returned as one structured response. Before October this was a UX argument. After October it's also a cost.
The same logic applies to automated agents. An agent that asks three clarifying questions before answering costs four messages. An agent that makes a reasonable assumption and offers a correction costs two.
2. Stop splitting messages
The cheapest change available.
Many implementations send several short messages in a row because it reads more naturally: a greeting, then the answer, then a follow-up question. That's three charges where one would do. Batching a reply into a single well-structured message with clear line breaks reads about as well and costs a third as much.
Look for this in automated flows especially, where teams often built in a “typing” rhythm on purpose, to feel human.
3. Watch your template categories
From Chapter 1.1: Meta can reclassify a template from utility to marketing, and the first you hear of it may be the invoice.
Two defences. Keep transactional templates strictly transactional: no promotional language, and no upsell tacked onto the end of a shipping notification, since that's what triggers reclassification. And monitor the category reported on your status webhooks against the category you expect, so a change surfaces in days rather than at month end.
4. Decide deliberately about the window
When a conversation is nearly resolved and the window is nearly closed, you have a choice: send now as a service message, or let it close and re-open later with a template. Neither is free any more, so the two have to be compared.
Broadly: if the follow-up is soon and short, staying in the window is usually cheaper. If it's days away, you'll be paying for a template regardless, so there's nothing to be gained by keeping the window alive with filler messages.
Where to start
Take your longest support conversation type by message count and ask a single question: which of these messages exist only because the previous one left something out?
Multiply that count by your monthly volume and your rate to size the saving.