How to Run Your Brand With AI This Black Friday

Say it's 2 p.m. on Black Friday and the campaign behind your best deal has just run out of budget. Should it get more?

The answer sits in four places. The ad console tells you whether the campaign is still inside its ACOS target. The inventory report tells you how many days of cover the ASIN has at today's sell-through. The deals page tells you whether the deal is still live, and the offer itself tells you whether you still hold the Buy Box. Two of those can flip the answer outright. With two days of cover left, more budget sells out the deal stock before Cyber Monday traffic arrives. If the offer has lost the featured spot, the extra budget won't serve at all (Amazon Ads).

Until recently, answering that properly meant three exports, a spreadsheet and a lookup you hoped was joined on the right column. Half an hour later you had a number for one campaign, which had been dark the whole time. A dashboard answers the questions its builder thought of. In our experience, brands on the same software end up looking at the same standard charts, each number on its own screen. The question that matters at 2 p.m. on Black Friday needs all four at once.

At Texin we run Amazon and ecommerce for brands, and this year we built how we run our clients' busiest weeks around AI tools. Our clients' data is connected to AI, and so are the controls: the ad and listing settings it can change once we've approved the rules. That 2 p.m. question now takes a sentence to ask, and the setup below, in the order we'd build it, is what lets a small team ask ten of them a day.

Your deals are locked. How you run the week isn't.#

If you sell on Amazon, the big holiday decisions were made weeks ago. Deal submissions closed in October, and so did the cutoffs for getting inventory into Amazon's warehouses in time. We start our clients' holiday planning in the summer for exactly that reason. By November, the units you'll have and the deals you'll run are set.

What's still open is how you run the week. Amazon's Black Friday and Cyber Monday event window for sellers runs from November 19 to November 30. Ad budgets and bids can change any time. Prime-exclusive discounts can still be submitted until 12 hours before the event ends. And you can decide today what you'll watch, what you'll do when something goes wrong, and who acts on it. With stock and deals fixed, most of the levers left decide where traffic goes and how fast you react when a product stops converting.

It's worth getting right. Adobe expects Cyber Week, Thanksgiving through Cyber Monday, to bring in $47.5 billion online. That's 17.3% of the whole season's spending, packed into 5 of its 61 days (Adobe). An average Cyber Week day is forecast to bring in more than twice the online spending of an average day in the rest of the season, so a problem that takes half a day to notice costs more that week than almost any other.

Start with the goal#

Everything else depends on knowing what you're trying to do. Every plan we build for peak season starts with a written goal for each brand or product line, and the goals can point in opposite directions.

One brand we work with is using the season to clear excess inventory, and pushing a few high-margin items that are well placed to win sales against their competitors. Another wants to move inventory and grow total sales, so it spends up to a set share of sales on advertising, which buys it more visibility while keeping the economics in check.

Same week, opposite budget logic. For the first brand, ACOS on the overstock ASINs is measured against what it costs to carry those units into the new year, so a campaign can run well above the brand's usual target and still be the right call. The high-margin items get spend only where they win the side-by-side on price and reviews, and the rest of the catalog gets very little. The second brand runs on a TACoS target, so its budget rises with the day's total sales. On Black Friday it can spend far more in dollars and still be on plan, as long as the ratio holds and organic sales climb with the paid ones.

The goal also writes the watch list. The clearance brand tracks units left on each overstock ASIN against days left in the event; the growth brand tracks TACoS through the day and whether total sales are rising with the spend.

Connect your data and ask it questions#

When your sales, ads, inventory and deal data are connected to an AI like Claude or ChatGPT, the 2 p.m. question costs a sentence. So do the questions that cut across systems, which in our experience are the ones dashboards handle worst. "Which deal ASINs have under four days of cover at today's sell-through, and how much ad spend are they still getting?" You ask, you get the answer from current data, and you ask the next question.

The real gain is depth, not saved hours. On a peak day, a person used to have time to dig properly into one or two things. Now they can dig into ten, across the whole business, on current data. In our experience, the time between the data arriving and a decision being made shrinks to minutes. When a deal draws traffic and doesn't convert, we look at the page and the offer first: our price on other channels, other sellers' prices on the listing, delivery date, images and reviews. A team that finds it at noon still has most of the day to fix it.

