Introduction
It’s been a long time since I last wrote an article. Sure enough, the older I get, the less I like to share. In just two years, I feel like a wonderful adventurer. I have experienced plug-in development in the startup community of ‘midjourney’, app development in the financing stage of ‘MyShell’, Prompt editor development by the Pure Technology Core Team of Web3 Company, and development of ‘Li Yizhou Technology Service Provider’.
Of course, except for the last one, everyone else is living well and making good profits. But I am fascinated. It seems that when it comes to these products, I am just a participant. I can't tell the story. I can only say that I was the first to write the community plug-in for midjourney DC. I was the first to write the MyShell App but I was too bad at it, hahaha and so on.
I feel that even if I am in a particularly strong team, I am just a means of production. In a technical and core team, I cannot be the core (such as mj). In a strong KOL team, I can only develop (such as Li Yizhou) and so on.
So I have to become stronger. As a means of production, my growth is already very limited. My background prevents me from having continuous entrepreneurial experience like some geniuses (such as David and Ethan) starting from the world's top schools and universities.
So I started chasing my dream. I ran to Beijing with my luggage 2 months ago, got some meager living expenses (salary), and started a startup. If I succeed, I can show off and tell stories again. If I fail, I will definitely grow tremendously (because this product is your child, and its revenue is your income).
The core of the product
Before all the reading sessions begin, the general direction of the product must be determined. The products I make are AI products, so I will first ask a few core questions. You can try to think about them first and then see my answers (the answers must be different from person to person because of different experiences).
1. When the team resources are limited, should we do it domestically or abroad?
2. Should we strengthen cash flow projects or raise PV through financing?
3. What is the core competitiveness of the product?
4. Is the current team capable of holding such a product?
5. Have competing products on the market completely wiped out the market, causing your customer acquisition costs to be extremely high?
Of course, we encounter many, many problems. These are just a few that I think are more important.
1. When the team resources are limited, should we do it domestically or abroad?
Don't do domestic work when the team resources are limited, or even when there are more.
Because at present, domestic manufacturers are very high-volume. You said you charge money for models, but the big manufacturers are all free.
You said you want to advertise. All major platforms, including Douyin, are very strict about advertising review in the AI category (except for your own).
If you want to join the stream, the competition for big words is too fierce.
Whatever you want to achieve, friends in China can quickly come up with a set for you.
Then there are the audit and qualification requirements that are particularly troublesome.
In the end, domestic players are not willing to pay, and the only ways to do this are through "paying for knowledge and selling courses" (but that doesn't work anymore).
Even if you overcome all the "unfavorable factors" and finally do it, you will still be eaten.
At present, few domestic AI companies are making money, and that’s not the case all over the world.
2. Strong cash flow or financing
In my opinion, there are only 3 types of players that can use financing
1. The background is particularly good.
2. Strong technology that is almost irreplaceable at the current stage.
3. I have very good relationships with various organizations.
The best situation is that everyone is rushing to invest for you and a large institution leads the investment. The worse situation is that you have to work hard to find people to write financing proposals, etc. The worst situation is that you have to keep going and still only get small amounts of financing.
Therefore, we can only do projects with strong cash flow.
What I understand as "strong cash flow" is cash cows, that is, projects with particularly strong short-term "ROI" (return on investment) that can directly cover investment and inflow costs and generate a large amount of wealth every month.
Under this premise, it is very important to look at the numbers. Every number must be considered and considered, such as bounce rate, registration rate, registration payment rate, retention payment rate, prepayment rate, etc. When we look at the numbers, we must be much more detailed than this (including various channels, average stay, intention conversion, etc., etc.), but this article will only talk about the last four.
3. What is the core competitiveness of the product?
The core of selling products is to solve people's needs, just like some AI are specialized for law, some AI are specialized for Web3, some AI are specialized for images, and some AI are specialized for companionship (pseudo needs).
So what is a strong need for people?
