Sammalkko has reviewed more than 200 companies in AI HR tech since we started investing in 2022. Thirteen made it into the portfolio. This is a write-up of the questions that most reliably separated the ones we invested in from the ones we passed on. These are not a checklist to be optimized by founders who want to impress us — they are honest indicators of company quality that are hard to fake if you haven't done the work. They are also the questions we should have been asking more consistently from the beginning; some of our earlier passes and some of our closer calls would have been easier decisions if we had asked these questions earlier in the process.
1. Is the AI doing something that was impossible before, or just faster?
The clearest indicator of genuine AI-native architecture is whether the product's core output was simply not achievable without machine learning — not just more efficient, but categorically different. A recruiting tool that uses ML to match candidates to jobs 10x faster than a human reviewer is a productivity tool. A recruiting tool that surfaces candidates from signals that no human reviewer would have the bandwidth to process — behavioral signals from internal mobility, skills inference from project histories, career trajectory modeling — is doing something categorically different. We ask founders to describe what their product produces that human labor genuinely cannot, at any cost or time. If the answer is just "speed," we probe harder for the structural AI advantage.
2. Who actually loses if you don't exist?
This question identifies whether the product is addressing a painful enough problem to survive the enterprise procurement cycle. The answer we look for is a specific person in a specific role with a specific consequence: "The Head of People Analytics at a 500-person European scale-up currently produces the workforce planning report manually in four days every quarter. Our product does it in four hours and includes forward-looking skills gap signals that the manual process can't produce. If we don't exist, that person spends 16 days a year on a report that is less useful." Specificity matters here. "Every CHRO needs better people data" is not an answer — it is a category observation. The person who loses when the product doesn't exist has a job title, a recurring pain, and a measurable cost to that pain.
3. What does the enterprise procurement path actually look like?
HR tech sells into complex buying committees. At a 500-person company, a meaningful HR software purchase typically involves the CHRO, the VP of People Operations, at least one member of the Finance or IT team for security review, and sometimes a Works Council or employee representative body in European jurisdictions. Founders who have sold into enterprise before understand this and have a clear account of the stakeholder map — who is the champion, who is the economic buyer, who can kill the deal in legal or IT review, and what each of them needs to see before signing. Founders who have not yet sold an enterprise deal often have a vague account of "the CHRO makes the decision," which is almost never accurate. We ask for the specifics of their last deal and listen for evidence of stakeholder management sophistication.
4. What is the data requirement, and do your target customers have that data?
This is where we have seen the most gaps between product promise and deployment reality. An attrition prediction model requires clean historical data on employee departures, demographic attributes, performance signals, and compensation history. A skills intelligence model requires connected records across HRIS, LMS, and performance systems. Before we invest, we want the founder to articulate specifically what data their product requires, what format it needs to be in, and whether the enterprises they are targeting actually have that data accessible in a form their product can use. The answer "we have APIs with SAP and Workday" is necessary but not sufficient — the question is whether the data in those systems is clean enough to train meaningful models. Founders who have run pilots know this from experience. Founders who haven't sometimes discover it the hard way after the contract is signed.
5. How does the product handle European data protection requirements?
This question is not secondary. It is a first-order qualification question for products targeting European enterprise customers. GDPR places specific requirements on how employee data can be processed, retained, and used for automated decision-making. The EU AI Act adds further requirements for AI systems used in employment contexts — transparency obligations, human oversight requirements, and documentation standards that affect product design. Works Council consultation requirements in Germany and Scandinavia affect the deployment process independently of the software's technical capabilities. Founders who have thought through the European regulatory context — not just GDPR compliance in the abstract but the specific provisions that affect their use case — are significantly more fundable from Sammalkko's perspective. Founders who have US-centric regulatory experience and plan to "add GDPR compliance later" should expect a longer conversation about this before any investment decision.
6. What does the customer's experience look like at month three versus month one?
The most important indicator of retention is whether the product gets more valuable over time. We ask founders to describe what a customer's experience looks like at month three that is materially different from month one. Good answers involve the product having more data to work with (training on the customer's specific patterns), the customer having made more workflow integrations, or the model having demonstrably improved its accuracy based on outcome feedback. Weak answers involve the product having been set up correctly and now running consistently — which is fine, but not a retention story. It is a commodity feature story. Products whose value increases with use tend to retain customers. Products whose value plateaus at initial deployment tend to face churn questions at the first contract renewal.
7. Which enterprise will sign a contract in the next six months, and why?
At Seed stage, we want at least one named enterprise customer in active conversation — not just "warm interest" but a prospect who has seen the product, has engaged with pricing, and has a plausible reason to be inside a procurement process by the end of the year. This is not about revenue — the actual contract value at seed stage is less important than the evidence that a real enterprise buying committee is seriously evaluating the product. Founders who have this conversation in process have gone through at least one cycle of the enterprise procurement gauntlet and have learned things from it that are irreplaceable. Founders who don't yet have an enterprise prospect in this stage are earlier than we typically invest, and we tell them clearly.
8. If the largest HR platform in your category copies your core feature tomorrow, what happens to your business?
We ask this because we want to understand the founder's mental model of defensibility. The products that we are most confident in are the ones where the answer involves data assets, customer relationships, or compounding model improvements that a large platform would find difficult to replicate quickly. The products where the answer is essentially "they probably would copy it if it got traction, and that would be a problem" are ones where we think hard about whether the window of differentiation is long enough to build a durable position before incumbents catch up. There is no wrong answer to this question — some of the best opportunities in HR tech are in the window between "this feature doesn't exist" and "the large platform adds it to their roadmap." But the founder needs to have a clear view of how long that window is and what they intend to build during it.
What These Questions Are Not
These eight questions are not a pass/fail test. A founder who can answer all of them perfectly has probably been coached through enough investor conversations to know the right framing. What we are actually looking for is the texture of the answer — the specificity, the acknowledgment of what isn't yet known, the evidence of genuine customer conversations rather than market research summaries. The founders who answer these questions with honest uncertainty about things they haven't figured out yet are often more fundable than the ones who have a polished answer to everything. The former are the ones who are actually in the problem. The latter sometimes aren't.