Alert only on what you can still change#

Once the data is connected, have your AI watch the whole business and send you only what you can still act on this week. Everything else goes in a daily summary.

Area	Alert when	First response
Stock	A deal ASIN or top seller has fewer days of cover than days left in the event, at today's sell-through, counting inbound that will be live before the event ends	Shift spend to in-stock variants and alternatives before the item goes dark; flag the reorder for January
Deals	A live deal changes status, or a price Amazon checks it against moves	Find which price moved: your own promo on another channel (end or align it) or a reseller on the listing (a person decides whether to go lower); if the deal won't come back, move spend to live deals
Pricing and Buy Box	A top ASIN loses the featured offer, or shows up in Pricing Health	Check who holds the offer and at what price; align your own lower price elsewhere; a person decides whether to match a reseller
Ads	An on-target campaign will run dry before evening at its current hourly burn, or clicks keep coming with no orders	Add budget to the campaign that's on target; cut bids on the spend that isn't converting
Listings	Title, images, variation family or a key attribute differs from the approved version	Restore the approved values, then find the source: an Amazon catalog update, another seller's contribution or a bad upload
Questions and reviews	A question or complaint repeats on a deal ASIN, or a low rating lands	Draft the answer and the listing fix for approval

For Halden Outdoor, a made-up brand, a day of those alerts might read like this, each with what to do next:

• "Trail Pack 32L: 4 days of stock left at this week's pace, and the event runs 7 more days." Move ad spend to the 40L and slow the ads on the 32L.
• "Ridge Headlamp deal went inactive at 2:10 p.m. Your Walmart price dropped to $24 an hour earlier." Align the Walmart price; if the deal won't come back today, move its budget to the live deals.
• "Featured offer on Summit Stove lost to a reseller at $54.99 (your price $59.99)." Whether to match is a pricing call for a person.
• "Camp Chair is $39 on your own site and $44 on Amazon." End the site promo before it costs you the deal or the featured offer; a new Amazon price is a person's call.
• "Campaign 'Packs, exact' runs out of budget by 3 p.m. and is under its ACOS target." Add budget within your limits.
• "Search term 'waterproof hiking backpack' spent $86 since Monday with no orders." Cut the bid or add it as a negative; on a deal ASIN, check the offer first.
• "Trail Pack 32L: three questions since Monday ask whether it fits a 17-inch laptop." Approve the drafted answer and add laptop size to the bullets.
• "Trail Pack 32L in Green dropped out of its variation family." Put it back and find out what changed it.

Claude can run a scheduled task in Cowork against your connected tools, as often as hourly (Anthropic), and ChatGPT has scheduled tasks too (OpenAI). Write the table's triggers into one prompt, set it to run every hour through the event, and tell it to report only when something crosses a line. The checks only see what's connected, and Portal connects your Amazon sales and ad data to Claude or ChatGPT and keeps the daily history current.

The questions to ask your AI this week#

Alerts flag what moved. We use these questions to find out why and to spot what's working, written close to how we'd say them.

In the morning#

• "Which products are below plan, and is it traffic or conversion?" Traffic down points to ads, rank or a lost deal. Conversion down points to price, the offer, the delivery date or a fresh bad review. It's usually our first question of the day, because traffic problems and conversion problems have different fixes.
• "Which products are ahead of plan, and do they have the stock and budget to keep going through Cyber Monday?" A winner that sells out or runs out of budget on Thanksgiving night costs as much as a broken listing.
• "Over the last two weeks, which hours carried the most spend for the fewest orders?" The answer shows which hours to bid down so the budget lasts into the evening. It needs Amazon's hourly ad data connected.

Through the afternoon#

• "For any deal that dropped today, what changed first: our price somewhere else, a seller's price on the listing, or the stock?" If it's a reseller priced below you, the clock matters: the lowest price from any seller on the listing counts against your deal price (Amazon Seller Central), and Amazon re-checks discounts every 24 hours (Amazon Seller Central).
• "Which deal items are getting traffic but converting below their normal rate, and what's different on the page?" Look at price, delivery date, images and new reviews before touching the ads.
• "Which of the products we compete with most dropped their price this week, and where are we now the more expensive choice?" A competitor's cut matters where your conversion fell with it. Everywhere else, holding price protects the margin you'll want in December.

