Selin Daldal on Why the Next B2B Software Companies May Look Like Services

Selin Daldal is an Investment Manager at Alstin Capital, a Munich-based early-stage fund focused on B2B software.

Alstin invests in fast-growing late seed and Series A companies across Europe, typically writing €2-8m initial tickets. The fund focuses on categories like fintech, insurtech, regtech, cybersecurity, climatetech, and other B2B software markets, and is known for taking an active role after investment, especially with support across sales, social media, PR and marketing, expansion strategy, and network.

Before Alstin, Selin studied Management and Technology at the Technical University of Munich, worked on green and digital transformation projects as a Consultant at McKinsey, analysed sustainable fintech at Exton Consulting, and worked in project financing at Pacifico Energy Partners. At Alstin, she has moved from Investment Analyst to Investment Manager and works closely with companies including etalytics, Blockbrain, Secfix, and NORBr as investor and board observer.

What makes Selin’s perspective interesting is the question she has been spending more time on recently: AI-enabled services. It is a broad label, but the underlying shift is specific. In some B2B categories, the customer does not only want a better tool. They want more of the work taken off their plate. Compliance handled. Financial workflows completed. Customer requests resolved. Energy systems optimised.

That changes how these companies need to be understood. Some will look like service companies in the beginning because humans are still involved and the customer is buying an outcome. But if AI increasingly changes how the work is delivered, repeated, measured, and improved, the company can start to behave very differently over time.

That creates a new question for investors: when is an AI-enabled service a scalable company in the making, and when is it still a better-run service business?

We sat down to talk about why AI-enabled services are becoming more interesting, what separates a scalable model from a better-run service company, how to think about humans in the loop, and when B2B customers may prefer buying the outcome rather than another tool. 

Q&A

1. You mentioned that you have been spending more time looking at AI-enabled services. What made the category interesting to you in the first place?

What initially drew us to the category was a simple observation. Services have always represented one of the largest markets in the economy, yet they have largely been outside the scope of venture investing because revenue scaled almost linearly with headcount. AI is starting to change that.

As automation takes on an increasing share of the work, professionals can focus on supervision, judgment and the exceptions AI cannot yet handle. Over time, revenue can grow faster than delivery headcount. Revenue per delivery professional increases, gross margins expand, and services start to exhibit characteristics that were previously reserved for software businesses.

There is also a resilience argument that I think is underappreciated. Companies selling AI tools become more exposed every time the underlying models improve because parts of their product risk becoming native model capabilities. AI-enabled service providers benefit from those same improvements. Better models make delivery faster, cheaper and better, while the customer relationship, domain expertise and accountability remain with the provider. Every improvement in AI strengthens the economics of the business instead of threatening its position.

2. In some B2B categories, customers do not only want a better tool. They want more of the work taken off their plate. Where do you see that most clearly?

We see it most clearly in industries where the biggest obstacle is not the technology itself, but getting incumbent service providers to adopt it. The organizations are often not particularly tech-savvy, there is no internal champion driving adoption, workflows rely on implicit knowledge and fragmented processes, change management is difficult, and adopting AI often means disrupting an existing business model. We see this in legal, insurance claims processing, accounting and property management.

These markets are also structurally attractive for new entrants. Customers already outsource the work, which means a replaceable budget already exists and there is no behavioural change required. In many cases, switching providers is operationally far simpler than reducing internal headcount.

That is the pattern behind companies like Lawhive, which originally sold software to law firms before becoming a law firm itself, and Harper, which started with AI tools for insurance brokers before becoming the broker. Where software adoption fails, the door for AI-enabled services opens.

3. What separates an AI-enabled service that can become a scalable company from a better-run service business?

One of the clearest indicators we look for is whether revenue per delivery professional is increasing over time. That tells us whether automation is absorbing an increasing share of the work and whether the underlying economics are improving as the business scales.

