When should I stop making changes to Black Friday campaigns?

Set a freeze date early enough that everything you want running on the peak day is already live and already out of the learning phase before it arrives. After the freeze, structural edits, new ad sets, bid strategy switches and tracking changes stop. Budget steps and, if genuinely needed, target loosening are the only moves left.

Should I lower my target ROAS for Black Friday?

Recalculate first, then decide. A discount cuts contribution margin, which raises the ROAS you need to break even rather than lowering it. In the worked example above, a 25% discount moved break even ROAS from 2.5 to 5.0. Whatever the correct target turns out to be, change it several days ahead so the system absorbs it before the auction shifts.

Can I launch new campaigns during Black Friday weekend?

You can, but a campaign launched inside the window starts with no delivery history at the point when auction prices are highest and mistakes cost the most. If a campaign matters for the weekend, launch it early enough that it has stabilised before your freeze date.

Why did my CPMs jump during peak week?

More advertisers bidding for the same inventory raises the price required to win an auction. That is a structural feature of how the auction works, not an account fault. The useful comparison is your own account in the same week of a previous year, not a published industry average.

What should I do with budgets in the week after peak?

Step budgets down gradually rather than cutting them off, and expect a period where automated bidding is still calibrated to peak conversion rates. On Google, look at whether a data exclusion for the anomalous days is appropriate; check the current official documentation before applying one.

AI for Black Friday Ads: Peak Season Bidding

AI for Black Friday ads is, in practice, mostly a question about bidding. For four or five days, automated bidding and automated budget allocation make frequent, rapid spending decisions against an auction that no longer resembles the one they learned on. Creative generation is the visible half of the story; the half that decides whether the weekend is profitable is the bidder and the budget engine underneath it. So this piece is about how those systems behave when auction conditions move fast, what you freeze before the weekend starts, and what is still safe to touch while it runs.

What AI for Black Friday ads actually controls

Three separate automated layers run in your account, and they fail in different ways. The bidding layer decides what to pay for a given impression or click, using a predicted conversion rate for that specific auction. The budget layer decides which campaign, ad set or asset group receives the next unit of spend. The delivery layer decides which of your creatives gets shown, which is why one asset can absorb most of a campaign's budget within hours.

Your job during peak is not to out-guess the three. It is to set the boundaries they operate inside, then leave them alone long enough to work. Most peak season damage comes from someone intervening at the exact moment the system was mid-correction.

Why the model's recent history stops describing the market

Automated bidding works from what it has recently observed. Over a peak weekend, two things change at once and pull your economics in opposite directions. More advertisers compete for the same inventory, so the price to win an auction rises. At the same time shoppers arrive ready to buy, so conversion rates rise. A bidder calibrated on an ordinary week meets both changes in the same hour.

The system does adapt, but it adapts by spending: it discovers the new conversion rate by buying traffic at the new price and watching what happens. That discovery period costs real money, and it lands on the most expensive days of your year. Conversion lag compounds it: a purchase that began as a Thursday browse and closed on Friday reports late, so the model optimises against an incomplete picture of the day it is reading.

Google exposes two named controls built for this situation. Seasonality adjustments tell Smart Bidding in advance to expect a conversion rate shift over a defined short window. Data exclusions tell it, after the fact, to ignore a period. Both sit in Google's official documentation, and the current guidance is worth reading yourself because they are built for short sharp events and are easy to misapply. Meta offers no direct equivalent switch, so on that side your levers are preparation and structure.

Your discount changes the number the bidder is aiming at

This is the part that quietly breaks peak campaigns. If you run a target ROAS strategy and you do not move the target when you cut prices, you are asking the system to hit a number that no longer means break even.

Take a product that normally sells for 1,000 TL and costs 600 TL to buy and box. Contribution is 400 TL, a 40% margin, so break even ROAS is 1 divided by 0.40, which is 2.5. Now apply a 25% Black Friday discount. The price becomes 750 TL, the unit cost is still 600 TL, so contribution falls to 150 TL and the margin becomes 150 divided by 750, or 20%. Break even ROAS is now 1 divided by 0.20, which is 5.0, exactly double the 2.5 you needed a week earlier. Same product, same cost, one discount, and the bar the bidder has to clear has doubled.

