tl;dr: AI investment is missing a huge opportunity in the form of industry-specific AI apps designed to integrate with legacy business processes in medium-sized firms in trad industries. This investment should go into industry-specific incubators that bridge the capability and capital gaps that prevent startups from building these real-world AI solutions that drive immediate impact.
So much time and money is flowing into AI today, but it’s nearly all focused on building foundational technology: massive models or compute infrastructure. The promise driving this investment is that AI will transform the real world: AI will make healthcare (or logistics, or law enforcement, or copywriting, or whatever) cheaper or better (or faster, or in more languages, or available to more people, or whatever).
This transformation will take much more than just good models and huge compute. These will have to be translated into actual applications that real-world people and organisations can use. Applications that fit into existing business processes in existing industries. Without an ecosystem of small tech companies doing this translation work, AI’s promise of transformation will go unfulfilled.
So this is the problem: Most startups are trying to become the next unicorn, chasing enormous markets that promise outsized returns. Too few startups are focused on building translational AI applications that make a tangible, immediate impact on the existing processes of medium-sized firms. This is a major market opportunity that almost no one is systematically exploiting.
I’ll unpack some of the reasoning below.
Why AI apps for traditional industries?
Because traditional industries are an enormous market in any country. Traditional industries — like construction, logistics, and insurance — form the backbone of every national economy. They don’t have the razzmatazz of digital and other new industries, but they nonetheless produce the majority (always more than 70%) of GDP. These industries predate widespread modern digitalisation, so they usually operate using legacy processes that could be transformed by AI applications.
The kinds of applications I have in mind are not general services that multiple industries can use, like project management tools (e.g., Asana or Slack). Instead, I’m thinking of applications that work only in one particular industry, that address the issues and processes specific to that industry.
An example from the construction industry: An application for construction site operations that processes drone, GPS, and mobile device data to automate building progress updates and material input/output management. Medium-sized construction firms using this application improve project management efficiency by up to 20%. These savings come from things like increasing the accuracy of just-in-time materials delivery to construction sites. Cumulatively, this reduces costs from project delays and increases overall project profitability. (BTW, this is an application built by a company called Unearth, which was acquired by Procore in 2023.)
Why medium-sized firms as clients?
Medium-sized firms in traditional industries are the sweet spot for startups building industry-specific AI applications. They’re large enough to see significant benefits from business process improvements using new technology and be able to pay for lightly customised products.
Why not target small or big firms? Big firms don’t need these applications; they’re large enough to staff in-house tools teams (or hire consultants) to build applications that are heavily customised to their own firm-specific processes. At the other end of the spectrum, small firms in traditional industries can’t afford lightly customised applications; they often lack the resources to buy them and the internal capacity to use them effectively.
The capability and capital gaps.
Medium-sized businesses in traditional industries are high-value targets for startups, but few startups are going after them. Why? Two reasons: a capability gap and a capital gap.
The capability gap.
It’s hard to build AI products that integrate into legacy business processes. This takes a different kind of product management approach, one that focuses on building products for small actually-existing markets, not large potential ones.
The problem is that most startups use a product management approach that is implicitly optimized for venture capital (VC) funding. The VC approach designs products to have enormous total addressable markets (TAMs). Enormous TAMs necessitate general utility with speculative business models and execution plans. Those lead to high failure rates in pursuit of unicorn status.
This VC approach to product management doesn’t work for startups trying to build AI applications for medium-sized firms in traditional industries.
The TAM for such an AI application is necessarily much smaller because it’s built to address the particular needs of a specific industry. This makes it valuable for firms in the industry, but less valuable and harder to sell to firms in other industries. (Though the TAM is small, finding product-market fit is easier, getting to positive cashflow is faster, and the small TAM naturally deters competition.)
Addressing industry-specific needs means deeply understanding and integrating with existing business processes. To do this requires a different approach to product management centred on an awareness of what I call meaningmaking.
Previously, I’ve written that “meaningmaking is the act of making subjective decisions about the relative value of things,” and that only humans can do meaningmaking work. Meaningmaking is relevant here because it separates what parts of legacy business processes must be done by humans and what AI systems can do better than humans. Meaningmaking is a powerful framework for understanding how to build AI applications that work interdependently and optimally with humans.
