Your AI knows exactly and only what you’ve taught it. That's either the good news or the bad news.
There was a point last year where every customer support conversation somehow ended in exactly the same place. "We've got AI."
Ticket volumes? AI. Hiring challenges? AI. Long wait times? AI. Mercury in retrograde? AI. If someone had suggested AI could perhaps personally chase down delayed packages and have a word with the warehouse, most of the room would’ve probably agreed.
To be fair, some of the excitement was justified. AI has been one of the biggest leaps customer experience has seen. It takes care of all of the boring and repetitive stuff so your people don't have to. Used well, it's great.
However, the angle has changed since the massive hype cycle fizzled out. Over the past few months, we've heard far less blind optimism and a lot more healthy skepticism. The conversation isn't really about AI as a general thing anymore; it’s about the actual experience it produces. We’ve collectively accepted that AI belongs in modern support operations, but now we’re talking about where it belongs.
Customers don't care whether the first response came from a chatbot or a person, they care whether it solved the problem quickly, accurately, and without making them fight through three automated flows before finally finding someone who could actually help. That's especially true during BFCM.
If you've been getting away with something criminal operationally, peak season will find it. Your chatbot is nothing but part of that operation, which means, whether you intended it or not, it inherits all of your strengths... and all of your weaknesses.
It's easy to start thinking about your AI as though it's its own, siloed little project. The chatbot gets updated, the knowledge gets refreshed, someone spends three weeks tweaking prompts. Everyone nods approvingly when the bot successfully answers, "Where's my order?" for the fourteenth time in testing, and the project gets a pretty green tick. Job done. Except... your chatbot doesn't exist in isolation.
It's just yet another front door into the exact same operation your human team uses every day. It relies on the same documentation, the same policies, the same workflows, the same escalation paths, and the same collection of "temporary" workarounds that nobody remembers the original reason for.
Your experienced people have spent months, sometimes years, learning the rhythm of your business. Not the one on paper, the actual one. They know which issues are genuinely unusual and which ones happen every Thursday. They can spot the difference between a customer who's confused and one who’s about to churn. They know when following the process is the right answer, and when preserving the customer relationship is worth bending it just the right amount.
AI doesn't know any of that, nor should it; it works with whatever you've taught it. If the information it's pulling from is incomplete, contradictory, or spread across four different systems and one supports lead's collection of mental notes, the chatbot isn't making “bad decisions”; it's making the only decisions it can make.
That's why "getting the chatbot ready for BFCM" has always felt like an oddly narrow goal. You can refine prompts, test conversation flows, and optimize every intent you can think of until your fingers fall off, but none of that changes the operation sitting behind them.
Peak season will drag every problem your whole system has into broad daylight. All the AI can do is faithfully reflect what's already there.
One of the more interesting things we've noticed is that prospects have stopped asking us how much AI they should use and instead, they almost immediately start qualifying the conversation.
"We'll only use AI if it doesn't feel to the customer like we're cheaping out." That's a very revealing sentence, because it has almost nothing to do with AI and almost everything to do with trust.
Twelve months ago, the industry was mostly obsessed with raw capability. How many conversations could AI handle? How much volume could it deflect? How many people could it replace (jail.)? Somewhere along the way, the industry became so fascinated by what AI could technically do that we forgot to ask whether customers would actually enjoy experiencing it.
Turns out, they're surprisingly reasonable when the situation is reasonable. If automation gets them the right answer to a really simple question faster than a person could, most customers are perfectly happy to take the win and move on with their day.
What they do notice is intent. Customers can tell the difference between a business that's using AI to make life easier and one that's using AI to make payroll smaller. One feels like convenience; the other feels like being trapped in an escape room.
We've all experienced it. The chatbot answers a question you didn't ask, confidently points you towards the same help article you've already read twice, apologizes with impressive sincerity, and somehow concludes that what you really needed was another link to the FAQ.
Five minutes later you're typing "PLEASE LET ME TALK TO A HUMAN" into the chat window, hoping somebody, somewhere, takes the hint. That’s the operation showing through.
Businesses talk less about just “replacing humans” or “deflection” these days. They're much more interested in removing the repetitive conversations that never needed a human in the first place, so the people on their team have more time for the wonderfully weird edge cases that arrive with peak season.
