*"You have a new customer inquiry that has not been responded to for more than one minute. To protect the rights of your store, please reply as soon as possible."*
If you have ever opened a takeaway shop and read the above line, you may have subconsciously sat upright.
In the past six months, whenever I hear this female voice anywhere - in my own store, on my own mobile phone, or even on other people's mobile phones on the street - my whole body becomes tense. That conditioned reflex is exactly the same as when an office worker hears an alarm clock. The only difference is: your alarm clock rings once a day, and mine rings every day.
I got PTSD from an AI landing.
The reason why I fell into this problem is because in the past six months, I did something that many colleagues think is a bit "performance art" - instead of sitting at my desk writing PPT, adjusting skills, and competing with Claude Code's quota, I actually opened a flower shop.
One of the most controversial questions on the Internet recently is: Can AI be put into practice? It’s been three or four years since 2022, and it’s time for an explanation. I thought that instead of facing off against others in the group, it would be better to go out on the field myself. So today's article is not another fantasy about "AI changing thousands of industries", but a review with blood and cocoons. I will show you every pitfall I have stepped into in the past six months as it is.
If you are also working on AI implementation and are stuck in some unclear confusion - I hope you can see yourself in it.
Why is it a flower shop
First, let’s talk about the background: why we want to do this.
The reason is simple. There is really not much space online. The increment is very limited and all is stock; the cost of attracting customers is getting higher year by year, and acquiring customers on the Internet is already ridiculously expensive. I've said this countless times on the show.
Then an idea came to mind: Can those of us who work on the Internet, with the set of things we have learned and the set of playing methods we have accumulated, turn around and add AI to help those who do business offline? Help them reduce costs and increase efficiency?
This story seems familiar. There was a time of "mouse and cement". But this time and the other - the reason for adding cement back then was that the Internet itself had not yet found a decent business model; but today it is just the opposite. The Internet's business model has become extremely mature, and the traffic economy has reached its peak. The only problem is: it is full and it is expensive. **Since online is so expensive, does the offline cost appear cheaper? Can we use our online skills back to offline?
After choosing and choosing, I finally landed on an industry that didn't look sexy at all: terminal flower retail—the kind of flower shop you see every day on the roadside. There are big chains like The Brutalist and Roseonly that are open in big shopping malls, and there are also mom-and-pop shops that are just a shabby shop with a lady sitting inside to sell you flowers.
Why flowers and not something else? I admit that there is an element of intuition here, but I do have a few hard standards stuck in my mind. I won’t end up missing any of them:First, it has to be small enough and specific enough that I can open one myself. If AI is used to empower and sell systems, then if I have never done it myself, why should I tell others that my set of things can help you make money? I have to go through all the pits myself first. This rule excludes all asset-heavy and high-threshold industries: those that I can’t get out of and can’t afford to lose; and the start-up cost and complexity of a flower shop fall within the range that I can personally verify.
Second, it had better be a "digital desert". The more humble, low-level, and untouched the industry is by IT and automation, the greater the space that AI can leverage. If you go to a flower shop - you have to hand-write orders to take orders. There is no system in the shop, and the only thing that has some "Internet" touch is takeout. Precisely because it started almost from scratch, it was the turn of an outsider like me to talk about "empowerment"; if it were a mature industry that had been plowed several times by SaaS, what more could I do if I got in?
Third, it must have a ready-made online entrance. My Internet style of play has to be inserted somewhere to make it effective. The flower shop is highly dependent on the food delivery platform to gain customers - this just means that it has been half pushed and half "online": there is traffic, data flow, delivery, customer service, and contract fulfillment. These links are all our home ground as people working on the Internet. To put it bluntly, a flower shop is an "offline business that has always been online," rather than a piece of iron that is completely insulated from the Internet. This is where I can enter.
Fourth, it must be able to afford a set of tools. For an industry that cannot even pay for tools, no matter how big the space is, it is still a mirror image (such as the Chinese podcast industry). But the gross profit of the flower shop is astonishingly high (this will surprise you later) - high gross profit means that the merchant has the ability to pay, and the unit economic model will not collapse as soon as it starts. Coupled with the fact that there are such mom-and-pop shops all over the country that have not been well served, this is a long-tail market that is large enough.
