It's difficult for AI to be implemented in enterprises: The collective dilemma of software companies in the AI wave

"Aren't you in the AI business? Why can't you even meet such a basic requirement?"

"Nowadays, AI development is all the rage. Why do you still charge hundreds of thousands?"

"Can you start the work first and we'll pay you if the results are good?"

Do these words sound familiar? Whether you're a veteran in a traditional software company or a newbie in an AI startup, you must have been confronted by clients like this.

"The AI wave is here, and everyone says it's a golden opportunity. But for us software companies truly in the thick of it, we know the reality better than anyone," the founder of a software company complained to Lao Yang. "It's not that there are no opportunities, but there are pitfalls behind every opportunity. It's so difficult to implement AI in enterprises!" Lao Yang, from the perspective of a client company, really empathizes. Today, Lao Yang will talk to you about the dilemmas of software companies in the current AI wave.

In Lao Yang's view, there are two types of software companies surviving in the current AI wave: traditional software giants and emerging AI startups. The former rely on the existing market, while the latter rely on AI narratives to secure enterprise orders.

Traditional Software Companies: Hard to Turn the Big Ship

Let's first talk about those established software companies that have been in the industry for more than a decade. They have a market, customers, and a technical team. On the surface, they have a solid foundation, but in fact, they face the most painful transformation. Not to mention the strategic investment risks of betting on AI, just these common problems are difficult to solve:

1. The old architecture is filled with a mess of code, and there's no place to put AI functions

As a client, Lao Yang often pressures the cooperating big companies: "Hurry up and add AI functions. AI tools are everywhere in the market!" But all he gets in return is a sigh from the other side: "Alas! Everyone understands that the product architecture of some big companies has been running for more than a decade. The people who wrote the code back then have long left. After numerous patches and bug fixes over the years, it has become a mess of code. It's simply impossible to add AI on top of it! Talking about refactoring with the enterprise? The cost alone is enough to scare away 98% of the customers. But the problem is that enterprise leaders don't care about these technical debts of the software company. They only see that others have it and you don't, which means your technology is backward."

2. Old customers have become a "sweet burden"

For traditional software companies, having a large number of old customers used to be an advantage, but now it's a double - edged sword. Because old customers are using old versions and old architectures, if you want to develop AI functions for them, you have to be compatible with more than a dozen versions at the same time, and the maintenance cost increases exponentially. New customers without historical burdens, on the other hand, can start with a clean slate. In the current market environment, most old customers tend to be conservative and have no energy for innovation. Even if they want to innovate, they are constrained by costs. At this time, a contradiction arises. If you can't serve old customers well, there's a risk of cash - flow interruption. If you want to co - create AI with enterprises, but enterprises don't have enough funds. Looking for new customers means a large amount of market investment.

3. Difficulty in retaining talent

People who understand the old technology stack don't understand AI, and those who understand AI don't want to touch the old code. This is the organizational dilemma of current software companies. For example, if you recruit an AI engineer, as soon as they see that they have to maintain a codebase from more than a decade ago, they'll leave immediately. Training old employees internally to learn AI is slow, and as soon as they learn it, they may be poached by Internet companies with high salaries because there is a great shortage of people who understand both business and AI in the industry.

4. The pricing logic has collapsed

Previously, software companies quoted prices based on function points and man - days, and enterprise customers would accept it. But now, customers will ask: "Your AI function is just calling the interface of a public large - scale model. Why do you charge me an extra 200,000?" At this time, if you explain that data needs to be managed, the model needs to be optimized, it needs to be integrated with the old system, and user training needs to be provided... enterprise customers may not even listen. They'll say: "Isn't there AI? It's very simple to do these things. Can you do it for 20,000?"

The result is that if you don't lower the price, you'll lose the order; if you lower the price, you'll make a loss! No matter what you do, it's wrong!

AI Startups: Small Boats Fear the Waves

Now let's talk about those emerging startups born in the AI wave. They have no historical burdens and use the latest technology stack, and they are very flexible. But after cooperating with them, Lao Yang really understands their hardships, which are of a different kind.

1. Can only do "tools", can't enter the "core"

What are startups good at? Telling stories with new AI technologies to impress enterprise bosses. But after cooperation, it's found that the AI functions they can provide are limited to knowledge bases, intelligent Q&A, work reports, and report generation. These tool - type applications seem fresh to enterprise leaders at first, but after using them for two months, they'll think "it's just so - so". If you want to go deeper, such as helping customers with sales forecasting and supply - chain optimization, sorry, these startups don't understand the industry at all. They can't figure out the enterprise's core business processes, historical data, and unwritten rules.

So when AI startups cooperate with enterprises, their product capabilities can only stay at the surface level of simple tools, making enterprises feel that "it's not enough", and they themselves feel "frustrated" due to the lack of in - depth understanding of the business.

2. POC (Proof of Concept) is a bottomless pit

To solve the survival problem, most startups will do POC for enterprises for free to win orders.

