Author: Deep Thinking Circle

Have you noticed a paradoxical phenomenon? On one hand, there's an overwhelming flood of AI success stories, funding announcements everywhere, and AI products being released daily. On the other hand, there's the real predicament faced by businesses: IBM research shows that 75% of AI solutions fail to deliver the expected ROI (Return on Investment), and an MIT report is even worse, stating that 95% of AI projects have no measurable returns. What exactly is causing this huge gap? Why is it that AI technology, which seems so glamorous, is so difficult to implement?

A few days ago, I watched a video where Ben shared an in-depth analysis of AI business models in 2026. Ben has been running an AI agency and AI software business for over two years, and his observations resonated with me. He pointed out a reality that many people overlook: the AI ​​businesses that truly make money and bring real value to customers are often not the coolest pure product companies, but rather those that seem "boring" yet offer a combination of services and products. This viewpoint completely overturned my understanding of AI entrepreneurship.

Why do most AI solutions fail?

In the video, Ben cited a shocking data comparison. While ChatGPT usage is growing rapidly and enterprises are frantically trying various AI solutions, very few are actually seeing commercial value. According to MIT research, only 5% of pilot projects of AI solutions sold by vendors eventually reach production environments. Deloitte found that only 15% of organizations reported a significant, measurable ROI from AI. A PwC report showed that 76% of enterprises have not yet seen any impact on profits. These numbers are truly appalling.

But at the same time, we also see completely opposite examples. Clara reports that their AI assistant helped cut customer service costs by 40% without reducing customer satisfaction. Intercom resolves over one million customer support conversations weekly. Freshwork used AI to reduce IT help desk ticket resolution time by 76%. Why do some companies reap such amazing returns from AI while others reap nothing?

Ben summarized three key factors, which I think are very insightful. The first factor is customization and process reengineering. AI can automate the workforce, but it only creates ROI when it's truly embedded in actual workflows, not simply adding another tool to the tool stack. This means that a certain degree of customization, integration, or even redesign of existing processes is often required. The workforce is closely tied to a company's unique data, edge cases, tools, and definition of "what's good." A McKinsey study confirms this, finding that among 25 attributes tested, redesigning and customizing workflows for AI had the greatest impact on whether organizations could see a real EBIT (earnings before interest and taxes) impact from generative AI solutions.

I understand this all too well. Many companies believe that buying an AI tool will yield immediate results, just like buying an Excel program. But AI doesn't work that way. Every company has a different data structure, business processes, and definition of quality. Without deep customization, AI is like a new employee who doesn't understand the company's situation—it simply can't perform its job. This is why out-of-the-box AI products often fall short, while deeply customized solutions are what truly create value.

The second factor is team training and a shift in mindset. Ben emphasizes that AI is a new technology; traditional software is deterministic, while AI software is probabilistic. People need to relearn how to critically evaluate the output of AI software, rather than blindly trusting it. Many people see a single erroneous output and conclude that the AI ​​solution is immature, rather than understanding the nature of this new technology. If teams don't learn how to use AI software, how to evaluate its output, when to validate it, and what constitutes a good result, adoption rates within the company often plummet.

Ben gave a good example: their AI SEO software was a productized solution, but it couldn't be properly adopted without training the team on how to use the system itself and how to collaborate with AI. I think this is particularly important because it reveals a truth that many people overlook: AI isn't magic; it requires humans to learn how to collaborate with it. Just like when users transitioned from command-line interfaces to graphical interfaces, they needed to learn new ways of interacting. The same learning curve is needed now when transitioning from traditional software to AI software.

The third factor is ongoing operation and human oversight. Because AI solutions typically promise results rather than just productivity tools, this means there's usually a need for someone to be responsible for and own the system's operation. Jobs change, businesses change, and AI evolves very rapidly. All of these factors mean there's often a need for someone to monitor quality, to be part of the human-in-the-loop process, handle edge cases, tighten guardrails, update cue words and logic, and overall maintain its alignment with the business.

Ben likens AI to a smart intern who still needs hands-on guidance and mentoring to produce results, rather than software that can be set up and forgotten. I completely agree with this analogy. Many companies expect AI to run automatically after deployment, like traditional SaaS (Software as a Service). But AI is more like hiring an employee, requiring continuous management, feedback, and adjustments. A Gardner study also supports this, showing that regularly evaluating and optimizing AI systems can increase the likelihood of obtaining high value threefold.

What do successful AI businesses do?