At the end of the day#

• "What changed on our listings in the last 24 hours that we didn't change, and where did it come from?" If Amazon's catalog or another seller changed it, restore it and check it again tomorrow; if it was your own upload, fix the file.
• "What changed in the account today, by us or by the system, and did each change hold?" The answer opens the daily summary.

Ask them out loud#

Most of those questions run a sentence or two, with an ASIN, a time window and a condition in them, which is quicker to say than to type. Said out loud, a question picks up the context a typed one leaves out. Typed, it's "days of cover on gift set deal ASINs." Spoken, it's closer to: "For the three deal ASINs on the gift set, how many days of cover do we have at today's pace, and if any are under four, which in-stock variant should pick up the spend?" Last week's issue of Signals, our newsletter, was about shoppers who talk to AI in paragraphs (Voice-to-Text Plus AI Is Rewriting Product Discovery). Operators get the same gain when they talk to their own data, and on Black Friday that means asking mid-call or while you're looking at the deal page. During peak week, I ask most of these out loud myself.

The setup takes minutes. Dictation is built into every Mac: switch it on in Keyboard settings, then press the microphone key (or Fn twice) and talk into the Claude or ChatGPT box (Apple). Wispr Flow does the same in any app on Mac, Windows, iPhone and Android (Wispr Flow). Claude and ChatGPT also have voice modes that talk back (Anthropic; OpenAI), which we use away from the desk. At the desk we dictate into the regular chat, because we want the answer on screen as a table we can scan.

Asking faster only helps if acting is fast too, which is where the controls come in.

Connect the controls too#

If the AI can only tell you what to do, a person still has to click every button, and that person becomes the bottleneck on the busiest day of the year. The setups that work for us connect the controls as well, so approved changes actually happen. Every change goes through official APIs: the system previews the change, makes it, then reads it back to confirm it took. Variations break more often than you'd think, and for one brand we run they now get fixed the same day instead of reported in next week's meeting.

A change that waits hours for approval on a peak day has missed much of the day, so we agree before the week starts on who decides what:

• The AI handles small, routine changes and reports them. For example, adding up to about 25% to a campaign's budget once a day when it has run out and is still meeting its goal, or restoring a listing detail that changed from the approved version.
• Bigger changes get a quick yes from a person. Larger budget moves, shifting money between products, or spending near the week's ceiling.
• New prices, deals and any new customer-facing content stay with people. Restoring a value a person already approved is the one exception.

The goal sets the limits. Budget goes up where a campaign is meeting its goal, which is also how Amazon's own budget rules work (Amazon Ads): for the clearance brand that means the overstock is selling through at an ACOS it can live with, and for the growth brand it means TACoS holds. Amazon's new Seller Assistant workflows can keep watching "even when the seller isn't logged in to Seller Central," and when a workflow acts, the seller approves each change first (About Amazon). We let small, reversible changes go through without approval and report them in the summary, because on a peak afternoon an hour in an approval queue can mean an hour of a campaign sitting dark.

Three things to do before November 19#

1. Write the goal, and the one number that tells you it's working. "Clear the overstock without going below this margin." "Grow total sales while ad spend stays under this share of sales." Every limit you set comes from it.
2. Write your questions and pick your alerts. Start from the list above and rewrite it around your goal. For each alert, write the trigger as a threshold and give it an owner and a first response.
3. Connect it, set the limits, and test it. Connect the data through Portal or your own build, and write down what the AI can change and by how much. Then trigger each alert once on purpose this week, so the first time you see it isn't on Black Friday.

Keep it running through December and January#

The week doesn't end on Cyber Monday. Last season, Cyber Week brought in $44.2 billion of the $257.8 billion spent online over the holidays, so most of the season happens outside it, including December gift shopping (Adobe). January brings returns, with retailers expecting about 17% of holiday sales to come back (NRF). It also brings leftover inventory, and Amazon's holiday peak fulfillment fees run until January 14. The same goals, questions and limits carry straight through.

Whatever you set up this month is also the first real test of how you'll run Prime Day and next year's holidays. Write down what you watched, what you changed, and what you wish you'd seen sooner. Start next year's setup from that list.

If you'd rather have a team that already runs it this way, that's what we do.

---

Texin.ai runs Amazon for brands and builds the content that helps them get found in AI search and by shopping agents. Talk to us.