Unfortunately, a meaningmaking-based approach to product management is not yet systematised or commonplace — this is the capability gap which needs to be filled.
The capital gap.
Beyond the capability gap, startups building AI applications for medium-sized firms often fall between the cracks when it comes to funding. They’re a poor fit for VC funding because their TAMs are too small (see above). They’re also a poor fit for private equity, because they’re not yet cashflow positive.
These startups would become cashflow positive more quickly if they had access not just to capital, but to industry expertise and connections. Specifically, industry insiders who know which business processes are most in need of AI support and the decision-makers who control procurement budgets.
The capital gap I see is for finance, industry-specific expertise, and access to procurement.
Filling both gaps with industry-specific incubators.
The answer to both the capability and capital gap is an incubator focused on building AI applications for medium-sized firms traditional industries. Such an incubator would do two things:
Develop product management capabilities: Train startups in a product management approach optimized for small but high-value markets. This means focusing on building products that integrate seamlessly into existing business processes, prioritizing cashflow-positive business models, and understanding the specific needs of traditional industries (like construction, logistics, insurance, brick and mortar retail).
Provide structured access to industry leadership: Give startups access to industry operations leaders who can help them identify high-potential processes for automation and efficiency improvements, as well as purchasing decision-makers who can shorten sales cycles.
The tech Mittelstand: A new vision for AI startups
The rewards are huge if we can successfully build this ecosystem. What will result is a a technology Mittelstand: The Mittelstand (“the middle estate”) is the large population of SMEs in Germany, known for their innovativeness, long-term stability, prosperity, and contribution to Germany’s industrial success. Instead of only trying to build tech unicorns, we should be trying to create conditions that enable a large number of startups to emerge, which mature quickly into a large number of small-headcount, stable-cashflow technology companies which provide a highly diverse range of industry-specific AI applications to serve medium-sized firms in traditional industries.
Medium-sized firms are under-optimized, not because the technology doesn’t exist, but because the right kinds of startups don’t exist in sufficient numbers. By addressing the capability and capital gaps, we can unlock a large, underserved market—one that could increase efficiency and profitability for thousands of businesses, ultimately strengthening the broader economy.
Why now?
AI has the potential to transform how businesses operate, but the current focus is skewed toward the development of underlying technology. AI’s missing middle is the application of that technology in ways that drive immediate value for real businesses. Medium-sized firms — too large for off-the-shelf solutions, too small for custom-built AI from large tech providers — are the ideal target. And the tech startups that will serve these mid-size firms will necessarily be small and scrappy, and there will have to be many of them.
So we need to rethink both how AI products are developed and how they are funded. By focusing on specificity, cashflow, and real-world application, we can create a profusion of AI companies that are individually small but collectively crucial to the future of work and the economy.
We should start building that future now.
What size constitutes medium?
The amount of effort to build the amount of data engineering, data negotiations (as stated brilliantly by Sarah Constantin at https://sarahconstantin.substack.com/p/the-great-data-integration-schlep)
is almost the same as a large enterprise, so might as well go after the big one
I can see someone doing the same thing for the medium sized firms for no other reason than
1. client's leadership is desperate and see the benefits of this AI integration,
2. these medium sized clients are the biggest firms where the supplier knows the leadership personally.
3. the supplier's costs fit these medium sized firms' budgets
One obstacle to this approach is that historically getting the sales approach right for organizations of this size has been quite challenging.
- Direct sales of more than $100 on anything that is not a standardized product used the same way by everyone is tough to do without some sort of salesperson involvement (either direct or through a retail/reseller)
- Phone-based selling can work well for tailored offerings up to about $10K, but beyond that buyers typically want to interact with someone in person (even post-COVID)
- Field-based sales forces typically need offerings above $100K to be sustainable
Midsize organizations budgets for this kind of thing tend to fall in between, which is one of the reasons why they struggle to find solutions tailored for them. It's not obvious to me how AI tools change this dynamic, as this is primarily about psychology rather than technology.