So how do successful AI businesses ensure these factors are met? Ben's answer is simple yet crucial: typically by adding a service layer on top of the AI ​​solution or software. This is the core of that seemingly "boring" but extremely effective business model. We see all types of successful AI businesses, companies using AI-native software, increasingly offering a combination of consulting, education, and customized implementation.

Ben provides a detailed analysis of several key business models. The first is the increasing presence of consulting departments in AI startups and AI software businesses. Forward deployed engineers, or solution engineers, are now among the most sought-after and highest-valued positions in AI startups. Dozens of Y Combinator startups are offering these services through these forward deployed engineers to ensure deployments actually happen. Depending on the solution, these engineers help continuously optimize and integrate the product into each specific business. They sometimes provide consulting, helping businesses prioritize and restructure processes, and sometimes educate and train teams on how to collaborate with AI and use these tools effectively.

I looked at the Y Combinator companies Ben mentioned—Harvey AI, Strata AI, Sakana, Collectwise, Furai, etc.—and they're all hiring heavily for these kinds of roles. Even large companies like n8n, Relevance AI, or Make.com typically have service departments for large clients and a network of partners that can provide these services to smaller clients. Consider n8n's success; it's largely due to YouTube bloggers educating many business owners on how to actually use these tools. What does this tell us? It shows that even the best products need an education and service layer to truly deliver their value.

Depending on the specific software, some lean towards customized services, some towards training and empowerment, and some towards consulting. Sometimes it's a mix of all three, but for almost all of these AI-native software businesses, this service layer remains essential for delivering real ROI to enterprises. This completely shattered my previous understanding of the software business. In the traditional SaaS era, the most successful business model was a fully self-service, scalable product. But in the AI ​​era, even the best products require service layer support.

The second business model is AI-first service agencies, such as marketing or lead generation agencies, which heavily utilize AI in their internal processes to automate services delivered to clients. Ben mentioned Called IQ, an AI-first lead generation agency that uses AI to automate content creation, email, and LinkedIn outreach processes, delivering these services through account managers or GTM engineers (entry-to-market engineers). These agencies have an advantage because they are the AI ​​operators themselves. Therefore, they typically don't need to train their client teams to use AI software. However, this is inherently a service business, providing consulting and customized strategies, usually done by these account managers, who are increasingly needing to become more technically proficient. This is why AI GTM engineers have become a new, high-demand position.

I think this model is particularly clever. Instead of trying to persuade clients to change the way they work, it directly delivers results. Clients don't need to learn how to use AI; they just need to see better marketing results or more potential customers. This model completely hides the complexity of AI behind the service; clients are buying results, not tools. This also explains why many traditional service organizations have significantly improved their profit margins by introducing AI—because their delivery costs have decreased, but the price clients pay hasn't decreased accordingly.

The third type is AI automation agencies, which offer the highest ROI and the greatest traction for businesses. These agencies don't just focus on building; they become AI partners, providing a complete service layer, including consulting through AI audits, customized implementation, and team training to teach businesses how to collaborate effectively with these systems. A highly valuable role within these agencies is delivery managers, who possess a combination of business understanding, AI expertise, and communication skills. They can provide ongoing consultation, refactor processes, identify inefficiencies, train teams, and deploy AI operators.

Ben shared a very insightful experience. When he first started his agency, they focused more on the implementation, which often resulted in AI solutions not being used or adopted by companies. Later, they adopted an approach that combined strategy, education, and implementation, and brought in delivery managers. Since then, they've generated significantly higher adoption rates and ROI for businesses. This shift is crucial, illustrating that technology implementation is only a small part of success; the real value lies in ensuring the solution is used correctly and produces results.

The fourth high-value role is that of AI officers or fractional AI officers. They possess a comprehensive skill set combining business acumen and AI technology understanding, enabling them to offer the same service package to help businesses transform in the AI ​​era. Ben mentioned that this role has many names, such as fractional AI officer or AI transformation officer, but ultimately it refers to those with a particularly high-value skill set who can deliver real ROI for businesses from AI solutions.

The boundaries between products and services are blurring.

One of Ben's key points particularly struck me: even though we can now build excellent software in hours using Claude Code, if you want to build an AI product business, most of the time (not all the time, but most of the time) you need to invest heavily in providing services. Many people see products and services as black and white, but Ben believes that in the AI ​​field it's more like a spectrum. There may be some completely self-service AI SaaS, and there are also some completely customized ones like AI transformation agencies.