The opposite can look deceptively healthy. Emergence Capital coined the term "Mirage PMF" to describe companies with strong revenue growth and solid customer retention where every incremental dollar of revenue still requires another delivery professional. Revenue and delivery headcount grow in lockstep, gross margins remain flat, and what looks like an AI-enabled business is ultimately a very well-run services business. There is nothing wrong with that model, but it is not venture-scale.

What determines which path a company follows is how it handles edge cases. Every AI-enabled service encounters situations the models cannot yet handle. The question is not whether those edge cases exist, but whether they become a ceiling or a learning opportunity. If every case a professional resolves feeds back into the system, automation improves, professionals become more productive, and margins expand over time. If those cases remain permanently manual, the business gradually starts to resemble a traditional services business.

4. A lot of these companies still have humans in the loop early on. When is that a strength, and when does it become a limitation?

Human involvement in these businesses is not a bug, it is a core feature of the model. Customers are not just buying labour, they are buying accountability, trust and responsibility.

In some cases, human oversight is required by regulation. A lawyer must sign off on legal advice, a CPA must authorise a tax filing, and an insurance adjuster remains responsible for the final coverage decision. In others, it is inherent to the nature of the work. A property manager needs to respond to emergencies on site, and a recruiter still has to assess cultural fit. In both cases, the human layer is not incidental to the product, it is part of the value proposition.

That is also what gives these businesses a structural advantage over pure software. Every improvement in AI makes the service faster, cheaper and higher quality, but it does not replace the accountability the customer is paying for. The provider owns the customer relationship, the domain expertise and ultimately the responsibility for the outcome.

It becomes a limitation when responsibility starts scaling faster than automation. If a company expands into increasingly complex or higher-liability work without automation keeping pace, human effort begins to grow faster than productivity. At that point, margin expansion stalls and the economics gradually converge with those of a traditional services business.

5. At late seed and Series A, what does an AI-enabled service need to prove before you believe the model can scale?

The first is whether the economics are improving in the right direction. Revenue per delivery professional should be increasing over time, but the mechanism behind it matters just as much. Is edge-case data being captured and fed back into the system so that automation continuously improves? That is what creates a genuine data flywheel rather than just a narrative around one. And if the company has a price advantage over incumbents, is it driven by structurally lower unit costs through automation or by subsidization? 

The second is execution. Unlike many early-stage software companies, AI-enabled services rarely have to create a new market. Customers are already paying someone to deliver the outcome, just through a different provider. That means the challenge is less about proving demand and more about winning on execution. The companies that win acquire customers quickly while simultaneously building compoundable assets such as edge-case data, domain expertise and track record.

The third is the retention and expansion path. At that stage, we do not necessarily expect proven expansion, but we do want to see the foundations for it. Is the underlying workflow structurally recurring? Is the provider becoming embedded in the customer's operations through accumulated context, workflow ownership and switching costs? And is there a credible path to expanding into adjacent services for the same customer over time?

6. If we sit down again in a year, what do you think will be clearer about when B2B customers prefer buying an outcome rather than another tool?

The biggest open question is whether the economics actually hold at scale. The entire category is being priced on the expectation of venture-scale outcomes, but services are historically home turf for PE. Can AI-enabled services demonstrate the margin trajectory and growth velocity that justifies software-adjacent multiples, or will PE absorb these companies at services-level valuations before they get there? In a year we will have much better signal from how growth rounds are priced and whether gross margins are actually converging as these companies scale.

The second thing that will become clearer is the competitive dynamic between AI-enabled services and incumbents adopting AI. Can established providers successfully layer copilots and agents into their existing operations? Do the two models end up competing for the same customers, or does the market naturally segment? We could see AI-enabled providers dominate high-volume, standardized work while incumbents retain more bespoke or liability-intensive matters. We could also see segmentation by customer size, with AI-enabled services winning SMBs while traditional providers remain stronger in the enterprise. Or we may find that both models coexist across the same markets, competing for the same customers. Either way, it will become much clearer over the next year whether the category can live up to the expectations currently being placed on it.