Being small enough to survive, being in a digital desert, having access to the internet, and being able to afford the tools—all these four factors taken together, a flower shop is almost the only answer. So, we seriously opened a flower shop in Tianjin.
But I have to add honestly: the above four points are all reasons that I can see clearly when I stand outside the door. And in the past six months, there have been places that are the same as what I thought; there are even more places that are completely different from what I thought - and these "differences" are exactly the beginning of all the stories that follow.
The first slap in the face: I thought we should help them reduce their losses first
What is the first reaction to reducing costs in this industry? Reduce losses.
This sounds so logical - flowers are fresh, and isn't the vitality of freshness just loss? The flowers themselves are not expensive, so why are they sold so expensive? Isn’t it just to cover the cost of the batch that rotted in your hands? I think so, even the first suggestion AI gave me was this. Make sense, right?
It turned out not to be the case at all. Flower shops don't care about losses at all. Why? Because the profits of this industry are high enough. So high that they don't feel any pain at all.
Don't look at the prices listed at the flower markets in Beijing's Wangsiying and Tianjin's Caozhuang. They are all "individual customers' prices" that have already been marked up for you. The real purchase price is much lower than what individual customers can get. This is only a regional terminal wholesale market. For example, the prices of upstream wholesalers in Dounan are even lower.
How can a wholesale market owner tell if you are an insider? You go in with a black plastic bag and say "Boss, give me a dozen orange tulips" - This is called buying flowers, because no florist shop owner only buys one type at a time, just like when you go to the clothing wholesale market, if you don't buy all sizes, you must be an individual customer. Therefore, insiders purchase goods like this: orange, white, and yellow, five bundles each. With a gesture, the boss will know whether you want this bowl of rice, and the price you will be given will be different.
I can give you a reference: when I buy a batch of mixed and matched goods, I can recoup my investment after selling about five or six bunches, and the rest is pure profit. Therefore, for them, loss does not hurt at all, and selling the flowers is the first priority. Only when the losses are so expensive that it makes people feel distressed, "cost reduction" will be established; otherwise, "open source" will always take precedence.
This is the first slap I have received, and it is also the first sentence I want to send to all my colleagues:
The pain points you think are often discovered by patting your head. And if you ask the AI, it will be very considerate and accompany you to take pictures. So if you ask the AI to make market adjustments, what it comes up with is basically hallucinations and nonsense.
Later, at the Beijing Flower Show, I saw the only company in the audience that made IT systems for this industry. One of its main functions is "inventory". We asked them on the spot: What is the usage rate of your inventory function?
I won’t describe the other person’s expression. You taste.
Second slap: People cannot be standardized at all
The road to reducing losses is unavailable, so what about improving efficiency? Before improving results, you must first understand "people".
I first discovered a counter-intuitive fact: it is impossible for florists to look at their mobile phones and computers while they are working. **
Those of us in offices take it for granted – aren’t you allowed to stare at a screen and take orders? Aren’t you allowed to read takeout orders? Later I realized that when people were cutting flowers, both hands were occupied and they couldn’t free their hands at all. That’s why all food delivery software is equipped with such harsh voice announcements (the one at the beginning that gave me PTSD), and at the same time spitting out receipts.
But the receipt from a flower shop is different from that at a restaurant. In addition to ordinary thermal receipts, it will also spray a whole piece of A4 paper - in addition to the order information, there must also be a picture of what the bouquet of flowers looks like. Why? Because there can be five to six hundred types of flowers hanging in one store, the florist can't remember them at all; what's more terrible is that the flowers are non-standard products. The chef knows how to make Kung Pao Chicken, because all the side dishes of Kung Pao Chicken are similar all over the world; but for "red roses with baby's breath", if you ask a hundred florists to make it, they can make a hundred variations. If a customer takes a pair of original pictures and they are different, they can reject them.So the easiest and most helpless way is to give you a piece of paper and just do it exactly the same. Don't think, don't act, don't use your brain.
