The BFCM
AI Agent Playbook

This is the first Black Friday where AI sits firmly and comfortably on both sides of the counter. On one side, merchants are handing more of their store to AI agents: answering customers, watching checkout, writing copy and even changing what's live on the site. On the other, shoppers are sending their own AI assistants to research products, compare offers and decide where to buy.

That raises a question most ecommerce folks haven't had to answer before: how much of your store are you willing to hand over, and at what point do you want a person back in charge? Get it right, and your agents take pressure off your team during the busiest days of the year. Get it wrong, and one bad decision repeats across thousands, causing more harm than it could ever cause good.

Why this BFCM is different

The shopper side of the counter is growing fast. Adobe says AI-driven traffic to U.S. retail sites had already grown 127% year over year by August, and it expects another 130% increase over the holidays. Shopify's data points in the same direction: in Q2 2026, AI-referred sessions to Shopify storefronts grew 197% year over year, orders from the channel roughly tripled, and half of those sessions landed directly on a product page rather than a homepage or collection page.

The merchant side is under more pressure than ever, too. Shopify merchants sold $14.6 billion over BFCM 2025, with sales peaking at $5.1 million per minute, and Adobe expects U.S. online holiday spending to reach $275.1 billion across November and December 2026, with Cyber Week alone accounting for $47.5 billion.

At that pace, there's very little room for an agent to get something wrong.

So this BFCM, there are really two AI questions.

The first is the obvious one: what should you let an AI agent do for your store while traffic is peaking?

The second is newer: what does your store need to do differently now that your customer may have an AI agent of their own?

We put the first question, and variations of it, to thirteen experts who build for Shopify merchants. Their answers follow the path a shopper takes over BFCM:

• Before the purchase: Tamir Eden (40rty) on AI shopping agents, Martin Cox (Native App Co) on content and creative, Moran Khoubian (Jurni) on ad landing page optimization, and Tina Donati (ContentCurve) on brand storytelling.

• During the purchase: Stoyan Dimitrov (Consentmo) on privacy and consent, Natasha Shine-Zirkel (Loox & Reviewers.com) on reviews and trust, Luca Borreani (Zipchat) on pre-sale conversations, Jake Casto (Layers) on the buying flow, Blake Waldron (Checkout Components) on checkout, and Michaela Lundberg (MindArc) on agent playbooks for peak week.

• After the purchase: Andrew Le (Cleverific) on order changes and "where is my order?" questions, and Connor Blakey (GoKwik) on post-purchase messaging.

Here's a summary of what they told us.

This is not an argument against autonomous agents. The point is to give AI freedom, but set clear limits.

That is also where security guidance is heading, and this October showed why. Attackers made unauthorized changes to the settings of several popular third-party apps on Shopify, and used them to redirect shoppers from checkout to fake payment pages. These were established apps from trusted developers, proving that this can happen to anyone, now more than ever. None of this in particular was caused because of overuse of AI, but the lesson carries straight over: anything that can change what shoppers see at checkout, whether it’s an app or an agent, is part of your risk. And it can happen to anyone, now more than ever.

It's also worth looking at your app stack as a whole. The surge of new Shopify apps shows no sign of slowing: there were 17,591 active apps in April 2026, up 43% from January 2025, and AI-assisted coding now lets a small team ship a working app in a fraction of the time it used to take. Plenty of these apps are excellent, but when shipping gets easier, testing often gets shorter, and not all of them have been through a Black Friday yet. Before peak week, check which apps touch your storefront and checkout, how long they've been around and whether they've handled peak traffic before.

OWASP lists excessive agency and prompt injection among the main risks when AI systems are connected to tools and can take real actions. Its guidance recommends least-privilege access and human approval for high-impact actions, instead of trusting the model itself to decide whether an action is allowed.

During a normal week, a poorly scoped agent may create a support headache. During BFCM, the same mistake can repeat thousands of times and be extraordinarily costly since you’ve already spent so much on getting visitors to your store.

Automate the effort. Be much more deliberate about automating the judgment.

In the chapters that follow, each of the thirteen experts explains what they'd hand to an AI agent this BFCM, what they'd keep with their team, and what they'd do before the traffic arrives. Read them in order, or go straight to the part of the shopper journey you want to learn more about.

This Black Friday, some of your shoppers won't start on your homepage, or even on Google. They'll ask ChatGPT, Gemini or Perplexity to find a gift, and an AI assistant will decide which stores make the shortlist. Tamir Eden, founder and CEO of 40rty, thinks that's where merchants should point their own AI agents first: at the content those assistants read. What he wouldn't hand over is the power to create discounts.

Tamir's biggest concern about Black Friday isn't the storefront itself. It's everything connected to it after the purchase: payments, fulfillment, shipping, reviews, email and SMS.

"Over BFCM, systems can see up to 100x their normal load," he says. "Your store may hold up, but the apps around it often don't."

In the worst cases, that means long delays, downtime or lost data.

Your store is being read by machines

When an AI assistant researches a holiday purchase, it doesn't only look at product pages. It reads your FAQs, your shipping and returns policies, your gifting information and your About page, and uses all of it to judge whether your store is a safe recommendation.

"If your return window or delivery cutoff is unclear, the agent will recommend someone else," Tamir says.

That's why auditing and fixing this non-product content is the first job he'd give an AI agent this BFCM: "This is low-risk, high-impact work that agents do well."

This is what agent-readiness looks like in practice. You're not writing for robots, you're making the facts a shopper needs, like price, stock, delivery dates and the return window, clear enough that a person and an AI assistant both get them right.

He also expects AI shopping to ease some of that pressure over time. As more people let personal AI assistants do their buying, stores will serve a growing share of traffic through protocols like MCP and Google's Universal Commerce Protocol (UCP) instead of full page loads.

"An agent doesn't need heavy images and videos for every session, just structured product data," he explains.

The stores that offer clean, structured data will handle peak traffic better.

Keep discounts away from customer-facing agents

The one thing Tamir wouldn't let an agent do is create coupon codes or promotions as part of customer support. With the right prompt, a shopper can talk an agent into a discount it should never give. His advice is to keep discounts behind human approval or strict rules set in advance.

