I registered Kvistlund in November 2020. That required explaining, repeatedly, to potential LPs why a seed fund focused specifically on AI-native logistics software in Europe made sense when the largest logistics tech investors in the world were based in San Francisco. This is that explanation, written at the time, for the public record.
The short version: I spent 7 years building logistics optimization systems at a Nordic 3PL and 4 years as VP Engineering at a Series C freight-tech company that was acquired in 2019. I have a specific and detailed view on where the software leverage is in European logistics. That view is worth more deployed as a fund than as anything else I could do with it.
What I observed building logistics software
When I joined the 3PL in the early 2010s, the optimization software we ran was rule-based. Route planning was done with heuristic algorithms that a team of operations researchers had tuned over years. Demand-driven replenishment used safety stock formulas from industrial engineering textbooks. Carrier selection ran on rate sheet lookups. These systems worked adequately for a stable, predictable operating environment. They did not adapt well to variation.
Over seven years, I watched three specific patterns that convinced me the entire stack would need to be rebuilt. First, the gap between what the rule-based systems recommended and what experienced human operators knew to be right was large, constant, and impossible to close through rules. A route planner who had worked a specific urban delivery network for 15 years knew things about that network — building access patterns, traffic at specific times, driver performance characteristics — that no rule could capture because no one had ever systematically encoded them. The human knowledge was better than the software knowledge, and the humans were retiring.
Second, the data we were generating was not being used. Every shipment, every route, every inventory movement, every carrier interaction produced data that went into a database and was used only for reporting. Nobody was training models on it. The irony was clear: we had millions of labeled examples of good and bad logistics decisions, and we were using them only to produce management dashboards, not to improve the decisions themselves.
Third, the vendors weren't going to fix this. The major ERP and TMS vendors selling to European logistics operators were enterprise software companies whose revenue came from implementation services and annual maintenance fees, not from product improvement. The incentive to cannibalize a profitable professional services business with a better ML-based product simply wasn't there. The established vendors would add AI features on top of their rule-based cores, market them aggressively, and continue generating the vast majority of their revenue from the existing stack. Genuine ML-native logistics software would have to come from startups.
Why the moment was right in 2020
The pandemic compressed a decade of supply chain digitization into 18 months. Operations that had been managed through phone calls and spreadsheets discovered, under extraordinary pressure, that they needed better software. The business case for supply chain software investment became visible to boards and CFOs who had previously treated it as a back-office operational concern. Budget opened up in ways it hadn't for a decade.
Simultaneously, the cloud migration of logistics operations — WMS, TMS, ERP — had accelerated to the point where the data was now accessible in ways it previously wasn't. A startup building ML-based freight visibility in 2015 would have needed to implement EDI integrations with each carrier manually over years. By 2020, the API coverage for major carriers had expanded substantially, and the cloud TMS vendors had webhook-based event streaming that made data aggregation tractable for a small engineering team. The infrastructure cost of building logistics AI had dropped significantly.
And the founder quality was there. The generation of engineers who had started their careers in supply chain digitization and spent a decade fighting with the limitations of the legacy stack was now experienced enough to start companies and young enough to have strong ML engineering credentials alongside their operational knowledge. I had spent years talking to people in my network who fit this profile. Several of them were thinking about starting companies. That was the early dealflow signal.
The European specific bet
The case for European logistics AI is a geographic complexity case. The EU is a single customs area with highly variable operational environments across member states. A 3PL operating cross-border freight from Hamburg to Warsaw to Budapest is navigating three distinct road infrastructure profiles, three driver labor market contexts, three fuel pricing environments, and regulatory requirements that vary at the national level despite EU harmonization efforts. The software that handles this complexity well is more valuable than the software handling a comparable US domestic corridor because the problem is harder and the switching cost of replacing it is higher.
Scandinavian logistics operators specifically have been early adopters of supply chain software relative to their European counterparts. Nordic 3PLs, FMCG distributors, and retail chains have historically invested in operational technology at higher rates than comparable Southern European operators. This creates a stronger early-adopter market for seed-stage logistics AI products in the Nordic region than you would find starting from comparable products in other European markets.
The thesis was straightforward to articulate but took genuine conviction to execute in 2020, when most logistics tech capital was flowing to US-based companies addressing US markets. The bet was that European logistics AI companies, built by European operators with European distribution relationships, were better positioned to capture the European market — and that the European market, despite being less liquid in terms of venture activity, was large enough and structurally interesting enough to produce fund-returning outcomes at seed stage. Four years in, with Fund I deployed and Fund II actively investing, I think that bet was right. The evidence is in the portfolio. The full outcome won't be clear for another several years.
What we got wrong at launch
We are transparent in our investor reporting about the assumptions that have proven incorrect, and this record should reflect the same honesty. We expected enterprise adoption velocity to be faster than it has been. We expected the legacy system replacement cycle in large European 3PLs to be more advanced by 2023-2024 than it is. Large logistics operators move slowly in enterprise software procurement. We knew this going in, and we still underestimated how slowly. The thesis is intact. The financial model timelines need more conservatism than we applied at launch.
We also underestimated the degree to which European logistics AI founders would face US investor pressure to expand to US markets before their European business was mature. Several founders in our network have navigated this badly — taking capital from US investors who pushed US GTM before the European business model was proven, and ending up with a capital-intensive US go-to-market motion in a market where they lacked the operator relationships to compete effectively. We try to be explicitly useful in helping our portfolio founders resist this pressure until the timing is right. It's one of the specific ways that operating in the European market for a decade matters for the support we provide.