At this point, I suddenly understood why Convenient Bee failed.
What Bianlifeng wants to do is to use machines to control every movement of people - down to giving instructions to the clerk on the mobile phone: "It's time to sort out the goods in container No. 5" and "Move the bread in container No. 3 to container No. 4." It wants to completely turn offline people into execution units driven by algorithms. It wasn't done. I don't think it can be done either.
Because offline work cannot be so standardized, there are countless operations that require on-the-spot judgment and responsibility for goals. You can't write everything into the machine's processes and turn everything into Skills - you can tell him to move the bread to a container, but you can't let him judge on the spot whether it would be easier to sell it by moving it to the door. This is something that the machine can't calculate, and the calculation can't cover such a fine granularity.
It has to be said that for most people, it is difficult to be responsible for the goal. Even operating according to the SOP is a very difficult "extra work" for them, and they should not do it if they can be opportunistic. The same goes for flower shops. Those who work according to the drawings cannot actually be called florists. They should be called "big workers". He has no design or art, he only cares about the drawings. This group generally has a low level of education and generally does not have the ability to delay gratification.
I've tried all the "internet-style" incentives and failed:
Issuing shares? Useless. You said you could realize 100,000 yuan next year, and he said, "No, no, no, you give me 10,000 yuan today."
Give commission? Raising the commission from 10% to 15% will have approximately zero incentive effect. Because when a person's income is very low, he cannot see that far - he may not even know where he will be in a month. He only cares about how much money he can take home today.
Talk about the long term? Don't. For this group, pie is a negative asset.
For offline business, as long as you want to recruit people, there are only three truly effective management methods: first, give enough money; second, be able to keep an eye on them; third - don't believe it - if you talk nicely, he won't listen, so he has to scold you. **Hitting is no longer allowed, so only scolding is left. This is a very helpless thing. He is barefoot and is not afraid of wearing shoes. You have no long-term interests to bind him. If he is unhappy today, the worst he can do is leave.
So what model can really work? At the end of my trip, I found that it was still the same old answer given by 711 when it came to franchising: Couple and Wife Store Franchise. Because he owns the store, he will be responsible for the whole thing. You don't need to use a ruler to measure his KPIs. He has an OKR in his mind - "I opened this store just to make money." So every action he makes is automatically aimed at making money.
I am talking about this to make a judgment that is crucial to the implementation of AI:When an SOP cannot even drive carbon-based life, it is futile for you to expect AI to solve it. Don’t think that Agents can solve all problems.
The most typical one is the "self-service inventory" function we designed from the beginning - allowing store clerks to count inventory before closing every day. This function was cut off by me personally. Because I know it all too well: they don’t know how to use it, and you can’t hold their hands to use it. This is a typical head-slapping request. (Later, I used AI to redo it: the clerk took a photo of the refrigerator before leaving get off work and left, leaving the rest to AI to identify - changing "let him do it" into "let him hardly have to do it." We will talk about this later.)
If you don’t go deep into an industry, you won’t even see what the real demand looks like.
The first really useful function: let limited flowers be transformed into 500 styles
After receiving two slaps, I touched the first one that I really needed.
It's hidden in a link that I completely ignored before. As I said before, a store can hang five to six hundred types of flowers. But you can’t really prepare five or six hundred kinds of flowers—it’s impossible to prepare all the flowers in the world. You can only use the limited types of flowers you have at hand to arrange and combine them into countless styles to put on the shelves.
How did you do it before? In the past, it all relied on the ingenuity of real florists, who would arrange, photograph and put each item on the spot. Some stores hire a florist specifically to do just this, and they can make as many designs as you want in a day. But it’s too expensive, and the cost changes every day—tomorrow this variety is no longer available, the price has gone up, or it’s out of stock, and you’ll have to tear down all your designs and start over. Of course, you can also go to Xianyu and buy stock photos to use, but those same pictures will only make you look uncompetitive among a bunch of stores.