Because, if BFCM teaches us anything, it's that customers have an uncanny ability to invent entirely new problems the moment order volume triples. And that's where human support is becoming the star of the show again.
Not because AI has suddenly become worse somehow; because the novelty has worn off, and the judgement has kicked in. We have become much better at recognising where AI genuinely improves the experience, and where it absolutely doesn’t.
One of the unintended side effects of the AI boom is that it's become much easier to spot what your best support folks were actually doing all along. Spoiler: it wasn't answering password reset questions.
Those were never the conversations that made experienced people valuable. If they were, we'd have settled the "AI versus humans" debate about fifteen minutes after ChatGPT launched and all gone home considerably earlier.
The value has always lived somewhere else. It's in knowing that this particular customer has already contacted you twice this week and probably deserves a little flexibility. It's recognising that a delayed parcel isn't really the problem because what's actually happened is someone's anniversary gift is currently enjoying a tour of Nebraska.
It's looking at a perfectly reasonable request, a slightly unreasonable policy, and deciding that preserving the customer relationship is probably worth more than winning an argument. None of those decisions come with a neat flowchart; they're judgement calls.
And, rather inconveniently for anyone hoping AI would solve absolutely everything, BFCM generates far more of them than a normal Tuesday in March. What happens when AI reaches the edge of what it's good at? How quickly can customers reach a person? Does the handoff actually make sense? Will somebody actually own the problem?
Customers don't judge your AI separately from your support team; they judge the experience as one continuous journey. They don't remember that the chatbot successfully answered three straightforward questions before the conversation went sideways.
They remember that, by the time they finally reached a human, they were explaining the whole thing for the fourth time and seriously considering whether the refund was worth the emotional investment. AI didn't change the support job; it just made everyone finally notice which parts of it mattered most.
The businesses seeing the best results from AI aren't necessarily the ones yapping about it most; they're the ones doing all the boring operational work nobody wants to put on a conference stage.
They're updating documentation before somebody asks for it, they're reviewing policies before customers discover the contradictions. They're making sure Marketing, Operations, Fulfilment and Support are all working from the same version of reality instead of four slightly different interpretations of it.
They're building an operation that already makes sense way, way before asking AI to sit on top of it. That sounds almost disappointingly obvious until you remember how many companies do the opposite.
It's much more exciting to launch a shiny new chatbot than it is to spend an afternoon updating returns documentation that hasn't been touched since someone redesigned the website eighteen months ago. Nobody's posting triumphant LinkedIn updates because they finally cleaned up their internal knowledge base.
Unfortunately, customers have an almost supernatural ability to notice the boring stuff. They don't care whether the inconsistency came from Marketing, Operations, Support, or your chatbot. They only experience one brand, which means every disconnect becomes your disconnect. That's why the phrase "AI strategy" has always felt misleading: AI doesn't really have a strategy of its own; it inherits yours.
If your operation is well documented, well maintained, and built around clear ownership, AI tends to look competent. If your business runs on outdated articles, Slack archaeology, and the collective memory of three people who've somehow survived every reorganisation since 2019, AI has a much harder time looking intelligent in any situation.
The irony is that none of those improvements were ever just for AI. Better documentation makes onboarding easier. It improves quality assurance. It helps new team members become productive faster. It gives experienced people somewhere reliable to check before relying on memory. AI just happens to benefit from exactly the same things your human team always has.
Which also means that preparing AI for BFCM is, in many ways, exactly the same as preparing your operation for it. You don't start with prompts; you start by making sure the business actually has the right answers.
The AI conversation has made human support more important, not less; and not because customers suddenly stopped liking automation. If anything, they're happier than ever to let AI handle the repetitive, predictable parts of the experience. The difference is that, once those conversations disappear, what's left are the ones that actually define your brand.
BFCM has always been a stress test for customer experience, and AI hasn't changed that, either. If anything, it's made the test a little more honest. Your chatbot isn't really being judged on how clever it is, it's being judged on whether it helps customers reach the outcome they wanted. That depends far less on the technology itself than it does on the operation standing behind it.
So, is the operation standing behind your chatbot actually ready?
There’s one fairly painless way to find out. Take our 2-minute BFCM Readiness Assessment and get a personalized receipt showing where your operation is solid, where the gaps are, and what’s worth fixing before peak season starts asking much less politely.