4. Can the current team and resources hold such a product?
This is the conflict between dream vision and reality. You say let’s make a chip. Is it possible? But the reality is actually like this. A month ago, a hospital here asked me to do an AI full-process hemodialysis (that is, blood test)... Obviously I don't have this ability, even the ability to make connections. His budget is 300W... If you have to eat, you should eat it. If you can't eat, don't eat it. Evaluating the strength of yourself and your team can sometimes reduce a lot of wasted time.
5. Have competing products on the market completely wiped out the market, causing your customer acquisition costs to be extremely high?
You have to know this when you really start streaming and looking for KOL/KOC. But I believe that most of the time I don’t. The cognitive gap between people is huge. In fact, it is not available in current products.
Data
Most things must be supported by data, and decisions should not be made based on subjective judgment.
This is something I learned recently, because I found that the performance of data is often very different from my understanding.
1. Start with data
Everything starts with data collection. If you cannot collect data accurately, it basically means you are blind.
Google Analytics
Generally speaking, GA is the one that is most seen in terms of growth. Its focus is also very simple. Just throw Gtag Script into it and then report it automatically. Usually used to look at some retention channels to retain my words, it is free.
Matomo / Mixpanel / Thingslog
They are both friendlier and better configured platforms. Among the three, Matomo and Mixpanel are more oriented towards viewing and configuring products and growth, ThingsLog is more oriented towards technology. I currently use Mixpanel because it looks better~, it is basically free for about 100,000 users (estimate), and it charges based on the number of events.
User peek screen and function heat map
Hotjar and Clarity, currently using Clarity, free.
2. Registration rate
There are generally several factors that affect registration rates:
1. Channels
2. Website performance and speed
3. Website content
4. How to register
What we can control are the latter three.
1.Channel
This part is completely controlled by growth. If his delivery channel is poor, for example: pop-up pages in novels. Then you will find that user retention and registration rates are particularly low.
2. Performance and speed
This is what development should do. What users can generally feel is FCP and LCP. You can go to pagespeed.web.dev/ to test the speed. This has something to do with SEO.
There are many solutions, including SSR, caching, js loading after hydration, large script delay, and image CDN. In short, I have tried a lot myself, and it depends on my specific technical solution.
The other part is the optimization of the login process and the speed of registration. Login means registration. For the process on the UI, refer to some better ones. character.ai/
On the other hand, all registration solutions are directly integrated through native integration. Even a large third-party registration integrator like firebase will suffer some speed loss.
3. Website content
Either find artificial ones (big IPs and creators), or copy them.
But many times it is very difficult to communicate with people. At the beginning, the way our boss and I communicated with people was to reward the creators with 20$ before communicating, and then listen to them.
Then I also found a lot of Southeast Asian writers (also expensive).
Later I said that I would go directly to the competing products, because the competing products must have been tested by the market. This must be the fastest way in the early stage of the product. Yes, it is not fast to produce more than 1,000 works in 20 minutes.
4. How to register
It is enough to implement as many mainstream and as many registration methods as possible.
####Finally
If the above content is good enough, you also need to review the registration rate. The average registration rate of similar competing AI character websites is above 20% (an underestimated value), and the registration rate of good peers can even be 50% or higher (I am not responsible for the accuracy of the data).
Looking at the process of registration rate + stay time every day, if it is stable enough at 2 and 4 points, it is actually analyzing the quality of the channel to decide whether to continue to use this channel, as well as the current user satisfaction with the content and whether content needs to be released.
2. Prepaid rate/paid rate
These two indicators are usually viewed together.
Prepaid rate
This indicator is actually the indicator before payment rate. My understanding is the rate of payment intention. It can be characterized in many ways. For example, if the user first performs operation A and then performs operation B, he is deemed to be a user with the intention of paying.
For example, if a user directly enters the webpage and jumps to the Pricing product display payment page, does this mean that he has the intention to pay? Obviously no, it may be because your Pricing page Seo is well done, causing users to go directly to the Pricing page as soon as they enter. Maybe he just wants to take a look, you can't be sure.
What is a better process? The user enters the Feed stream content distribution page on the Home page, then clicks to enter a certain role, chats about N (an average indicator line), and then his free quota is used up, and he enters the Pricing page. At this time, it is regarded as a prepaid user.