"Over BFCM, when margins are already thin and volume is huge, one exploited prompt can spread fast," he says.

What to do now, before Black Friday traffic hits?

Shop your own store the way an AI assistant would. Open ChatGPT, Gemini or Perplexity and give it a real holiday shopping task, the kind your customers might type. For example: "I need a gift for my sister who loves hiking, under $80, that will arrive before December 20. What should I buy?" Swap in your own category, price range and delivery date, and try a few versions of the question.

Then look closely at the answers:

1. Does it find you at all? If your store doesn't come up for a request it clearly matches, the assistant either can't read your content or doesn't have enough information to recommend you with confidence.

2. Are the facts right? Compare the price, stock status, delivery cutoff and return window it gives with what's actually on your site.

3. Who does it recommend instead? If a competitor comes up, ask the assistant why it chose them. The answer often points to something they explain more clearly, like delivery dates or a gift-friendly returns policy.

"If it can't find you, gets your prices or policies wrong, or recommends a competitor, fix that before BFCM," Tamir says.

The good news is that the fix is usually in your own content: an FAQ that never mentions holiday shipping cutoffs, a returns page that doesn't say whether gifts can be exchanged, or product pages that leave out size, compatibility or who the product is for. Update the source, then run the same prompts again once the changes are live. But keep in mind - AI answers don't update overnight, which is one more reason to start in October rather than the week before Black Friday.

Martin Cox has a refreshingly calm view of Black Friday. "You shouldn't encounter technical issues on Black Friday if you are running a Shopify store," says the co-founder of Native App Co. For him, a strong BFCM comes down to the basics working together: inventory, merchandising, fulfillment and customer service. He'd happily use AI to improve the content that helps shoppers, and the AI tools they use, find your store. What he'd keep human is the creative work, like your ads, offers and campaigns, that makes your brand stand out.

Let AI optimize your content for search and AI discovery

Martin's first choice for AI this BFCM is optimizing your store's content for search and AI discovery, the same problem Native App Co works on with Product Pelican. Much of what sits on a store, like product descriptions, specs and FAQs, gets read more often by search engines and AI assistants than by people.

"Most of your content is stuff that humans are never going to read. That doesn't mean you should write for machines, though. Good GEO content is generally good content for humans to read anyway. So go hard with it."

His suggestion is to research what customers ask ChatGPT about products like yours, and get those answers onto your site. That's work AI can do quickly and at scale.

The timing matters too. Traffic from AI assistants is still a small share of most stores' visits, but it's growing quickly.

He compares it to the early days of SEO and social media, when the small and medium businesses that moved first saw the biggest gains.

"No one knows just how significant it will be this BFCM. If you get on the front foot, there's lots of possible un-forecasted growth to take."

Creative is where you stand out

The one thing Martin wouldn't hand to an agent is creative.

"Cut-through is getting harder every year, and you need great marketing that isn't 100% data and AI driven to cut through.”

BFCM is the noisiest week of the year. Inboxes, social feeds and ad slots fill up with near-identical "up to 50% off" messages. Now that every brand can use the same AI models, trained on the same data, to write its ads and emails, much of that content starts to sound the same. AI is very good at producing a version of what has worked before. In a week when everyone is doing that, "good enough" is easy to make and just as easy to scroll past.

Martin's answer is to put people on the ideas that are hard to copy.

"Whether it's founder-led, strong creative ads, use of ambassadors or just crazy offers, you need to push the boundaries."

In practice, that could look like:

• Founder-led content: the founder explaining why the brand is running a Black Friday sale (or deciding not to), or showing what goes into making the product.

• Ads with a real idea behind them: a concept customers will remember, not just a discount badge on a product photo.

• Ambassadors and real customers: people your audience already trusts, talking about the product in their own words.

• Bold offers: something unexpected that gets people talking, like a limited bundle or a gift with purchase, instead of another percentage off.

AI still has a role in all of this. It can help research angles, draft variations and resize assets for different channels. But the idea itself, and the call on whether it feels like your brand, should come from a person.

What to do now, before Black Friday traffic hits?

Understand your numbers. That's Martin's most important advice for this BFCM, and it has nothing to do with AI.

"Black Friday is a known quantity and you should have several years of historical data. Traffic is going to come, so make sure you know what traffic is going to generate the most profit for you."

AI can optimize whatever you point it at, very efficiently. If you point it at the wrong goal, like revenue at any cost, it will chase that too. So before you add another tool or agent, spend some time with last year's numbers:

1. Which products made you money, not just sales? Look at margin after discounts, shipping and returns, not only top-line revenue.

2. Which channels and audiences brought profitable customers? Compare what you spent on each channel with the orders and repeat customers it brought in.

3. Where did you leave money on the table? Check which best sellers ran out of stock, which offers ate too much margin and which campaigns drove traffic that didn't convert.

Once you know where your profit actually came from, use AI to speed up the work around it, from content and SEO to product pages, and keep your budget and creative energy on the products and channels that pay off.

Tina Donati, founder of ContentCurve, uses AI every day to get creative ideas out of her head and in front of her team faster. But there's one thing she won't hand to an agent, especially in the busiest week of the year: deciding what your brand actually says.

Tina doesn't think BFCM breaks anything new. It puts pressure on systems and teams that work fine the rest of the year. Support teams get more questions and complaints than they planned for, and fulfillment systems built for a few hundred orders a day can struggle when that number triples or quadruples.

She has felt it as a shopper, too:

"I've personally placed orders during major sales events, only to receive an email a few days later apologizing because the item was actually out of stock. There's nothing more frustrating as a customer, especially when you've passed up other deals because you thought your purchase was secured."

Use AI to go from idea to rough visual in minutes

One of Tina's favorite uses of AI right now is turning ideas for campaigns, ads and landing pages into rough concepts and wireframes. She doesn't publish what the AI produces, but instead uses it for a baseline, something to show her designers to illustrate what she has in mind.

"Anyone who's worked with designers knows how challenging it can sometimes be to explain exactly what you're picturing. Now, instead of sending over a long paragraph trying to describe a concept, I can show them a rough visual of what I'm thinking, and they can take it from there."