Can AI solve this problem? Your first reaction was exactly the same as mine:
Isn’t this just a permutation? The graph-generating model is so good now, is it finished?
That's what we did too. Then, it flipped over again.
Why? Because the graph model trained based on general data does not know so many flowers at all. **
I told it that my store has red roses, white roses, white peony, cappuccino roses, calla lilies, teddy... it was immediately confused. What is "cappuccino with teddy"? It literally puts a cup of coffee on the dog's head for you. It can't tell the difference between calla lilies and calla lilies. What's more, flower shop owners are also full of slang and abbreviations. The name of the flower in their mouth has nothing to do with the Latin scientific name.The final solution is not an easy task: you have to first take an open source, smaller graph model, do a round of pre-training, and tell it step by step - "Cappuccino" is not coffee, but this low-saturation rose; Teddy is not a dog, but a double-petaled sunflower; although this one looks similar, it is called a calla lily, and that one is called a calla lily. After the reference picture is produced, the reference picture is fed to the SOTA model and produced a second time. You can't dismantle the closed-source SOTA model and retrain it. You can only use this stupid method of "small model to base, large model to produce film". But it works.
At the same time, you have to guide the store owner to take pictures of the flowers in his store and put them into the system by hand - because the color, openness, and status of the flowers in each season and each batch are slightly different, and everything must be based on the actual SKU in your store. Fortunately, I have a lot of wholesaler resources at my disposal. They will take pictures of the flowers and send them to us as soon as the goods arrive. This is the cleanest training material: whatever the market is, it will be in the system, and the resulting pictures will be whatever it is.
After this function is completed, it is the first system we have launched and it is also a system that has been widely praised.
But have you noticed that to be able to make this function, the key behind it is: you need to understand the industry, have the data and resources of the industry, and have some model training. This is the threshold. If you really want to beat the question, you really can't do it.
"You can do it even if you are a bean bag" - PTSD common to every AI entrepreneur
When we take the system to sell, too many shop owners will ask the same sentence:
*"Can the bean bag also do this function of yours?"*
I said, OK, you can come on the spot. He took a bean bun and made one, stared at it for a long time, and muttered: "Oh...it doesn't look like it."
I believe that this sentence - "You can do it too" - is the biggest PTSD of many AI entrepreneurs today.
My partner worked on an internal knowledge base system for a government department in the South, and the other party said, "You are not as accurate as Doubao in asking this question." As a result, it turned out that the question they asked about the beanbag was the original text of the report of the National Congress of the Communist Party of China. Of course, anyone can answer it. I asked you to ask for peripheral and thin ones. Do you have bean bags? When asked, no. Only then did the other party react: "Oh, you are better."
If you are also working on AI implementation, you must have hit this wall countless times. The customer's words "Others can do it too" or "You can do it for free" will make you feel discouraged in an instant. But you have to understand - all the cumbersome and invisible work you do is essentially to reply to this sentence. **Reinforcement, pre-training, industry data, and hard work are all designed to create a distance between you and the general large model that customers can vote with their feet at that moment when "bean bags don't seem to work".
If you cannot hold this level, your customers will be driven away by generic (free) models such as bean bags. Only when you hold on to this level can you truly "land".
The same ability has grown a second leg: Flower Changing Customer ServiceWhat’s even more interesting is that after this picture-generating ability became solid, it developed a second function that I didn’t expect – used in pre-sales and in-sales customer service.
The scene is this: the user ordered a bouquet of flowers, which included a tulip, but it was not in stock today, or there was no such flower at all in this season.
The easiest way is to ask for a refund directly. But then won’t my money just go away? Therefore, a responsible merchant will call and communicate: "The flower you ordered is not in season now, and the condition is indeed not good. How about I replace it with calla lilies? The shape will be better, do you think that's okay?"
Usually, at the end of the communication, the user will say something to you:
*"You make it and I'll take a look."*
This sentence has nothing wrong with users. But the florist was dumbfounded on the spot - I slipped all the flowers out and finished them. If you say you don't want them after reading them, who should I sell them to? Reheat? The branches have been pruned and cannot be brought back to the furnace.
