In our last article, The Generative AI P&L, we wrote about how Generative AI needs to create $3 trillion of value in order to justify the flood of VC investment in the space. This number is astronomical, but if startups are strategic to their end customers across the entire P&L, we can see a path to the Generative AI market surpassing the software market. Moreover, we expect to see a whole new breed of startups that will emerge, creating products and services that were not even conceivable prior to LLMs.

We also believe AI agents and copilots are critical to unlocking the $3 trillion market.

When we first wrote about Gen AI following the release of Chat GPT, we already knew that true winners would have to change how we all complete work. We wrote that winners would need to reimagine workflows, meaning they would need to provide a solution that was 10x better than existing solutions — or make the previously impossible, possible.

In our article on next gen automation, we wrote about why agents mean this can finally happen. Agents don’t just follow a preset set of rules, they encounter a choice and make a decision, which means that they can take actions on behalf of humans. Today, these actions are relatively simple tasks such as email or call follow-ups, but in the future these could be more complex personalised decisions previously only possible to execute with a human brain.

In practice, agents mean that multiple repetitive tasks — broadly defined as services — could be automated away. Deloitte’s 2024 Global Outsourcing Survey showed that enterprises were continuing to heavily outsource core services such as IT, cybersecurity, legal, tax, HR and finance, all of which are perfect candidates for agentic applications.

The opportunities are immense across verticals as well. In the US, for example, services account for ~78% of US GDP, which is roughly $22 trillion. As such, the potential for service‑sector automation is extraordinary.

Where is the agentic B2B gold?

Although we are still several iteration cycles away from completely autonomous agents, these cycles might come fast. The release of Manus, an autonomous agent, is a case in point. The latest GPTo3 model is also capable of extremely thoughtful reasoning. But when it comes to production grade B2B applications, today the most successful agents remain those that are great at focused tasks, and, frankly, also keep humans in the loop as they iterate on accuracy.

As such, at Dawn we are particularly excited about AI applications that focus on functions or verticals. We believe that the highest value businesses will be able to hone in on highly complex, high volume and repetitive workflows, with copilots as a wedge to build customer trust.

Below, we highlight some exciting opportunities we see across functions and verticals.

Some example killer use cases

App Development

We recently wrote about vibe-coding and the future of software, where we highlighted the opportunities that remain in the software development life cycle. Just taking the ten largest service providers in this space (the likes of Accenture, TCS, Infosys) amounts to $180 billion in a service revenue opportunity. We’ve seen the best builders in the space start by focusing on the still service-heavy elements of the process. Cogna, for example, decided to focus on more complex, under-digitised industries that rely on large consultancies for product development. By honing in on requirements-gathering for enterprises, Cogna can accelerate time to development by 10x. Lovable, on the other hand, has started by accelerating the front end software development process for both technical and non-technical users, and is focused on delivering a ‘wow moment’ for users as soon as they trial the tool, but plan to increasingly own the manual more complex parts of development.

Customer Support & Experience

Zendesk, one of the leading customer service software providers, estimates their TAM to be $165 billion. The opportunity is gigantic because almost every company will eventually use some tooling given the high volume of use cases. As such, we believe differentiation will come from the ‘deflection rate’ of these businesses — a metric that measures how many customer enquiries are resolved without needing to involve a human.

Focusing on a specific vertical within customer support could lead to higher deflection rates (this is the bet being made by Gradient Labs and Fini). Another approach is to scale fast so that every single use case can drive volume that in turn leads to higher deflection rates.

In addition to customer chats, we’ve recently seen companies focus on retention. Customer feedback collection and LLM-powered companies are growing — here, Magic Feedback is leading the charge. Pavlov, on the other hand, provides customers with agents that support onboarding, expansion and retention.

IT Ops & Services

In our recent article on AI in IT Services, we addressed how AI will restructure the $400b market of outsourced IT and security providers. These businesses are necessarily human-centric and are currently suffering from inefficiencies and structurally low-margins. We have seen a range of new interesting AI-based tooling that will transform how they can deliver their services. These include tools for managing your tenants at scale like Inforcer, improving your IT support like way.so, running an AI-first security operations centre like Qevlar, and revamping third party vendor management like Platformed.

Also In IT services, we see a large opportunity in incident management and site reliability engineering. Here agentic co-pilots can improve the speed of remediation and manage post-incident processes, gradually replacing manual human-only processes. Here, we’re excited about companies like Phoebe.ai, Anyshift and incident.io.

HR & Recruiting

HR is one of the largest outsourced functions for enterprises globally, with Delotte estimating that 57 % of companies outsource at least part of their HR activities. The main driver is the complexity of local geographic regulations, creating a $175 billion market. JetHR tackles this manual, intricate space with agentic products that automate the work of local employment consultants. In Italy alone, we estimate a $10 billion opportunity.

Within HR, recruiting is equally ripe for disruption. Companies such as Dex are pioneering models that serve both candidates — by providing best‑in‑class search experiences — and employers, by delivering highly qualified applicants.

Finance

Morgan Stanley estimates that the market for financial services ERP is $220b, growing extremely fast, and agents are about to disrupt it. Rillet and Light, next-gen ERPs, are examples of companies that combine both LLMs and traditional machine learning to enable bank reconciliation and outpace legacy providers like Netsuite and SAP. Other players are building AI into the book-keeping process on-top of existing ledgers, such as Briefcase, Stacks, and Haydn, or into the procurement process such as Omnea.