For her, the whole point of AI is not to replace the experts on the team, but to make their jobs easier and help them work smarter and faster. During BFCM, that speed can make a huge difference.

"BFCM is chaotic enough as it is. If AI can help us cut down on the back-and-forth, get creative ideas moving faster, and give our teams more time to focus on the work that actually requires their expertise, I'm all for it."

Your brand story needs a person behind it

Ask Tina what she'd never let an agent do on its own and she answers right away: brand storytelling and messaging. AI can help with brainstorming, refining copy and producing content faster, but the core message behind your brand and your promotions should come from someone who understands your business and your audience.

At around times like Black Friday, when you and 5 (or 50) other competitors are going after the same shopper, if your Black Friday email sounds like it could have come from any other store, it wouldn't stand out. AI can help you tell a story, but Tina wouldn't trust it to independently decide what that story should be.

She also points out that the biggest AI companies are hiring full-time storytellers and brand marketers, sometimes at salaries well above what content marketing roles usually pay. If the companies building these tools still invest that much in human creativity and judgment, merchants shouldn't be quick to hand theirs over either.

What to do now, before Black Friday traffic hits?

Stress-test the systems, and the people, responsible for delivering on your promises. That's Tina's advice for the weeks before BFCM, because, as she puts it:

"Getting the sale is only half the battle."

Her out-of-stock story shows why. A shopper who buys from you on Black Friday has often skipped other deals to do it. If the order then can't be fulfilled, or a support question goes unanswered for days, you lose more than one sale - you risk losing that customer for good.

Before the traffic arrives, test the parts of your business that have to deliver on the sale:

1. Plan for three to four times your normal volume. Ask your support team, warehouse or 3PL how they'd cope if orders and tickets tripled overnight, and what would break first.

2. Check that your stock levels are accurate. Make sure what your store shows matches what you actually have, especially for best sellers and bundles, so you never sell something you can't ship.

3. Prepare support before the rush. Write replies for the questions you know will come, like delivery times, returns and out-of-stock items, and set up AI tools to handle the repetitive ones so your team can focus on the harder cases.

The moral of her story: brands spend weeks planning how to win the sale, but it's those few hours spent on what happens after that keep the customer coming back. That might seem like a post-Black Friday thing to think about, but it's in fact something you should take care of now, before the orders start coming in.

Brands spend more on ads over BFCM than at any other time of the year. Then most of them send every shopper who clicked a specific offer, from a specific creative, to the same product page everyone else sees. Moran Khoubian, VP Growth & Strategy at Jurni, sees that as one of the most expensive mistakes of the weekend. She'd happily hand an AI agent the job of building a landing experience for every ad. What she wouldn't hand over is that page to a tool that doesn't know your brand.

On a normal week, a mismatch between the ad and the page costs you some conversion. Over BFCM, with a year's worth of buying intent packed into a few days, every mismatch gets multiplied, along with slow load times, confusing offers and checkout friction. And there's no fixing it next week, because the traffic and the money behind it - they are all happening now.

Let AI build a landing experience for every ad

Over BFCM, growth teams launch dozens of new creatives across Meta, TikTok, Google and email, each with its own angle, audience and offer. No team has the designers and developers to build a dedicated page for each one, so most of that traffic ends up on the default product page.

"That's the work I'd hand to an agent without hesitation, as long as it knows your brand and your products."

In Jurni's case, the team sets the strategy and the offer, and the agent turns each ad into a matching landing experience in minutes, then tests variations and recommends what to try next. This way, every expensive click lands on a page built specifically for it, while the team stays focused on the decisions only they can make.

Moran sees speed as the biggest change AI brings to peak season. Brands used to plan BFCM weeks in advance, launch everything at once and spend the weekend watching dashboards, because any meaningful change needed constant oversight and a designer, a developer or both. By the time it shipped, the moment had passed. With AI, that’s no longer the case: you see what shoppers respond to, find where conversion is leaking, generate a new experience and send each type of traffic to the page most likely to convert.

"Instead of one BFCM site for everyone, stores can react to intent as it happens."

Don't hand your most expensive traffic to a tool that doesn't know your brand

The one thing Moran wouldn't hand over is the page the most expensive traffic lands on, at least not to a tool that doesn't know the brand. General-purpose AI is great for a lot of things, but building a personalized landing experience takes context: your ads, your products and data on what actually converts, and ultimately - the expertise in knowing what makes a landing page a high-converting one.

"Over BFCM, when every session counts, that context is what keeps the page on-brand and true to the promise the ad made."

A generic tool can produce a page that looks fine and still say the wrong thing: an offer worded differently from the ad, a product angle the ad never mentioned, or a tone that just doesn't sound like you. A shopper who clicked because of a specific promise notices straight away when the page doesn't live up to it.

What to do now, before Black Friday traffic hits?

Make the landing experience part of the creative process. That's Moran's advice for right now: when you write the brief for an ad, brief the page it lands on at the same time.

"Most brands spend weeks perfecting the creative, then send it to a page that was never designed for it."

To get there before peak traffic arrives:

1. Pick the campaigns that matter most. Identify the campaigns, audiences and hero products that will carry most of your BFCM ad spend.

2. Brief the page with the ad. For each one, write down the promise, offer and message the ad makes, and make sure the landing experience carries all three through to purchase.

3. Build the destinations alongside the ads. Don't wait until the creative is finished to start on the pages. Build them in parallel so they launch together. And on time.

4. Test before the traffic hits. Run each ad-matched page on real traffic before BFCM, so you know which version converts while there's still time to change it.

In Moran's words, the brands that win BFCM "will be the ones whose page keeps the promise their ad made." Jurni's BFCM page has more on how they approach it.

Privacy rarely makes the BFCM to-do list, but Stoyan Dimitrov thinks it should. As Director of Product & Technology at Consentmo, a cookie consent and privacy compliance app for Shopify, he sees the same thing every year: the weeks before Black Friday add new risks, and the busiest weekend of the year is when they show up. His rule for AI is simple. Let it read, watch and summarize anything privacy-related, and keep every change in human hands.