Run that calculation on your own catalogue before you set any target, after shipping, payment fees and expected returns, all of which push the figure higher again. The break even ROAS calculator handles the division across different margin assumptions. What you want out of it is one sentence: at my Black Friday price, the lowest ROAS I can accept is this. Every other decision hangs off that number.

The freeze list

Pick a freeze date and treat it as a deadline everyone touching the account respects. After it, these stop:

New campaigns and new ad sets. Anything launched inside the window begins with no history at the moment the auction is least forgiving.
Bid strategy switches. Moving from a cost based strategy to a ROAS based one mid weekend discards calibration you spent weeks paying for.
Optimisation event changes. Swapping the conversion event a campaign optimises toward restarts the learning problem from zero.
Structural edits: audience definitions, placements, and large budget jumps. Meta's own documentation describes significant edits as changes that can push an ad set back into the learning phase, and the current list is worth checking in the Help Centre.
Tracking and site changes. Pixel edits, server side event changes, theme updates, new apps. A tracking break during peak is not a small bug, it is the week your bidder goes blind.

The principle is simple: anything you want live on Friday should already be live, already out of learning, and already checked. That includes the discount code, the landing page, and whatever the site does when a variant sells out.

What you still change, and how

Freezing structure does not mean freezing budget. Budget is the one lever that has to move, and how it moves matters more than where it ends up.

Suppose the account normally runs 5,000 TL a day and you want 20,000 TL on the peak day, four times as much. Jumping there in one edit hands a stable campaign an entirely new spending problem overnight. Stepping up by roughly half each day gets you there instead: 5,000, then 7,500, then 11,250, then 16,875, then 20,000. That is four increases across five days, and the final one is only about an 18.5% rise because you have nearly arrived. The 50% step is not a platform rule but a discipline you choose; the useful output is knowing you have to start five days out rather than on Thursday night. Put the ramp into whatever automated scaling setup you use instead of trusting yourself to remember it at midnight. If raising spend without losing efficiency is the harder problem, the mechanics are in our guide to scaling PPC campaigns.

Targets move too, but early. If your break even figure doubles on Friday, your minimum acceptable ROAS target should have been raised days beforehand so the system absorbs the change while the auction is still ordinary. And set a real ceiling: a daily cap or lifetime budget that makes it impossible for an automated system to spend the weekend's money before Friday lunchtime.

Watching it run without breaking it

During the weekend you are diagnosing, not optimising. Three patterns cover most of what you will see.

Spend pinned flat against the budget all day means you are budget capped, and the only question is whether the next unit of spend still clears the break even number you calculated. Spend running well under budget on a target based strategy usually means the target is starving delivery, and loosening it is the lever, paid for with a fresh learning period. Spend on pace while ROAS collapses is a measurement question before it is a bidding question: confirm events are still firing and deduplicating properly, because a tracking failure and a performance failure look identical in a dashboard.

Whatever you do, change one thing at a time and note the hour you changed it. Standing automated rules earn their place as guardrails rather than optimisers, pausing on a hard loss threshold instead of chasing performance, and the practical patterns are in our piece on automated rules for Facebook ads.

The week after belongs to the same plan

Peak weekend data teaches your bidder that conversion rates are high and buyers are cheap to reach. December does not work that way and the model has no calendar. Expect a correction period in which the system spends against expectations the market no longer supports, and decide whether a data exclusion is appropriate for the anomalous days.

Budget should come down as a ramp rather than a cliff, for the same reason it went up as one. And hold your verdict on the results: revenue reported over the weekend is not revenue kept, because discount driven demand carries returns that land weeks later. The honest read on whether automation handled your peak season well is the net figure in January, not the ROAS column on Monday morning.