Playbooks · 6 min read
Black Friday Is a Load Event, Not a Creative One
Most Black Friday advice is about the offer. Discount depth, urgency, the countdown, the second email on Cyber Monday. That advice is not wrong, it is just about a different machine than this one.
On the comment-and-DM side, BFCM is not a creative problem at all. It is a load problem. The words you send are the words you would have sent in September. What changes is that four days of ordinary volume arrive in about six hours, on the one weekend where a delivery problem costs you the most.
Everything below is a thing to set before the sale, because during the sale you will be doing something else.
The shape of the weekend
A normal week gives you comments spread across days, arriving faster than you can answer but slower than anything can break. A sale weekend gives you a post that travels, a discount code people want, and several hundred comments in an afternoon.
Two consequences follow, and only the first one is obvious.
The obvious one is that you cannot answer them by hand. That is what the automation is for and it handles it.
The less obvious one is that an uncapped automation will try to answer all of them as fast as the API accepts. A large volume of near-identical messages leaving one account in a short period is exactly the pattern that produces a delivery problem, and the problem outlasts the sale by a long way. You would be trading a good Saturday for a bad December.
The cap looks like a limit on the busiest day of your year. It is insurance against the busiest day of your year.
The setting that matters most, and why the maths works
rate_budget turns the spike into a queue. Instead of six hundred messages in twenty minutes, they go out steadily over some hours, and the ones at the back of the queue are answered later in the day.
That is survivable because of which clock they are on. A comment can be replied to for seven days after it was posted. Somebody who commented at 11am and hears from you at 4pm is still comfortably inside the window, and on a sale weekend they are still inside the sale. The delay costs you very little and prevents the thing that costs you a lot.
It does not work the same way on DM traffic, where the clock is twenty-four hours and starts from their message. Know that difference before you set the number.
The pre-flight list
- Write the sale replies as templates, not as AI prompts. Templated replies are free and unlimited. A credit is spent only when the model writes something new, one credit per reply. On a weekend where several hundred people ask the same question about the same code, there is nothing to compose — and a template is also the only version you can read in full before it goes out to everybody.
- Set the rate budget deliberately. Not the default, not off. Pick a number you would be comfortable seeing sustained for several hours, and remember that the seven-day comment window is what makes the queue harmless.
- Put the sale keywords on a campaign automation, never on your evergreen ones. Any sentence containing a date, a price, a discount or a stock claim belongs to something you will switch off. Adding
BF30to your general price automation "just so it catches everything" is how a Black Friday code is still being quoted in March. - Write the negative keyword list before the trigger list. Sale weekends bring their own traffic:
where is my order,not delivered,cancel,refund,wrong size. A cheerful discount template is the worst possible answer to any of them, and those messages need a person. - Decide quiet hours in advance. Sales run overnight, and people comment overnight. Either you narrow the window and accept 2am replies, or you keep it and accept a morning queue. Both are defensible. Discovering the setting at 1am is not.
- Find the kill switch before you need it. It is the first check in the gate for a reason. Know where it is, and tell whoever else has access.
- Rehearse on a small post the week before. Build it, run Test on myself, put it live on one quiet post, then read Activity the next morning. A rehearsal that finds one wrong sentence has paid for itself.
Two things that go wrong specifically at volume
Duplicate arrivals. Instagram redelivers events, and people comment twice because the first one did not seem to do anything. dedupe and cooldown handle it, but if you have three automations live on the same post with overlapping keywords, you are asking one person to receive three messages. Before the weekend, read your live automations as a set and check that no two of them could fire on the same comment.
Old posts waking up. A reel from last month gets shared into the sale traffic and starts collecting comments again. If your evergreen automation is scoped to all posts, it is now answering sale-adjacent questions with pre-sale answers. Scope the campaign automation to the campaign posts, under Advanced, and check what the evergreen ones are attached to.
The Monday after
Two jobs, and both of them get skipped every year.
Export the leads as a CSV first, before you touch anything, because that is the irreversible step. Then switch the campaign automations off — all of them, from the list, not from memory — and delete the sale keywords from anything evergreen you touched.
Then write down, that same day, what you would change: which keyword nobody typed, which question came up eighty times with no template, what the negative list should have contained. In eleven months you will remember none of it. The teardown, in full, is its own discipline and it is the part that decides whether next year starts from something clean.
If you have not built the underlying flow yet, start with where it leaks rather than with the sale, and the full list of what can stop a send is the thing to read before you set a single cap.