In the run-up to the big weekend, merchants add new apps, pixels, popups and theme tweaks. Each one brings another script, and often a tracker that fires before the visitor has given consent.

On a normal Tuesday, you'd barely notice. On Black Friday, the consent banner loads late on mobile, Google Consent Mode is set up wrong so your conversion numbers are off right when ad spend peaks, and EU visitors get tracked without consent. Privacy requests keep arriving too, landing in the same inbox as "where is my order?", and the legal deadline doesn't move for Black Friday.

"Most of what breaks that weekend was already broken. You just didn't have enough traffic to see it."

Let AI watch, read and summarize

Stoyan's AI wish list is made up of tasks where the AI only reads, repeats or watches. Before BFCM, an agent can answer questions like "which apps I installed in the last 60 days added trackers?" or "which cookies aren't categorized?" During the weekend, it can send an alert when the consent rate drops after a theme change, a new tracker appears or a page slows down. It can tag support tickets, draft replies, route privacy requests to the right workflow and translate banner text into the languages you sell in, with a person checking the result.

His favourite is a daily summary, so the team starts the morning knowing what changed overnight "instead of opening five dashboards."

Consentmo has built this approach into its own product. Its new MCP connection lets merchants link their store's privacy data to Claude or ChatGPT and ask questions like "is my store compliant right now?" The AI only reads the data, so there's nothing it can break.

An agent should never touch consent settings

Anything that decides what customer data you collect shouldn't be left in the hands of AI. Things like consent banner settings, tracking scripts, Consent Mode and privacy policy text. The reason is simple: an agent can follow your instructions perfectly and still get you into trouble.

"If you ask an agent to fix a drop in tracked conversions, the fastest way to do it is to loosen consent. Fire pixels before the visitor chooses, pre-tick the boxes, hide the Reject button. Your dashboard will look better and you'll be breaking GDPR and CCPA for your biggest audience of the year."

The same goes for requests to access or delete personal data. An agent can draft and route them, but a person should check the customer's identity and press "Send", because those actions can't be undone. The AI suggests, but a person approves.

What to do now, before Black Friday traffic hits?

Audit everything, then stop changing things. Pick a freeze date, early November at the latest, and stop installing or changing anything on the storefront after it.

"Each script you remove makes the store faster and lowers your compliance risk, so you win twice."

Before that date, work through your store in this order:

1. List every app, pixel and script. Note what each one does and who on the team still needs it. Anything nobody claims is a candidate for removal.

2. Check what fires before consent. Open your main pages in a private browser window without accepting cookies, and see which trackers run anyway. Make sure every cookie is categorized and Google Consent Mode is set up correctly.

3. Test the banner on mobile. Check that it loads quickly on your key pages and that both Accept and Reject do what they should.

4. Decide what your AI tools may do. Write down who is on call over the weekend, which privacy tasks an agent can handle on its own and which always need a person, like changing consent settings or completing data deletion requests.

It takes some time, but as he already pointed out - it pays off twice: you get a faster store and much less risk during your biggest traffic weekend of the year.

Looking further ahead, he expects agents to watch the dashboards while people decide what to do about what they find.

"I think the stores that do well will be the ones that decided ahead of time what AI is allowed to touch, not the ones using the most AI."

On many stores, the biggest Black Friday risk isn't the traffic itself but everything that has piled up in the store before it arrives. Jake Casto, CEO of Layers, sees it most often on stores that test and change things all year. Over months of experiments, apps get installed and removed, and bits of them stay behind: old scripts, broken widgets and other leftovers that still sit in the buying flow when peak traffic hits.

Each of those adds a little latency, and smaller niche apps may not cope with Black Friday volume at all. An app that runs fine on any given day can still take the buying flow down the first time it sees peak traffic. That's why the first job Jake would give an AI agent is QA, checking every buying flow for problems before shoppers find them. What he'd never hand over is discounts, because during BFCM a wrong change reaches every shopper at once and there's no time to undo it.

Use AI for the QA nobody has time for

This is where Jake would put an AI agent to work without hesitation: QA.

An agent can walk through the same journeys your shoppers will take, from product page to cart to checkout, again and again, and flag anything that got slower or stopped working since the last run. It's tedious work for a person and exactly the kind of repetition AI handles well.

He sees this as part of a wider change in how stores operate.

"AI is enabling brands to move at a much more rapid pace, helping them move from reacting to on-site problems to finding and fixing them early. Let it go over all the buying flows and check for any oddities, slowdowns, etc.”

Why discounts stay with people

Discounts keep coming up across the different responses, and there's a reason for that. Jake is firm on this too: he wouldn't let an agent set up any discounts or promotions during BFCM.

"During BFCM, a bad change costs the most, and there is no time to unwind it.”

A "harmless" tweak can clash with another promotion or break what you agreed with a partner, and it reaches every shopper on the site at once.

What to do now, before Black Friday traffic hits?

Finish your changes, then freeze. That's Jake's top advice for the weeks before Black Friday. The closer you get to peak traffic, the less one more change is worth, and the more it can cost if it goes wrong.

Getting there takes four steps in the weeks before Black Friday:

1. Set a freeze date. Pick a day after which no new apps, scripts, theme changes or experiments go live, and tell everyone who can touch the store.

2. Clean out the leftovers. Go through your installed apps and theme code, and remove anything left over from finished tests or tools you no longer use.

3. Test the full buying journey. A homepage speed test isn't enough. Run the paths shoppers will actually take, from product page to cart to checkout, on mobile and desktop, with your real BFCM promotions switched on.

4. Keep testing during the freeze. Let an AI agent repeat those journeys every day and flag anything that gets slower or breaks, so you catch problems without changing anything.

A freeze can feel like lost time when there's still so much you could improve. But a store that stays the same through BFCM is one your team understands, and when traffic peaks, knowing exactly what's running on your site is worth far more than one last tweak.

"Trust breaks faster than infrastructure."

That's how Natasha Shine-Zirkel, CBO at Loox and Reviewers.com, sums up Black Friday. On any other week, a slow review widget or a referral code that won't apply is a small annoyance. During BFCM, shoppers have five tabs open and decide in seconds, so the same glitch sends them to a competitor.