After I personally went through it, I found a rough ratio: about one-third of the people refused to exchange it and either gave me the flower or refunded the money; one-third of the people were very happy, "You can do it, you don't need to show me the picture, just send it to your home"; the remaining one-third just said, "I'll see if I make it." In the past, florists could only lay out a rough product for people to see - but don't forget, the delivery platform has an assessment of the picking time. You go back and forth there, and as the time increases, the assessment points will be lost.
But now, in our system: circle the tulip and replace it with a calla lily - the picture will be generated in seconds and sent to the user for a look.
What used to take five or six minutes can now be solved in one minute. Moreover, it not only improves efficiency, but also reduces costs - because you don't touch the real flowers at all, and there is no loss. There is also a silent cost game hidden here: users are often inquiring about several stores at the same time, and which ones are still making orders now. Your picture comes out first - this order will most likely be yours.
You see, really useful functions are often not designed by you, but emerge on their own after you fall into a pit.
A word of advice: don’t build in public
At this point, I would like to intersperse a piece of specific advice I would like to give to my colleagues.
I have sworn to show this set of things only to people in the industry and never to people outside the industry. **
If you are in the flower industry, you come to me and I can see it; if you are not here, I am sorry, but it is useless to see it. Especially programmers - his first reaction is always "I can do this too" because he is curious, but he cannot understand the non-Internet, dirty, and earthy logic behind it.
That once popular word, build in public?
Offline business, this is not necessary. If you come out publicly, laymen will not be able to understand it, and your colleagues will come to harass you. For offline things, you are only responsible for your customers, that’s enough.
True Ceiling: Platform
After talking about people and functions, we must talk about the unavoidable and biggest pain point - the platform.The platform is valuable, and I will not take sides and criticize the platform. For the vast majority of takeaway flower shops, the platform is the most important means of production and the core source of customer acquisition. Without a platform, many stores opened in basements and nooks and crannies would not be discovered at all - you can't open every store in a prosperous shop on the street, right? If calculated carefully, the rent in that area may be higher than the fee you paid to the platform. As for the commission, it’s about 20%, which I think is really not high. After all, the gross profit in this industry is there.
The value of the platform is certain. But the real pain point lies precisely in the platform. **
The first pain point is that the vast majority of bosses do not understand the platform’s investment logic at all. He would only calculate the accounts very crudely: "I invested a thousand yuan today, how many orders were made, and how much money I made." - He regarded investment losses as costs and included them. But anyone who has worked on the Internet knows that traffic should not be a cost, but an expense: you have to rely on it to retain and promote repurchase, rather than investing one dollar and immediately expecting to get two dollars back. Not to mention that the streaming itself is extremely complicated - orientation, ranking, search, different seasons, different festivals, completely different playing methods. You have to calm down and calculate and compare these things one by one.
And this is precisely our pure home ground as Internet people. We tried it on Mother’s Day and May 20, two holidays that are close together. As a result, within one month, our store reached the top 7% in Tianjin. I told my colleagues: This is definitely not something that fell from the sky. This must be something that others have not done well, but we have done a good job.
I sighed in my circle of friends at that time: If Meituan can open this data into an interface for me, I can completely make this investment strategy into an AI product and tell every store owner how to invest today and how to invest tomorrow.
——But this is the second and most fatal pain point: the data on the platform is closed.
I can tell every colleague who is doing AI implementation with absolute certainty: the closed nature of China’s Internet platform is the biggest problem facing all domestic AI implementation manufacturers at the moment. Everything you do usually goes smoothly, but at the last step, "pop", the platform is blocked in front and you can't get in. If you do customer service, you will end up stuck on WeChat; if you do marketing, you will end up stuck on a vertical platform like Meituan.
Take Meituan as an example (other platforms are similar, and this is not just to say that Meituan is bad): its API is not non-existent, but its openness is extremely conservative. There is the concept of "service provider", but the threshold is frighteningly high - if you accept it as a store, you must be a national chain to be eligible; if you accept it as a service provider, the threshold is another hurdle, and you even have to pass the first-class insurance. Even if you go to great lengths to get in, the more detailed marketing data you want is simply not in the document. All you can get is your own order and product data. It is equivalent to saying that the most critical thing is nothing.