Of course, the specific actual situation and the determined branch flow are much more complicated than this.
Payment rate
Then in the next step, through the ratio of prepaid users and the ratio of final paying users, we can conclude that he intends to pay, but he may not pay for some reasons when he gets to the product page, so we have to think about these reasons.
For example, the products are too expensive, there are no products that he is satisfied with, the payment channels are too cumbersome, the payment process is difficult, he has to run away from the platform, etc.
We need to find the reasons and formulate appropriate strategies.
Of course, this is only part of it. We can also have many ways to improve users' payment rate, "better content", "more unique functions", "more consumption touch points", etc.
In a cash flow project, the pay rate is almost everything.
Retention rate
Needless to say, this is related to the question of whether he will pay for the second month. This requires constant content supply and the implementation of some retention functions.
The main purpose is to analyze the retention differences between different user groups (for example, paid users and free users). Draw the key points from it.
And identify the time points when the retention rate drops to locate the problem of user churn.
About the rule of large numbers and the amount of investment
The law of large numbers is an important theorem in probability theory, which states that in a large number of repeated experiments, the average outcome of an event will approach the theoretical expected value. It states that the frequency of a random event will approach its probability when the number of trials is large enough.
To put it simply, we need a certain amount of money to observe whether your ORI covers the cost (investment does not mean investing millions at once for a small team), and what adjustments you need to make currently. Then about 30 to 50 times is considered to be a "large enough" number under the rule of large numbers.
In other words, you need 30 paying users every day. Assuming the registration payment rate of the website is 1% (this number is very high in this category), you need 3000 registrations.
And when our ORI covers all kinds of costs, we will start to make a fortune! Infinite upward streaming (but this is very unrealistic, because the data is dynamic).
About Seo
Seo takes effect very slowly (6 months). Not only do you need to create various content, but you also need to constantly optimize external links and various details of the website. For example, Alt, TDK, Internal link, SITEMAP and so on.
The best way is to invest in streams, buy navigation sites, look for KOL/KOC, and buy POST at the same time to accumulate enough weight to reach a certain breaking point.
Ideal
The most ideal situation is definitely a small amount of traffic and a large amount of natural traffic, just like this. But this is impossible in the early stage. You need to keep settling and accumulating before you can harvest the vegetables on the last day.
Technology
In fact, I feel like there’s nothing technical to talk about, so I’ll give you some experience packs.
Front-end
Framework: I used to be a loyal Next and Frameless developer, and now I am forced to take over Nuxt for this project (the overseas business ecosystem of the Vue series is almost incomparable to React).
css: I am using unocss. I sincerely want to ask, when I use uno in nuxt3, how can I bring the css style embed style in the document of ssr instead of loading it through link. like this
instead of
.webp?updatedAt=1739894462774)
Image CDN service: imagekit, all images are webp, quality 0.8 (the monthly service bandwidth is so high that I don’t know how he calculates this broken CDN).
Login integration service: Supabase Auth (not as good as firebase, I don’t know how many times it has been transferred to the integration service).
Deployment: cloudflare is probably the only good news.
Front-end service: Nitro feels very easy to use. You can write the cloudflare adapter and do things like controlling cache (control query and then cache, etc.)
Crawler: express hand rub.
Image service: python hand rub.
Future app: Because Expo is used to embed H5, because Rn is a native webview, and flutter has pixel rendering performance which will be a little worse. Another way is that I don’t want to install xcode and android studio again on my computer. It’s too big.
Backend
The backend is Go's microservices + a bunch of integrated operation and maintenance (railway, supabase, etc., I can't write a little, and I hate this kind of integration very much, so I write work orders and complain every day).
##Finally
It’s already 6 o’clock in the morning when I finished writing this article (Wuhu is exactly 3 o’clock in the afternoon in California). After thinking about it, these are some “experiences” of a broken front-end developer who is not very proficient in products, not very proficient in growth, or even not very knowledgeable about technology. It is enough to make me laugh. It would be better to say it is a growth milestone of my own. Looking at my articles in each time period in the past, I really miss my old self.