"A missing star rating can cost you the sale."

That's why she sees AI as a big help with reviews, and why she'd stop it one step before it speaks for the brand.

Reviews are a pattern-spotting job

BFCM brings a flood of reviews and support tickets, and that's where Natasha would put AI to work first. An agent can sort incoming reviews and tickets by urgency and sentiment, flag anything a person needs to see quickly and draft replies to common questions.

The bigger win is spotting patterns while there's still time to act. If twenty reviews in a day mention that a product runs small, the value isn't in writing twenty thank-you notes. It's in telling your team before the next shopper orders the wrong size. In Natasha's words, it lets you "fix a sizing issue or a shipping complaint the same day instead of in a December post-mortem."

Why a person should approve every public reply

The step Natasha wouldn't hand over is publishing replies to reviews.

"An agent can draft a thoughtful reply in seconds, and at BFCM volume that saves your team hours. But one wrong reply, like promising a discount that doesn't exist or arguing with an unhappy customer, lands in front of 10x your usual traffic, and it's hard to walk back."

So the agent drafts and a person signs off. Natasha sees that split as the key to getting through BFCM with a lean team: AI takes the triage, drafting and pattern-spotting, and people keep the final say on anything that touches brand trust or money.

"The stores that draw that line before BFCM come out with happier customers and a team that isn't burned out. The ones that draw it mid-weekend usually draw it too late."

What to do now, before Black Friday traffic hits?

Collect real photo and video reviews for your best-selling products before the traffic arrives. That's Natasha's most important advice for right now.

"Seeing customers like them is what closes the sale. It works even when your price is higher or your discount is smaller."

During BFCM, shoppers compare your offer against a dozen others, and with so much AI-generated content around, they want proof that real people use and love what you sell. A few weeks is enough to build that up if you start now:

1. Pick the products that will get the most traffic. Your BFCM hero products and best sellers are where reviews will do the most work.

2. Ask recent customers for photos and videos. Send review requests to people who bought those products in the last few months, and make it easy to upload a photo or a short clip from their phone.

3. Put the proof where shoppers decide. Make sure star ratings show on collection pages and near the add-to-cart button, and that photo reviews load quickly on mobile.

4. Set clear rules for your review agent. Decide which reviews it should flag right away, like complaints about sizing, quality or delivery, and who on the team approves its draft replies before they go live.

Even a handful of real customer photos on your biggest products can do more for a hesitant shopper than any amount of polished copy. They show exactly what the shopper is about to buy, through the eyes of someone like them, and that kind of trust is hard to fake in a week full of AI-generated content.

Luca Borreani is comfortable giving AI agents a lot of room. As CMO and co-founder of Zipchat, which builds AI agents for ecommerce sales and support, he'd let an agent handle pre-sale questions, social replies, follow-ups, pre-orders and most customer support over BFCM. There's only one job he'd still keep with a person: approving refunds and returns.

Hand the agent customer support and pre-sale conversations

Luca would hand an agent the two areas where he sees Black Friday sales slip away most often: customer support and pre-sale conversations. In his view, this is exactly where an AI agent does an excellent job.

Support comes first. During BFCM, people buy quickly because they're worried the deal or the stock will run out, and a few days later many of them want a refund or a return. Answering fast isn't enough at that point. Support has to reassure people and, where it makes sense, give them a good reason to keep the order. That's hard for a team that isn't online around the clock and doesn't speak every language your customers speak, but it's well within reach for an agent.

Pre-sale is the second. A shopper clicks an ad, lands on a product page and has a question before buying: will it arrive before Christmas, does it work with something they already own, which size should they pick now that their usual one is sold out?

“If nobody answers within 3 to 5 seconds, the sale is gone. That's the moment Zipchat's AI product page assistant is built for.”

Beyond those two, his list covers most of the conversations around a purchase:

• Replying to every comment and DM on social media

• Handling every pre-sale conversation on the site, explaining why this product and why this promotion

• Booking pre-orders

• Following up on conversations that got interrupted

• Automating most customer support

The goal is that every shopper knows everything they need to know before buying, and ends up convinced to buy.

He'd also let an agent create small personal discounts on the fly, such as an extra 5% on top of the running promotion, valid for a few hours. If you try this, have the agent choose from offers you've approved in advance, with a fixed maximum, an expiry time and a limit per customer, rather than letting it invent new ones.

Luca expects agents to take on even more soon. Customer support, and the actions agents can already take, will run largely on their own, which takes a huge load off teams during peak days. He also sees AI-powered QA becoming standard: systems that keep checking the site for pages that don't load, pages that load too slowly and buttons that don't work, and alert the right people the moment something fails.

Refunds need a track record first

"A year or two ago I would have given you a long list here.This year, with agentic systems, there's only one thing I wouldn’t hand over: approving refunds and returns."

For him, it's about trust that's earned over time. Unless an agent has run the refund flow for at least six or seven months, across a lot of real cases, he'd keep a person on it.

"It's money leaving the business, and every case has its own context. One wrong call repeated a few hundred times over a weekend gets expensive fast."

Until the agent has earned that trust, the handover is what matters most. It can collect the order details, explain the policy and then pass the conversation to a person with all the context attached, so the customer doesn't have to start over. The same applies to anything outside the agent's rules, like a customer who's clearly upset or a question it can't answer with confidence. An agent that knows when to step aside is worth more over BFCM than one that tries to handle everything on its own. And fails.

What to do now, before Black Friday traffic hits?

Switch on an agent that can run the whole customer journey. That's Luca's advice for the weeks before BFCM. One agent that has full context on your brand, and remembers each customer and the channel they prefer, can sell more and handle most of your customer service. That leaves your team free to focus on what matters most over BFCM: bringing in traffic and turning it into sales.

If you're setting up or expanding an AI agent before BFCM, a few things decide whether it helps your team or creates more work for it:

1. Give it your BFCM details. Load your promotions, shipping cutoffs, return policy and stock updates before the sale starts, so it can answer shoppers in those crucial 3 to 5 seconds.

2. Set limits on offers. If you let it offer personal discounts, decide the maximum amount, how long the offer lasts and who qualifies.