Legal

The large amount of unstructured text makes law a perfect use case for the trillion dollar legal services market. Marketplaces like Lawhive are using LLMs to their advantage by providing customers with both new revenue, and higher efficiency through agentic solutions, making them a no-brainer for high street law firms. LegalFly, on the other hand, delivers much shorter deal cycles for law firms, directly helping firms with one of their biggest pain points — working capital. Avantia is building an AI-enabled law firm, delivering on improved outcomes for customers at higher internal efficiency rates.

Intellectual property management is also having a Generative AI moment, with contenders like DeepIP, Solve Intelligence, Lightbringer and Ankar accelerating the speed to outcomes across the entire IP lifecycle.

Financial Services

In financial services, tackling fraud is a $500 billion killer use case. Our portfolio company, Quantexa, quickly realised that the non-deterministic nature of LLMs means they cannot be used in isolation in mission critical use cases like fraud detection. In their blog post, It’s Not All Generative AI (But Gen AI Helps), they highlight how using LLMs in conjunction with more deterministic models (something they term ‘composite AI’), is what it will take to win large financial institutions.

We also see a range of agentic AI applications for research and due diligence such as in asset management (Auquan, Finster), AML and KYB (Arva, Spektr).

Insurance

We are also really excited about the service ecosystem being built around insurers, where companies like Claimsorted and Strala are making waves in the $380 billion claims handling market. These next-generation providers offer traditional claims processing services through an AI-enhanced, outside-in approach, using automation and intelligence to improve efficiency. Their approach is effective because insurers prioritise a) the experience of their end customers and b) cost considerations. Claimsorted and Strala solve for both, bringing down claims’ handling time significantly, whilst also reducing human error. This approach results in direct cost improvement. However, given the complexity of claims and the criticality to insurers’ business models, next-gen providers will have to prove themselves before expanding their remit over time, and will likely start by picking some less complex verticals.

Generative AI is also allowing for faster and better insights throughout the underwriting process, with solutions like Nettle and Novee emerging to streamline data analysis, triage, and risk assessments.

Construction

Construction is an immense market accounting for 13% of GDP, and we estimate that construction firms spend $100b on software globally. In the construction sector, Generative AI is an enabling technology. Generative AI makes it possible to digitise data gathering and reconciliation in a way that was not quite possible before — either because data ingestion was too hard or because data reconciliation and cleansing was still very manual and required too much work from the foreman or back office. Solutions like Comstruct are digitising construction sites through AI solutions, and Dawn recently invested in Kraaft, which you can learn more about here.

Healthcare

Healthcare has also become an excellent use case for AI, with ambient scribes and downstream workflows being overhauled with AI solutions. As we mentioned previously, automating manual processes in US hospitals is a $350 billion market opportunity. We’ve seen companies like Phare.health transform how revenue cycle management is managed for hospitals through reducing the amount of time it takes for them to be reimbursed for medical procedures completed. We’ve also seen ambient scribes like Tandem and Voize.de become regional champions through leveraging local language, context and electronic health care systems to drive higher scribing accuracy.

The case for horizontal agents

Horizontal agents will have a big role to play in the agentic AI era, and we’re already seeing this happen with the early success of apps for solo entrepreneurs and micro SMEs. Here, emerging successes offer more generalised agents that solve mission-critical tasks while being simple enough to meet these businesses needs and price points. Companies like Sintra and Convergence are examples of European companies building to meet these prosumer and SMB needs.

Which European companies are building picks & shovels for the AI gold rush?

In order for agentic applications to scale — and for enterprises to confidently build in-house applications — the tooling that supports production grade applications still needs to mature.

Below, we highlight some unsolved pain points that startups are addressing.

Data readiness

Upstream of Generative AI applications, enterprises are increasingly voicing concerns about how unprepared their data is. Models that perform well require well-structured, relevant training data. Companies like Dataiku are leading the charge with their data application solutions. Grounding GenAI pipelines in graph and vector databases also helps, with companies like neo4j and Vespa being increasingly deployed in production to leverage internal company data.

LLM evaluation

To date, the most significant unsolved challenge appears to be evaluation — that is, determining whether a model performs as expected in real-world settings. A new generation of companies, such as Langfuse, is gaining traction by addressing this “last mile” with tracing and evaluation tools. The key challenge is turning product managers’ “gut feelings” into robust, repeatable frameworks that ensure Generative AI applications function as intended — a very difficult problem.

Governance and security

For enterprises, governance and security also remains a hot issue. A Gartner survey recently found that this was the number one issue for enterprises adopting Generative AI solutions. Here, Collibra, another of our portfolio companies, is addressing this need with its new AI governance product. We are confident that as enterprises buy more tools and build more AI applications, governance and compliance businesses like Collibra will gain even more traction. Earlier stage companies like Portia.ai are also creating a category around authentication for LLMs, which will address enterprise governance concerns.

On the security side, we already see a massive opportunity today from SMBs and enterprises needing to protect their business from AI bot attacks, with companies like Blackwall seeing significant traction. Other exciting companies in the space include Whitecircle and SplxAI, who are focused on red teaming and pentesting.

Orchestration and deployment

Finally, because it’s still extremely hard to build LLM-powered applications we continue to see an opportunity for teams to be supported with orchestration and deployment. Companies like C TWO, Doubleword, Deepset and Nexos.ai are already helping both small and large enterprises accelerate time to production for Generative AI applications. The fact that enterprise end users can readily access these services with the help of established AI companies — such as our portfolio company Dataiku — is already proving a successful model for growth.

Conclusion

Below is our market map of the 100+ European agentic companies we are excited about across many verticals and functions, as well as some of the picks and shovels empowering this build. If you’re one of these businesses — or you’re building in this space and want to have a chat — please reach out to shamillah@dawncapital.com, owen@dawncapital.com, nils@dawncapital.com and andrea@dawncapital.com. We’d love to hear from you.