I completely agree with his point. Ben's argument is that most AI businesses attempting to launch in 2026, regardless of their business model, will need to add some part of a service layer. Even if you have a fully self-service product, you'll likely need to invest heavily in education and onboarding training. With tools like Claude Code, product building is becoming increasingly democratized. While building products was difficult in the SaaS era, launching a successful AI SaaS now is less about code and more about AI deployment capabilities.

This insight is profound. It means that the technical barrier to entry is decreasing, but the service barrier is increasing. In the past, being able to write code was enough to create significant value because writing code was difficult. Now, simply being able to write code is no longer sufficient, because AI can write code for you. What truly brings value is understanding customer needs, designing the right solutions, and ensuring they are correctly deployed and used. These all require deep service capabilities, not just technical skills.

Ben says many people dream of building an AI product, keeping it streamlined, and selling it to thousands of people. But for most people without years of entrepreneurial experience, VC connections, or Silicon Valley networks, the reality is that even if providing services isn't the ultimate goal, it's the fastest path to delivering real ROI today. And services are also the best vehicle for productization. When the same patterns repeat across different clients—similar workflow tweaks, similar integrations, similar training issues, and recurring ROIs—these become signals that they should be repeated and productized.

I think this is the most important point. Good products usually come from evidence, not assumptions. A16Z (Andreessen Horowitz, a top VC firm) also published an article about product-led growth and service-led growth in the AI ​​era. They see the same trend: the companies that deliver the highest ROI and generate the most long-term revenue are those adopting service-oriented AI businesses. While this may mean lower profit margins and more work at the beginning, it usually means you can find product-market fit much faster.

Ben uses his own AI SEO software as an example. They've built customized SEO systems for multiple clients to understand what the product needs, what truly produces results, what integrations are required, and how to get people to use it efficiently. Through this process, they've been able to increasingly productize their solutions. But even after working with over 100 companies on the software, they still need to invest heavily in education and training for each client to see real results from these solutions.

My In-Depth Thoughts on AI Business Models

After listening to Ben's presentation, I have a deeper understanding of the essence of AI business. I believe we are experiencing not just technological change, but a fundamental shift in business models. In the era of traditional software, scalability was king. The most successful software companies were those that could serve the most customers at the lowest marginal cost. This is why the SaaS model is so popular, because once the software is developed, the cost of serving one customer is almost the same as serving ten thousand customers.

But AI has changed the game. The value of AI lies not in the software itself, but in how it is applied to specific business scenarios. Each company has different data, processes, and goals, meaning that the deployment and effectiveness of the same AI tool can vary drastically across different companies. This is why customization and the service layer have become so crucial. We can no longer think about AI business using the traditional software mindset.

I think the AI ​​business is more like a hybrid of consulting and software. It requires the client insights and customization capabilities of consulting, as well as the technical capabilities and scalability potential of software. AI companies that try to take a pure product route often encounter adoption problems because, although their products are technically advanced, customers don't know how to use them or can't integrate them into their existing processes. On the other hand, companies that take a pure service route can meet customer needs, but they lack scalability, and their profit margins are limited.

The most successful AI business models strike a balance between these two aspects. The success stories Ben mentioned—whether Y Combinator's AI startups, AI-first service providers, or AI automation agencies—all share a common thread: they offer a combination of products and services. They use products to provide core functionality and scalability, and services to ensure the products are correctly deployed and used. This hybrid model may have lower profit margins in the short term, but it is more sustainable in the long run because it truly creates value for customers.

I've also noticed an interesting trend: high-value positions in the AI ​​era are multi-skilled. They're no longer purely engineers or purely business personnel, but rather those who understand both technology and business. Forward-deployed engineers need to understand customer business processes, AI GTM engineers need both technical implementation and market strategy expertise, delivery managers need business acumen, technical skills, and communication abilities, and AI officers require well-rounded capabilities. This reflects the essential characteristic of AI business: the deep integration of technology and business.

From an entrepreneurial perspective, I think Ben's advice is very practical. For those wanting to enter the AI ​​field, don't start by thinking about creating a scalable product. Begin by providing services, using those services to deeply understand customer needs, accumulate experience, and discover repeatable patterns. Once you've solved the same problem for 10 customers, you'll know what's worth productizing. This service-to-product path may seem "boring," but it's the most stable and likely to lead to success.

I've also pondered why traditional product thinking has failed in the AI ​​era. I believe the fundamental reason lies in the probabilistic nature of AI. Traditional software is deterministic; given the same input, it always produces the same output. But AI is probabilistic; the same input may produce different outputs, and the quality of the output depends on many factors, including training data, prompts, and context. This uncertainty means that AI cannot "set it up and forget it" like traditional software; it requires continuous supervision, adjustment, and optimization.