As for why the platform is so closed - I will not speculate on the reasons, I will only state the reality.
So, I did a very hacker thing
If the data cannot be received, will the road be cut off? No. I thought of a very "wild" method.
Remember what I said earlier - the florist relies entirely on the A4 paper that is sprayed out to make the order. On that piece of paper, all the order information is included.
So I strung a box between the computer and the printer in the store. To put it bluntly, it is a man-in-the-middle attack: the content to be printed is cut off halfway.
(To tell you more about the fate here: My partner’s previous company started by intercepting the printing data of POS machines and helping merchants with BI and marketing. It was called Hubang International, and was later invested by Alibaba. Therefore, the matter of "threading something in the middle" is in our genes.)
In the past, this was quite difficult - to print it out in PDF or image format, you had to perform OCR yourself. But now with AI, OCR is so easy and can be done in minutes. The data is captured and put into my own system for processing, and the rest is easy to handle.
I connected an Agent at the back so that it can automatically broadcast more detailed information about the order by voice - so that the florist can just listen without occupying his hands; he can also use his mouth to ask: "How far is this order from me?" "When should I send it?" The Agent can answer the questions. In essence, this is reducing the burden on people. Later, when the data has been accumulated for long enough, it will be able to analyze the orders and give some business suggestions to the store owner.
Of course, I still can’t get the marketing data, so I haven’t filled in this gap. But at the order level, we used a box to tear a hole in the closedness of the platform.
So you see - if you can't break through the wall of the platform, then you have to try your best to dig out the data from the cracks in the wall. This is an unavoidable step for the implementation of domestic AI.
It only treats you as a "front warehouse"
In the past six months of working with the platform, my deepest feeling is: In the eyes of the platform, you are not a warm store, you are just a link in its entire link - the "front warehouse" responsible for delivery.
It doesn't really care about your carrying capacity, how good your flowers are, or whether you want to work slowly and carefully. What it wants is a standard product. Meituan once had a label called "Youhua", but that thing was useless. In the end, it still recognized the standard product it defined.
This kind of "front-end warehouse thinking" permeates every assessment detail:
You can't just close the store. We only give you a business suspension quota of more than ten days a year - these ten days also include the Spring Festival. The Spring Festival is gone in seven or eight days. Do you still want to take a break during the peak season? Don't even think about it.
24 hour operation is most encouraged. I foolishly actually tried it for 24 hours at first. As it turned out, someone actually placed an order in the middle of the night (you guessed it, they were all delivered to somewhere - to hotels, and they mostly wanted to add something you know at last), and the phone didn't stop all night long. Unless you really hire people to work three shifts, you can't handle it at all.Inquiries must be responded to within one minute. That’s the voice at the beginning that gave me PTSD. Because flowers are non-standard products, almost every customer has to chat with the customer service and ask for a real picture before placing an order. This "immediate response" assessment is one minute during the day and one minute at night. Points will be deducted even if you doze off in the middle of the night. Our florist has experience and taught me to add a disclaimer and push the orders from 9pm to 10pm the next day to the next day - you can reply to this immediately and the platform will not deduct a second for you.
The picking time and delivery time are all subject to assessment. When these two times are superimposed, the "delivery time" you display on the page is ultimately determined. It's averaged over your history.
And this gave rise to a phenomenon that chills me the most - reverse elimination:
I want to seriously tie a bouquet of flowers by hand. The craftsmanship is good, but it is slow and takes thirty or forty minutes. If your flower shop displays slowly, it will be immediately judged as a loss by the system, sink to the bottom, and no one can see it. So what's the result? Either I'll give you a bunch at random, or I'll just sell the ones I've already prepared in the refrigerator. Those who want to do well will not survive.