The enterprise leader says "try this scenario", so the software company immediately develops it. Before it's even really tested, the enterprise leader will say: "Try that scenario!" As a result, after trying three or five scenarios, either the business data is too dirty and messy to handle, or the enterprise leader says: "The budget is only 20,000. Can you do it?"

The dilemma of AI startups at this time is that if you don't do POC, you don't even have the entry qualification. If you do POC, you may be exploited for free. Because if you don't do it, the startup next door will.

3. The project cycle is uncertain, and the cash - flow may break at any time

AI projects require customers to cooperate in providing data, sorting out rules, and participating in testing. But in reality, enterprises say they are in a hurry, but once the contract is signed, they are not in a hurry during the project construction period. The data is delayed for two months, and the rules are delayed for three months. The startup's team is burning money every day, and the contract amount is fixed. The project cycle is difficult to control. As we all know, the longer the project cycle is delayed, the thinner the profit margin will be, and finally, it will result in a loss. If it's a traditional software company with a solid foundation, it can withstand it, but startups have a fragile capital chain and can't afford such consumption. If a large project is delayed for half a year, the company may go bankrupt.

4. The technology base changes every day, and the product becomes obsolete before it's even finished

You develop an application with Model A today, and tomorrow, Model B releases a new version with better performance and a lower price. At this time, enterprise customers will ask: "Can you switch to the new model?" Yes! But the problem is that it needs to be re - adapted. Who will bear the cost? Most enterprises won't. If the software company says no, the enterprise leader will probably say: "Change to another company!"

If you chase the new model, you'll always be on the go; if you don't, you'll lose the order. This kind of anxiety can only be understood by AI startups.

Common Dilemmas

First, enterprise leaders' expectations of AI are always in the clouds

Enterprise leaders have read too many self - media articles and think that AI can do everything. They'll ask software companies to predict the stock market, automatically handle all customer complaints, and replace humans in making decisions. If the software company explains that "this can't be done", the leader will think that the software company is incompetent. In most cases, some software companies will bite the bullet and do it, but if the result is not good, the leader will think that the software company's technology is even worse!

Second: The success or failure of a project doesn't depend on the software company, but on the customer's data and organization

No matter how well the software company writes the code and tunes the model, if the enterprise's data is dirty, messy, and incomplete, the final result will be terrible. But in most cases, enterprises won't blame their own data, but only blame the product. So some software companies will spend 80% of their energy helping customers clean data, promote processes, and provide training. These costs are not written in the contract, so they can't get paid. But if they don't do it, the project simply can't move forward.

Third: The price war is even more fierce at this time.

In the traditional software era, there were still functional differences, but in the AI era, everyone is just calling interfaces, and the products are highly homogeneous. Enterprise procurement departments will also use AI to compare prices based on the quotation schemes. As a result, the price is driven all the way below the cost line. If you still want to use "industry experience" as a premium at this time, the enterprise leader will say "make a more detailed plan". If you do, the enterprise leader will use it to pressure other companies for a lower price.

Fourth: Can't find a way out for the business model.

If you charge by project, enterprises think it's too expensive. If you charge by SaaS subscription, enterprises will say "our data is sensitive and can't be put on the cloud"; if you charge by results, enterprises will say "how to define the results? You define it and I won't agree, I define it and you won't accept it". If you charge by man - days of service, it's back to the old way of traditional outsourcing. So after trying all the charging models, it's found that none of them work.

From the above, it's not difficult to see that although there seem to be dividends in the AI wave, there are actually "pitfalls" everywhere after really entering the game. The summary is as follows:

Unlimited enterprise expectations vs limited technical capabilities - Customers think AI is magic, but it's a systematic project.

Old customers are assets vs old customers are liabilities - Old customers bring in money, but they may also kill innovation.

Projects require in - depth customer participation vs customers are reluctant to invest - Success depends on customers, but customers are irresponsible.

Fast - paced technology iteration vs long - cycle project delivery - The model changes every three months, but the project takes half a year to go live.

Value lies in industry knowledge vs customers are only willing to pay for functions - You're selling "capabilities", but customers only recognize "calling APIs".

Requires long - term investment vs capital demands short - term returns - It takes three years to develop AI capabilities, but investors want growth next quarter.

The above six problems make software companies feel like they're running in a swamp - the harder they try, the deeper they sink.

So don't be deceived by the "AI will disrupt everything" chicken - soup stories. No software company has an easy time in the AI wave. Traditional companies have their own hardships, and startups have their own difficulties. But Lao Yang thinks this is not a bad thing because only the best will survive after the waves of competition. So where is the way out for software companies? Lao Yang has talked a lot about this in previous articles. Welcome to leave a comment!

This article is from the WeChat official account "Xiangjiang Digital Review" (ID: benpaoshuzi), author: Lao Yang, published by 36Kr with authorization.