3. Keep refunds and returns with a person. Let the agent collect the order details and explain the policy, and pass the final decision to your team.

4. Test it like a shopper. Ask it the questions you expect most, in every language you sell in, and check that it hands over to a person when it should.

Get those four things right, and the agent takes the routine conversations off your team's plate exactly when there are the most of them, while the decisions that cost money stay with people.

Blake Waldron, founder of Checkout Components, has seen what peak traffic can do to a checkout. Last year, a merchant came to his team after their checkout extension provider couldn't handle the BFCM volume, which led to failed checkouts, slow load times and data that didn't add up.

Burst traffic exposes limits that never show up during normal trading. On Black Friday, inventory, discounts, targeting rules, analytics and integrations all run at the same time, so one weak link can disrupt the whole journey. That's why Blake treats every third-party service in the checkout as part of a store's capacity plan, and why he thinks monitoring is where AI does its best work over Black Friday. What he'd never let an agent do is make live changes to checkout on its own.

An extra pair of eyes on checkout

Monitoring is the job Blake would give AI first. An agent can watch checkout completion, response times, error rates, upsell attachment, discount behavior and data consistency across every provider at once.

"It can flag unusual patterns early, help isolate which extension or integration may be under pressure and summarise the likely impact for the team.”

It can also prioritize support requests, spot recurring issues and prepare suggested replies, as long as clear thresholds, escalation paths and human approvals are in place. For Shopify merchants, he has a specific tip:

"Sidekick is going to be your best friend this year, because it lets you build real-time dashboards for almost any question you want answered.”

Finding the problem isn't the same as fixing it

Where Blake stops the agent is at live changes. He wouldn't let it publish anything that affects checkout availability, pricing, discounts or promotion logic.

"If performance drops, an agent might correctly identify an extension as the source, but automatically disabling it could create another issue or remove something essential to checkout.”

During peak traffic, any change reaches a huge number of customers very quickly. So the agent monitors, diagnoses and recommends, and someone who understands the technical dependencies and the commercial context approves the fix.

What to do now, before Black Friday traffic hits?

Test the complete customer journey under realistic promotions and peak load. That's Blake's top advice, and he means the whole journey.

"Don't only confirm that a banner, discount or upsell appears.”

The test should also reflect the volume and number of shoppers you expect during BFCM, with every extension, integration and third-party provider in checkout switched on.

The most reliable way to do that is to go through checkout the way your Black Friday shoppers will, with everything switched on at once:

1. Test as different shoppers. Go through checkout on different devices, in each market you sell to and with each payment method, and check that the right customer sees the right banner, discount or upsell.

2. Stack your promotions. Turn on every BFCM offer at once and check that discounts don't clash and inventory updates correctly.

3. Ask your providers about load. Confirm that every app running in checkout has been load-tested for the traffic you expect, and find out how it behaves under pressure.

4. Give every critical system an owner. For each extension and promotion, decide who owns it, how you'll know it's failing and how you'll roll it back.

The aim, as he puts it, "isn't to remove people from BFCM operations, but to give them earlier warnings, clearer information and more time to make the right decision."

Michaela Lundberg sees Black Friday from the side of the teams who keep large Shopify stores running. As Head of Marketing & Partnerships at MindArc, ANZ's first Platinum-tier Shopify agency, her view on AI agents is practical: give them the same kind of rules you'd give a developer working on your live store. An agent can watch the store around the clock and even make small fixes, but only the ones your team has agreed on in advance and written down in a playbook. Anything that changes what a shopper pays stays with a person.

In Michaela’s experience, Shopify itself rarely experiences downtime during Black Friday. The apps and systems connected to it are a different story. Review widgets, search, shipping rate calculators, loyalty programs and ERP syncs all run on every visit, and at peak volume one slow app can stall a product page or checkout.

The one merchants tend to miss is inventory sync. It keeps up fine on a normal day, then falls behind under a burst of orders, and suddenly you're selling stock you no longer have.

"A sell out turns into a week of refunds and apology emails."

The sale itself goes through without a problem, so nobody notices until the cancellations start. By then, you've disappointed the shoppers who were most eager to buy from you.

Run peak trading around the clock, without a round-the-clock team

Most teams still find out about a problem when sales drop or a customer complains. Michaela sees AI agents changing that.

"Peak trading becomes a 24 hour operation without a 24 hour roster."

An agent can watch site speed, checkout errors and app failures through the night, point to the likely cause and flag it to the team, often before a shopper notices anything. Instead of waking up to a sales dip and a pile of complaints, your team starts the morning with a short list of decisions and a store that kept trading overnight.

She'd also hand an agent first-line customer service, since questions about order status, delivery times and returns spike over BFCM and an agent trained on your policies can answer them straight away. And she'd use one to tidy product data before peak, from titles to attributes and missing details, so AI shopping assistants can find and recommend your best sellers. That matters more every year: according to Shopify's 2026 Holiday Retail Report, 62% of Australians expect to use AI for at least one holiday shopping task this year.

Give your agent the same guidelines as a developer

The one thing Michaela wouldn't hand to an agent is anything that changes what a shopper pays. Pricing and discounts stay with a person, every time.

"Over BFCM one wrong rule gets multiplied across thousands of orders in minutes, and the cost lands in margin, refunds and customer trust before anyone notices."

Her rule for everything else is simple: put the same controls on an agent that you'd put on a developer. A good developer doesn't push a pricing change or a major design update to a live store during peak week without someone reviewing it first. An agent shouldn't either. It can flag that a discount looks off and suggest the fix, but a person approves it.

Where an agent can act on its own is inside a playbook your team has already agreed on. If switching off a non-essential app that's slowing checkout is in that playbook, the agent can do it at 3am without waking anyone up. The difference is that the decision was made calmly, weeks earlier, by people who understand the store, and not improvised by an agent under pressure.

What to do now, before Black Friday traffic hits?

Stress-test the path to checkout, then write the playbook. That's Michaela's most important advice for the weeks before BFCM.

1. Load-test for your expected peak. Run a load test that mirrors the traffic you expect on Black Friday, not an average day.

