Before founding Sammalkko, I spent five years building the people analytics function at a 4,000-person Finnish telecoms company, and then another five years building a workforce intelligence SaaS product that eventually reached 120 enterprise customers across the Nordics. The product was acquired in 2019. I started the fund in 2021. Three years of investing in this space later, I want to write about what I learned as an operator that I could not have learned as an investor reviewing decks from outside.
This is not a post about the fund or our portfolio. It is a personal account of what building and selling people analytics products taught me about where the value actually lives, where the resistance actually comes from, and what questions I now know to ask that I would not have known to ask before.
The Organizational Immune System
The most important thing I learned building people analytics from the inside is that the hardest problem is not the data or the models. It is the organizational immune system — the set of implicit norms, status relationships, and psychological contracts that determine which insights can be acted on and which ones sit in a dashboard forever.
When I was Head of People Analytics at the telecoms company, we built a fairly sophisticated attrition prediction model. It was technically good — better precision than anything we had used before, well-calibrated, with reasonable coverage across employee populations. The model was essentially useless. Not because the predictions were wrong. Because the moment you tell a line manager that one of their team members is at high attrition risk, you are implicitly making a claim about their management. Even when the model's signal was driven by compensation gap relative to market rate — a factor entirely outside the manager's control — the conversation was experienced as accusatory. The insights landed badly, were resisted, and over time stopped being surfaced by HR because they generated conflict rather than action.
This is not a data science problem. It is an organizational design and change management problem. The way we eventually made attrition insights actionable was to completely reframe the output — instead of surfacing individual attrition risk, we started surfacing team-level compensation positioning relative to market, alongside benchmarked turnover rates by role category and tenure band. The same underlying signal, framed as a workforce planning input rather than an individual risk flag, was acted on by managers who had completely ignored the original output. The insight did not change. The organizational entry point for the insight changed.
The HRIS Data Quality Problem Is More Structural Than It Looks
When I was selling the workforce intelligence product, we ran into a version of the same data quality problem at nearly every enterprise we onboarded. The problem was not missing data. It was definitional inconsistency within a single company's HRIS. The same field — "job family," for instance, or "performance rating" — meant different things in different business units that had been merged through acquisition, had different legacy HRIS implementations, or had different HR leaders who had made independent decisions about how to use the standard fields.
The consequence was that the first three months of every enterprise engagement was effectively an exercise in data archaeology: understanding what the data said on the surface, what it actually meant, and where the definitions diverged in ways that would corrupt any cross-unit analysis. This work was expensive, invisible to the customer, and not something we could accelerate without making the customer's own HR data team a co-owner of the problem.
When I evaluate people analytics startups today, one of the first questions I ask is: "Walk me through your data onboarding process for a customer whose HRIS data has field-level definitional inconsistencies across business units." The answers fall into two categories. Companies that have been through this tell me about the specific problem, their approach to resolving it, and usually share that it took longer than they expected the first time. Companies that have not been through this describe their integration API or their data schema and miss the question entirely. The second group will hit this wall with their second or third enterprise customer, typically after the first one has been absorbed as a learning experience.
The Credibility Window Is Short
One thing I understood much more clearly as a builder than I do reviewing pitches from outside: the credibility window for a new people analytics product is short. Enterprise HR leaders are skeptical buyers. They have typically seen multiple vendor promises that did not deliver. When they give a new platform a chance, the window during which they are willing to attribute disappointing results to normal implementation friction rather than product failure is maybe 90–120 days.
If the product does not produce a recognizable insight in that window — something that surprises the HR team, confirms something they suspected but could not quantify, or surfaces an opportunity they had not seen — the account goes into maintenance mode. The platform becomes reporting infrastructure rather than a strategic intelligence tool, the renewal conversation is difficult, and the expansion that would make the economics of the deal attractive never happens.
What this means for product design is that the fast-to-value question — what can the platform show the customer in the first 30 days, using whatever data is available at that point, before the full integration is complete — is not a nice-to-have. It is the question that determines whether the account becomes a reference or a cautionary tale. The best products in our portfolio have a deliberate "day-30 insight" designed into the onboarding process: something specific enough to be surprising, reliable enough to be accurate with partial data, and useful enough to produce an action.
What CHRO Buyers Are Actually Buying
I spent a lot of time in enterprise sales conversations with CHROs and VPs of People. The sophisticated ones — and most of the people buying workforce analytics tools are sophisticated, because they have survived long enough in their roles to have seen previous technology cycles come and go — are not buying the dashboard or the prediction model. They are buying internal organizational permission.
A CHRO who wants to rebalance headcount across business units, who wants to shift compensation budget toward retention rather than acquisition, who wants to reduce reliance on expensive external search for senior roles — these are political decisions within the organization, and they require CFO alignment, CEO support, and line manager buy-in. The people analytics platform gives them the evidence base to have those conversations with credibility. The product is not the analytics. The product is the ability to show up in the executive team meeting with a number that is defensible rather than a judgment that is disputable.
This shapes how you should think about the product's output layer. The CHROs who are most engaged with our portfolio companies are the ones who are using the platform's outputs in executive presentations, board presentations, and headcount planning conversations with CFOs. When the product is generating evidence that travels outside the HR function, it is doing something valuable. When it is only visible inside the people analytics team, it is at risk of being re-categorized as a specialty tool in the next budget cycle.
The Selling-to-HR Paradox
There is a structural irony in selling workforce intelligence to HR leaders: the people who understand the product most deeply are often the ones with the least budget authority, and the people with the budget authority are often the ones who need the most convincing that people data is as strategic as financial data.
When I was building the product, our most satisfied users were people analytics practitioners — HR data scientists, workforce planning analysts, HR business partners with quantitative backgrounds. They understood what the product could do, pushed its limits productively, and gave us the feedback that made the product better. But they rarely controlled the budget. The budget conversations were with CHROs or CFOs who had different information needs, different relationship with data, and different criteria for success.
The products that bridged this gap best had two distinct product experiences: a deep analytical environment for the practitioners who ran the analysis, and a clean executive summary layer that translated practitioner outputs into the format that budget-holders could evaluate and act on. Building both of these well, with the practitioner layer informing the executive layer, is harder than it sounds and almost always underscoped in the initial product roadmap.
We are not saying that people analytics products should dumb down their output for executive consumption. We are saying that if the executive layer is an afterthought — if the path from practitioner analysis to something a CHRO can put in front of a CFO requires manual reconstruction by the HR team — the product will be perpetually underfunded because the value creation is invisible to the people controlling the budget.
Building and selling people analytics taught me that the interesting problems in this space are not primarily data science problems. They are organizational behavior problems that require data science to solve. The companies that understand that distinction early build products that land differently than the ones that do not.