How does the real AI ecosystem look? Who are the parties that can be the actual winners in the AI business? Perhaps the answer lies in taking lessons from the winning models in other ecosystems, such as online advertising and mobile apps.
Artificial intelligence and machine learning have been hot terms in the tech and startup market for some years. But we are still often confused about what AI and AI applications actually mean, how they are used, and their business use case. It doesn’t help that corporate management, startup founders, and software vendors use the terms very loosely to get better valuations for their companies or products.
We have started to see more academic studies about AI ecosystems. For example, this study on how different companies in different countries can utilize AI has some interesting findings. It has mainly been the Internet tech giants (Google, Amazon, Microsoft, Alibaba, and Tencent) that have been able to utilize AI in their business when offering end-user services. The authors of the paper also expect AI development from tech giants to shift from horizontal solutions to more vertical solutions in the future. But while data analytics and AI can enable a tech company to expand to new verticals, it is not easy in practice if the company doesn’t know the vertical.
Actors in the AI ecosystem
Looking at the current AI business, we can see many kinds of companies with very different resources. There are at least six categories of companies that are working on developing AI:
- Consumer internet companies that offer services to consumers and businesses
- Embedded AI solution builders
- Independent software vendors (ISV) that offer AI solutions to enterprises
- Software solution providers and systems integrators (SI) that develop AI into their solutions
- Companies that provide AI development components tools to other software companies and developers
- Companies that enable consumers to use AI with their own data.
There are fundamental differences in how companies in these different categories access data. Companies in category #1 (e.g., Google, Amazon, Meta, Uber) collect and keep a lot of data for themselves. Companies in category #2 (e.g., automobile companies that make self-driving cars) use data for their products to collect and use. Categories #3, #4, and #5 don’t have access to actual production data; they merely offer tools and solutions to companies that have data. Category #6 is a new emerging category: companies that help consumers use their own data and have AI-based applications to utilize the data.
There are also differences between companies that have a horizontal or vertical focus. Consumer internet companies have done quite large-scale AI development, but they still typically utilize it primarily for their own end-users services. For example, Google uses data to make its advertising and search more effective, while Amazon and Alibaba use data to sell more products. Embedded AI solutions are typically vertical, whether it’s self-driving cars and airplane or automated manufacturing processes.
ISVs also typically make software for a specific vertical – for example, automated insurance claim processing or fraud detection. Categories #4 and #5 try to make more horizontal solutions to have tools and components to build AI and use them for all kinds of needs.
For the emerging Category #6 the focus has been especially on how consumers can obtain their own data and utilize it. But we can estimate that there will be more horizontal platforms to collect and pre-process the data and then make it available to vertical applications, such as health, financials, and entertainment.
The online advertising business has utilized data for 20 years, so we can probably learn something from it for the AI market too. Initially, many people believed that traditional advertising companies would dominate the online ad market because they had competence, customer relationship, and good resources. That’s why everyone thought newcomers like Google could not really be successful in that market.
Now we know the reality is different. Google has come to dominate the online advertising market, as has Facebook, while traditional advertising and media agencies have mainly continued in other roles in the ecosystem (e.g., creating content and specifying target audiences), and publishers have lost their business in advertising.
There have been thousands of tech and data companies that have tried to get into the online advertising market. Software vendors have created components to analyze site visitors, measure click rates, manage ad inventory, and target ads based on content, visitors, or context. But very few of them have been big successes; and often that success came in the form of Google or another giant either using their solutions or acquiring them. Successful companies have mainly been the ones able to manage the whole chain end-to-end, from buying ad inventory to targeting an audience and showing ads to visitors. Narrow tech solutions haven’t survived well in this ecosystem.
The mobile application market is even more centralized than the online advertising market, with Apple and Google dominating the marketplace. There are actually very few really successful apps (i.e. the ones that have hundreds of millions or more users) and a long tail of apps that have very few users. But that is not the whole picture. Mobile apps have an essential role for many other companies – supermarkets, gyms, airlines, public transportation services and all kinds of organizations offer them as a part of their customer service.
Can we learn something from these examples for the AI business? Most probably many things, and a complete analysis would take more time and space than I have. Still, we can make some observations and develop at least some hypotheses:
- Success stories in one vertical often require a full end-to-end solution, not just components or a specific technology for it. This means a company should be able to have access to data, implement a one-stop-shop for users and get enough volume in the market.
- It is hard for legacy or incumbent companies to transition to a new technology. I have earlier written how a digital company must be built on new digital processes, rather than trying to digitalize and automate old processes – i.e. a retail chain cannot become a new Amazon just by automating its old processes and the work their current employees do.
- Application platforms and marketplaces are volume games; the biggest ones dominate the market. But this doesn’t mean that it will be limited to only one or two platforms; there can also be vertical-focused success stories, such as a platform and marketplace for health services. And in the early days of a platform, the quality of apps and the user experience are more relevant than big numbers.
- Systems integrators and solution providers are mainly service companies: they must adapt to any technologies and platforms that dominate the market, and have the competence to build solutions on those.
- The objective is not necessarily to create services that are independently profitable businesses in their own right – many services can be a part of the whole customer experience. Even if they have no price tag or are not independent products, they are an essential part of the business and user experience.
All these points can give us some good ideas of what kind of AI companies we will see and which of them will be successful.
At the moment, the AI and ML markets are still fragmented. We have companies that build basic capabilities, companies that try to build very targeted tech solutions, and both new and old companies trying to understand how to utilize AI in their business. Many companies are still looking for their role in the ecosystem.
Where will the best success stories in the AI area come from? I would list the following: (1) companies that offer a total end-to-end solution to consumers or business users for a specific need or vertical, (2) new companies that build all their operations on data and AI to offer superior services for a vertical, and (3) general or vertical platforms and marketplaces for AI applications that offer the best access to high-quality data, apps and users.
I could also name the most likely losers in AI: (1) legacy companies that simply try to use some AI to automate their existing operations, rather than design new digital processes, (2) tech companies that make generic tech components for a small part of an AI process (although some of them could make reasonable exits to larger companies), and (3) generic AI development tool companies.
A new ecosystem always needs time to evolve, and many companies must learn by making mistakes. New ecosystems oftentimes tend to surprise somehow, and things don’t go as predicted. But those who can solve end-users’ problems, find a successful business model and offer excellent customer experience are always successful sooner or later. These fundamentals also apply to the AI business.