2. Buy something on your phone while it runs. Watch for slow scripts, timeouts and apps that fail with no fallback, then fix or remove anything that stops a shopper from paying.

3. List every integration and app on your product pages, cart and checkout. Note what each one does, who owns it and how to switch it off.

4. Decide what the agent may do on its own. Go through that list and mark the switches an agent can flip without asking, like turning off a non-essential app, and the ones that always need a person.

That list becomes your playbook if something slows down on the day, for your team and for your agent. MindArc's Black Friday 2026 checklist covers the rest of the preparation in more detail. As Michaela puts it, "A fast store that takes payment will outsell a beautiful campaign on a slow one."

Most BFCM advice is either about getting through the weekend or preparing for it. Viktoria Philbrick, Head of Community at Uxify, starts somewhere else: with the number you use to judge it. Black Friday is the biggest stress test your store gets all year, and the most useful thing it can show you is how much each visitor is actually worth. That's what revenue per visitor (RPV) measures: your total revenue divided by the number of visitors, which brings conversion rate and order value together in one number.

It's also a more honest number than conversion rate alone. Over BFCM, traffic and conversion rate both go up, but discounts pull order value down, and many of the shoppers you paid to bring in never come back. RPV makes you ask two more useful questions: how much is each visit really worth, and what's stopping it from being worth more? Answer those for your BFCM traffic and you've learned something that applies to every other week of the year. Uxify's view on AI follows from that: let agents do the continuous, measurable work of making every visit worth more, and keep people in charge of the decisions that set your margins.

Let AI agents make every visit worth more

Most of what drags RPV down is small and spread across thousands of sessions: a page that's a second slower on mobile, a wait between pages that makes shoppers lose interest, a cart left behind when someone closes the tab. No team can watch for all of it in real time, least of all during peak traffic. This is the work agentic DXP is built for: agents that observe real sessions, find the friction, act on it and measure the result, without waiting weeks for a test cycle.

In Uxify, that work is split across a few agents:

• Speed across the whole journey. Shoppers rarely leave because of one slow page. It's the delay that repeats page after page that wears them down. Navigation AI predicts the page a shopper is most likely to open next and prerenders it, so it appears almost instantly.

• Saving carts before they're abandoned. Carter steps in with a browser-native prompt when a shopper tries to close the last tab with items still in their cart, and merchants pay only when a cart is actually saved. Last month, it saved about 130 carts from being abandoned in the first place for one Australian fashion brand.

• Catching regressions the day they happen. Vita watches Core Web Vitals and flags a regression as soon as it appears, with a short note on the likely cause. On a weekend like BFCM, when a lot is happening on the site at once, only having to look through the changes that matter is a big help.

• Seeing where visitors and revenue leak. Funnels are built automatically from the paths real visitors take, so you can see where they drop off and how much revenue leaves with them.

None of this has to stop when you freeze your store before BFCM. Freezing your theme is good advice, but it doesn't have to mean freezing improvement. Uxify's agents work on top of your live store instead of changing theme code, so they can keep optimizing through the freeze without adding risk to it.

Keep the goal and the trade-offs with your team

What Viktoria wouldn't hand to an agent is deciding what "worth more" means for your business. An agent optimizes whatever it's pointed at, and some ways of lifting revenue per visitor today cost you later: a deeper discount that raises conversion but eats into your margin, an aggressive popup that wins the sale but annoys the customer, or a push on upsells that brings more returns in January.

That's why picking the right measure matters as much as picking the right agent. An agent told to raise conversion rate can do it by giving away margin, and the dashboard will still look like a win. Agents can find the friction and improve the experience around it, but your team should set the goal, decide which trade-offs are acceptable and keep everything commercial, from pricing and discounts to the promises your brand makes.

What to do now, before Black Friday traffic hits?

Start measuring revenue per visitor before the traffic arrives. That's Viktoria's main advice for this BFCM, because a baseline taken now is the only way to see what the weekend really did for your business.

1. Record your baseline. Note your RPV, conversion rate and average order value for the last 30 days, split by mobile and desktop.

2. Find your biggest leaks. Look at the pages and steps where most visitors drop off, and check how fast your key pages load for real visitors on mobile. Uxify recommends aiming for a Largest Contentful Paint of 1.5 seconds rather than Google's 2.5-second threshold, because shoppers are less patient than the average web visitor.

3. Switch on what can run through a freeze. Agents for speed, saving carts and monitoring can all go live before your change deadline and keep working while the theme stays untouched. If you want to try them before the traffic arrives, our BFCM page has everything you need to get started.

4. Compare after BFCM. Look at RPV for the weekend and for the weeks after it, and check which improvements kept paying off once traffic went back to normal.

Do this, and you'll come out of BFCM with more than a sales total. You'll know which pages, steps and visitors actually make you money, and where to keep improving once traffic goes back to normal.

Ask Andrew Le what breaks on Black Friday weekend and he doesn't hesitate: support. As founder of Cleverific, which lets Shopify shoppers edit their own orders after checkout, he watches support teams get buried every year under the same handful of questions. Most of them are "where is my order?" (WISMO) and tracking questions, followed closely by requests like "can I change my shipping address?", which take more work to resolve.

Address changes are more common than you might expect. In Andrew's experience, express wallets and autofill tools like Apple Pay and Google Pay now account for 40% to 60% of payments, and they often fill in an old or default address. Shipping address changes make up about 60% of all the order change requests Cleverific sees. None of these requests needs a person to solve it, which is why Andrew would hand them to AI and, wherever possible, let customers make the fix themselves before it ever becomes a ticket.

Let AI answer, then let customers fix it themselves

"The number one thing to offload to an AI agent is WISMO.”

Tracking questions, promotion questions and "what can I change in my order?" belong with AI too, since the answers are things you'd publish anyway.

But Andrew doesn't think AI should stop at answering. For a shopper who ended up with the wrong address, the slow route goes from the AI to a support agent to the operations team before the order is fixed. The fast route is for the AI to understand the request and send the shopper to a self-service page where they can fix it in a couple of clicks.