This explains why the service layer is so important. The service layer provides not only technical support, but also a continuous process of optimization and refinement. As Ben said, AI is more like a smart intern than an automated tool. You need to give it feedback, adjust its behavior, handle its errors, and teach it new skills. This process is not a one-time event, but continuous. This is why AI companies that only sell products and don't provide services rarely succeed, because they push the responsibility for continuous optimization onto the customer, and most customers lack the ability or willingness to take on this responsibility.

Outlook and suggestions for the future

Based on Ben's analysis and my own reflections, I have several predictions for the future of AI businesses. In the short term (the next 2-3 years), service-oriented AI business models will continue to dominate. Technology is still evolving rapidly, and each company's needs are different; standardization has not yet been established. During this phase, companies that can provide deep customization and continuous support will reap the greatest value.

In the medium term (3-5 years), we will see some successful models begin to be productized. Companies that discover repeatable patterns during the service process will begin to solidify these patterns into product features. However, even at this stage, fully self-service AI products will still be in the minority, and most successful AI companies will still retain some service components. Just like the current enterprise software market, although there are highly productized companies like Salesforce, they still have a large network of implementation partners and professional service teams.

In the long term (5 years or more), AI technology will become more mature and reliable, and users will become more familiar with how to collaborate with AI. At that point, we may see more fully productized AI solutions emerge. However, I believe the service layer will never completely disappear because the complexity and diversity of business are eternal. Even as AI becomes more intelligent, businesses will still need help to integrate AI into their unique business processes.

For professionals looking to enter the AI ​​field, my advice is to cultivate a multifaceted skill set. Don't just learn technical skills or only business acumen; strive for both. Learn some AI automation tools, such as n8n and Make.com, and also learn some coding, such as Claude Code. Simultaneously, develop business acumen, learn to identify company pain points, design solutions, and communicate effectively with clients. This multifaceted skill set will be extremely valuable in the future.

Ben suggests that professionals should gradually position themselves as AI operators or AI officials within their companies. Start by automating some of your own processes, then expand to other processes in the business, demonstrating and training others on how to use AI. This will not only make you indispensable in your current company but also build a valuable skillset for the AI ​​era. I strongly agree with this advice because AI won't replace people, but those who know how to use AI will replace those who don't.

For aspiring entrepreneurs, Ben suggests starting with an AI agency or fractional AI officer. This naturally builds three key skills: consulting, implementation, and training. Tools can be learned in a few weeks, but these skills require practical experience. Furthermore, most businesses lag behind AI trends, and if you're a few weeks ahead, they'll be willing to work with you long-term. Typically, just two to four clients are enough to generate a recurring monthly revenue of $10,000 to $20,000.

If you're already running an AI agency, Ben suggests truly investing in your service portfolio: consulting, training, and implementation. Many agencies focus purely on implementation, but adding a consulting and training layer through AI audits, workshops, and training sessions is key to driving the ROI of your solutions. And you can get these clients to sign long-term agreements, which is crucial for recurring revenue in this business model.

If you're an existing service agency, such as a marketing or lead generation agency, or have experience in these areas, you're in a great position. Don't let the hype outside deter you from transitioning to an AI agency or building an AI product business. If you can leverage AI internally to provide marketing or lead generation services to businesses, you can build a very good, high-margin business. The key is to recognize that AI can significantly reduce your delivery costs, rather than completely transforming your business model.

If you're an AI product business struggling to gain traction, Ben suggests seriously considering investing heavily in the service layer over several months or years before fully attempting productization. I think this advice is particularly important because many AI startups fail not because their products are bad, but because they rush into scaling too early and neglect the importance of services. Prove value through services first, find product-market fit, and then consider scaling.

Finally, I want to say that there are no true experts in the field of AI in 2026. Everyone is learning and exploring. This is both a challenge and an opportunity. Those who are willing to learn deeply, practice, and share have the opportunity to become pioneers in this field. As Ben said, take advantage of this huge adoption gap and jump into the AI ​​field today. Don't wait until everything is mature, because then the window of opportunity will be closed.

I believe the next few years will be a crucial period for the formation of AI business models. Those companies that find the optimal balance between products and services, those that truly create value for customers rather than chasing technological hype, and those that cultivate teams with multifaceted capabilities will be the winners of this era. And that seemingly "boring" service + product hybrid model may very well be the most sustainable and valuable AI business model.