By the way, here are some lightning protection tips for you as a consumer (also acquired with "dirty hands"):
Don't buy the kind that has a particularly short delivery time - the shorter it is, the better it is pre-made; don't order it delivered early in the morning, as it must be put in the refrigerator the night before and picked up by the errands the next day, and C-grade flowers will not survive for a few days in the flower mud; if you want to buy good flowers, order them at noon or in the afternoon. Don't place an immediate order during festivals, as no one has time to do it for you. Also, the flowers delivered to you outside are basically C-grade flowers; for the same bouquet, the flowers sold in physical stores are grade A, B or even imported.
The human world in the flower shop
After talking about so many hardcore pitfalls, I want to leave something soft at the end - the world that "Dirty Hands" has shown me in the past six months.
I discovered that flowers are not just flowers at all. It is a magic mirror that reflects people's most secret emotions.
I have observed a very objective rule (how to interpret it is up to you): the bigger the flower, the less surprised the person who receives it will be. **I usually deliver large bouquets in person, so you can observe that when 999 roses are delivered, the recipient often has an expressionless face, or even a little bored.
I also dare to give a reminder to all straight men: Never send red roses with baby’s breath. Although this thing is cheap, in the eyes of knowledgeable people, it looks "stinky" and "too popular" - it looks cheap and cold, especially if it is delivered to someone's office, it will not be the same if they take it out or not take it, which is called an embarrassment.
So how do you buy flowers? My suggestion is very simple: go to the physical store and chat with the florist privately. ** Tell him who you are giving it to, how old he or she is, what the occasion is, and what the purpose is, and let the florist design one for you. Even if you only want a flower, we won't actually give you a bare stem when you go to the store - we will add embellishments, and even place a lily or leaf underneath to make it look plump and decent, rather than dry.The festive nature of this industry is incredibly strong. Once 520 passed, on May 22nd, no one asked about roses anymore. And these days all I can think about is sunflowers - why? College entrance examination. Sending sunflowers is called "winning the first prize in one fell swoop", and pairing them with cheongsam is called "victory starts with a flag". There is also a new trend: the number of girls giving gifts to boys has obviously increased, and the flowers given to boys are often more thoughtful than those given by boys to girls - sometimes a pack of cigarettes is hidden in the bouquet.
Oh, yes, there is a very real compliance detail about hiding cigarettes: because we do not have a tobacco monopoly license, we must clearly inform you that we are "buying on behalf of others" and attach the invoice to the flower - otherwise it is illegal and your peers will report you. In the same way, snack flowers and toy flowers like those on Children’s Day must also have a license to sell pre-packaged food, or at least have a receipt.
You see, none of these things can be "photographed" at your workstation. They will only reveal themselves to you bit by bit when you actually stand behind the counter, being chased by the one-minute reply voice, and packaging the love of strangers with your own hands.
Conclusion: You have to get your hands dirty
After half a year, if I could condense all my words into one sentence and give it to every colleague who is still confused, it would be——
You have to work with dirty hands.
If your hands are not dirty, don’t even think about it. If you sit in front of a computer, you cannot capture the real needs of any industry, not any industry. To paraphrase the saying we often joke about: Give me a hundred years and ask me to take 10,000 data, and I can always get one right, right? ——Isn't that the same as "a monkey can write an operating system by typing on the keyboard"?
When it comes to the implementation of AI, the most taboo thing is to be patted on the head. In the past six months, I have used real money to verify: you think you should reduce losses, but you are wrong; you think it is easy to make a profit, but you are wrong; you think issuing shares can motivate people, but it is useless; you think you can get data by connecting to an API, but there is no way. Only when you really fall into the pit and take a trip, will you know - oh, this is the point.
Among the functions I have made in the past six months, some can really help them, and some, in hindsight, are purely for my own enjoyment. But I recognize both. Because only if you get your hands dirty can you tell which one is a real need and which one is just for fun - and this is exactly what most AI implementation projects never figure out until they die.
So, back to the much-discussed question at the beginning: Can AI be implemented?
My answer is: The answer to this question is not in any debate, it is in the pit.
Stop asking if you can. Go and open your own "flower shop". The day you suffer from PTSD, you will naturally have the answer.