The difference isn't only about speed. A shopper who spots a wrong address right after a Black Friday purchase is already worried, whether about the gift arriving on time or about the money they've just spent. Every extra question and every hour of waiting adds to that worry, and a small slip starts to turn into a bad memory of your brand.

Letting shoppers fix it themselves gives them back control straight away, so the purchase feels safe again. It also keeps a large share of these requests from ever reaching your support team, which matters most on the weekend when that team is stretched the thinnest. In the moment, that's all the shopper wants: as Andrew puts it, to "click a button or two" and move on.

Frustrated customers need a person, fast

The one thing Andrew wouldn't hand to an agent is a frustrated customer. A good agent setup has to read the customer's mood and hand over to a person as soon as it starts to turn. Otherwise, he warns, "your agent runs the risk of turning an unhappy customer into a frustrated customer, and then into a customer who never wants to shop with you again."

It's a useful rule for any BFCM setup: hand over when the risk goes up, not only when the AI doesn't know the answer. An agent can know exactly where a package is and still be the wrong one to continue the conversation.

There's another point that should be made here too. Over a busy BFCM weekend, a customer with a damaged order or a missed delivery often waits in the same queue as hundreds of people asking where their parcel is. By the time someone gets to them, they've had hours to sit with the problem, and by now they've convinced themselves to never shop with you again. When AI takes the simple questions out of the queue, those customers reach a person much sooner. For Andrew, that's the real change AI brings to peak season: not a smaller support team, but a team with enough time to handle the hardest conversations with care and empathy.

What to do now, before Black Friday traffic hits?

Make sure your AI agent can answer last year's most common questions. That's Andrew's advice for right now. Questions like where an order is, or how to change, cancel or update it, should get a clear answer and a direct next step, ideally a link where the customer can make the change themselves.

Last year's support history already shows what this BFCM will look like, so it's the best place to start:

1. Pull last year's top questions. Have AI export your tickets from last November and December and group them by topic. WISMO, address changes and order edits will likely be near the top.

2. Write one clear answer for each. Keep it short, and end every answer with what the customer can do next, like a tracking link or an order-editing page.

3. Let customers fix their own orders. Set up self-service for the changes you see most, like shipping addresses, sizes and quantities, within limits you control. As Andrew puts it, customers "just want to click a button or two."

4. Decide when the agent hands over. Set clear triggers for passing a conversation to a person, especially when a customer starts to sound frustrated.

For more on what happens after checkout, Cleverific has a guide to four post-purchase strategies that increase customer lifetime value.

For a lot of brands, Black Friday conversations happen outside the website: over email, text message and WhatsApp. Connor Blakey, Partnerships Director at GoKwik, sees AI agents doing much of the work there, from order updates and tracking questions to product suggestions and subscription changes. What he'd keep with a person is your brand's tone and any conversation with a high-value customer, where one badly judged automated reply can do lasting damage. But his biggest warning is about something far less exciting that can stop a WhatsApp campaign before it starts: the health of your account.

"Brands try to send massive, unmanaged promotional broadcasts to cold databases, triggering spam reports and instant Meta rate-limiting or daily account caps right when volume peaks."

Why does this happen? When a list is cold, people report the messages as spam, and Meta reads that as a sign of a low-quality sender. It ties how many messages a WhatsApp Business account can send to the account's quality, so a channel has to prove itself before it can send at scale. In Connor's experience, it takes around 60 days of steady utility messages, like order updates and shipping notices, to unlock the highest daily send tiers, which makes October and early November the time to build them up.

Let the agent handle the everyday questions

Once the channel is ready, Connor sees AI working on both sides of the purchase. Before checkout, an agent can recommend products based on what's in the shopper's cart. After it, the agent can answer order status and tracking questions and handle one-tap subscription changes, like pausing or skipping a delivery, right inside the chat.

For Connor, this moves support away from a reactive ticket queue and towards an ongoing conversation. In many stores, a question asked in a chat widget or a DM still ends with "leave your email and we'll get back to you," and the shopper has to explain everything again in a new thread hours later. When the agent can answer inside WhatsApp, the conversation stays where the shopper started it, with the order details already at hand. The agent resolves the most common questions on the spot and recovers abandoned carts in real time. If a person does need to step in, the ideal setup lets them pick up the same thread rather than starting over, so the team can focus on the conversations that really need them.

High-value customers deserve a human

Those are the conversations Connor would keep with people. He wouldn't give an agent control over brand tone, or over how escalations with high-value customers are handled.

The reason has a lot to do with how these customers feel about your brand. Someone who has bought from you many times doesn't think of themselves as a ticket number. They expect to be recognized, and messaging apps make that expectation stronger, because a message in WhatsApp feels personal, closer to a text from a friend than an email from a company. When something goes wrong for a customer like that, a generic automated reply tells them the opposite of what they want to hear: that nobody knows who they are. In Connor's words, a poorly context-aware response sent to a high-LTV customer during peak trading "can permanently damage brand perception."

A loyal customer whose problem is handled with care, by someone who understands their history with the brand, can end up trusting you more than before. That's why Connor wants a person in the loop at these moments: they're where loyalty is either lost or made stronger.

What to do now, before Black Friday traffic hits?

Warm up your WhatsApp channel and protect your account health now. That's Connor's most urgent advice:

"Warm up your messaging channels and secure your Meta WABA health score immediately."

The send limits you'll have on Black Friday depend on how you use the channel in the weeks before it, so this can't wait until November.

1. Check where you stand. Look up your current messaging limit, quality rating and approved templates in WhatsApp Manager.

2. Send useful messages now. Use WhatsApp for order confirmations, shipping updates and delivery notices through October and early November. Steady messages that customers expect and find helpful build your account's track record.

3. Clean your list. Only message people who opted in and are likely to engage. A big broadcast to a cold list is the fastest way to get spam reports and lower your limits.

4. Get templates approved early. Submit your BFCM promotional templates well ahead of time, so you're not waiting on approvals during the sale.

5. Plan the handover. Decide which conversations the agent handles on its own and when it passes high-value or unhappy customers to a person.

None of this is exciting work, and it's easy to put off until the campaign creative is ready. But on Black Friday, the brands whose messages actually get through are the ones that spent October earning the right to send them.