Something structural has shifted in financial technology delivery, and the numbers behind it deserve scrutiny. Projects that used to consume two years now reach production in six months. Espen Skogen, chief executive of RocketFin, has watched this compression happen inside trading platforms, risk visualization systems, and execution infrastructure for tier-one institutions. His reading of it runs against the popular narrative.
The compression is real. The question is what survives it.
When you take a two-year delivery cycle and squeeze it into six months, something gets removed. The evidence emerging across the industry suggests that firms getting this wrong remove the wrong thing. They remove the senior judgment, keep the process, and wonder why the output degrades.
Financial services carries an 82% AI failure rate, the highest across industries. When failure costs are amortized, the true expected cost per successful deployment lands near $5.2M. That figure includes roughly $800,000 in sunk costs per abandoned medium-complexity project before termination.
These are not technology failures in any meaningful sense.
An analysis of 140 enterprise AI implementations found that 77% of failures were organizational. Only 23% traced back to model performance, data quality, or integration complexity. The rest came down to strategy, governance, and change management.
Skogen has held this position for two decades, well before AI made it fashionable. If you think you have a technology problem, you've misdiagnosed it. It's a people problem wearing a technical mask. The AI failure data now provides the empirical backing for what he observed the hard way inside major financial institutions.
💡 The pattern worth noting: the models work. The tooling works. What fails is the human layer that decides what to build, how to constrain it, and when to stop.
Here's where the compression gets dangerous. AI handles implementation and testing at a pace no human team can match. Code generation, test coverage, boilerplate infrastructure. All of it moves faster than most delivery organizations know how to absorb.
Speed amplifies whatever quality of thinking sits behind it.
A team with strong architectural judgment now ships a two-year system in six months. A team with weak judgment ships two years of mistakes in six months. The velocity is identical. The outcomes diverge massively.
This is where the financial analogy becomes precise. Poor decisions made at AI speed accrue technical debt at AI speed, and the interest compounds. The 42% of companies that abandoned most of their AI initiatives in 2025, up from 17% a year earlier, paid that interest in full. They scrapped nearly half of their proofs of concept before production.
Skogen's reading of this data is clinical. The abandonment wave reflects firms that treated AI as a cost-cutting instrument. They asked how many people the tooling could replace. The firms delivering results asked a different question entirely: how much more can our best people accomplish with this leverage.
The evidence on specialization supports this. Institutions using finance-specialized AI talent report significantly higher success rates, with domain-experienced specialists achieving implementation nearly 80% faster than generalist counterparts. Generalist AI practitioners excel at model-building while lacking understanding of financial systems, compliance mandates, and operational constraints.
That gap matters enormously in regulated finance. A generated trading component that violates a regulatory constraint is worse than no component at all. The person reviewing that output needs to hold the technical architecture, the legal framework, and the business consequence in a single view.
AI produces output. Senior practitioners produce judgment. The value sits in the combination.
This is the structure RocketFin built before the AI wave arrived, and the wave has validated it. Small teams of senior specialists, no generalist padding, no headcount for optics. The model was already outperforming when humans wrote every line. Research on team size found smaller teams operated two or more productivity indices above larger teams, and adding staff to a project increased cost by 350% while generating 500% more defects.
AI multiplies that advantage. Give a five-person senior team the implementation capacity of fifty, and the coordination overhead that suffocates large organizations simply never appears. The senior team keeps its speed of decision and gains speed of execution.
⚠️ The inverse also holds. Give a bloated, process-heavy organization the same tooling and you multiply the bloat. AI does not fix a broken delivery structure. It accelerates it.
A pattern emerges across the implementations Skogen has observed. The firms extracting real value from AI compression share three characteristics.
1. They start with the problem underneath the problem.
Incumbent solutions get questioned before anything gets automated. Automating an inherited process locks in the assumptions that process was built on, many of them outdated. The willingness to rebuild from first principles determines whether the six-month delivery produces something worth having.
2. They staff for competence, not coverage.
The instinct in large institutions is to scale headcount when a project grows. The evidence points the other way. Strip the team down to people who can actually perform, then let AI carry the volume work those extra bodies used to absorb. The senior people stop managing coordination and start applying insight.
3. They keep quality non-negotiable at any speed.
Code quality and architecture remain the foundation. Compressing the timeline changes nothing about that. What changes is where the human attention goes. Implementation and testing shift to the tooling. Review, architecture, and consequence analysis stay with the people, and those people need the depth to do it properly.
For senior executives at large financial institutions, the strategic implication is worth sitting with. Only 38% of AI projects in finance meet or exceed ROI expectations. The gap between that 38% and the rest is a strategy and execution gap. The technology is available to everyone at roughly the same price.
What is scarce, and what the data shows getting scarcer in relative value, is the senior judgment that turns velocity into results. The capable hands.
Skogen's position is that this scarcity restructures the entire delivery market. When a small team with genuine depth delivers in six months what a large program delivered in two years, the traditional consulting structure loses its central justification. Bodies stop being the product. Perspective becomes the product.
The firms that understand this early gain a compounding advantage. Competence attracts competence, and a small senior team operating with AI leverage learns faster with each engagement. That learning becomes an asset that appreciates.
The two-year project is gone. The six-month project is here. What remains open is who your institution trusts to hold the judgment that makes those six months count.
So the question for anyone running technology delivery in a large financial institution becomes this: when the implementation work moves to the machines, do you actually know which of your people provide the insight that makes it worth anything?
https://www.pertamapartners.com/insights/ai-failures-financial-services