Future Processing

The Goodfirms interview with our CEO, Michał Sztanga, covers questions that come up constantly in technology delivery right now. One is the consulting layer before implementation: whether the business case is real, whether the data foundations can support the use case, and whether the organisation can actually absorb the change once the system is live. Another is accountability: the shift towards people who take ownership of outcomes, not just outputs. That is a meaningful distinction at a time when AI makes it easier to build things faster, but not necessarily easier to ensure they are the right things to build.   Enjoy the read - the link is in GoodFirms’ comments.

A look back at last week's AI in Claims panel in Munich. Justyna Szymańska-Laskowska shares her perspective on an evening that brought together claims leaders, a cybersecurity voice and a technology provider for a conversation that was open, honest and focused on real-world challenges. More personal insights from Justyna in the post below.

When an AI project gets abandoned, most reviews account for the financial cost. The organisational cost rarely appears in any of them. S&P Global puts the average cost of an abandoned AI project at $7.2 million. Gartner expects 60% of AI projects to be abandoned through 2026, most of them for want of AI-ready data. The financial figure is visible in hindsight. Paweł Pustelnik, our COO, addresses the cost that standard business cases do not capture: what happens inside the organisation when a project does not deliver. The "AI Scaling Paradox" report maps where this pattern comes from and what changes it. It covers the three cycles that keep organisations investing without return, the conditions that separate the 6% generating measurable EBIT from the rest, and what a structured decision looks like before the build budget is approved. There is also a section on UK and DACH specifically, where the window to act is still open. Read the "AI Scaling Paradox" report: https://lnkd.in/dAakqMKV

The business case for energy efficiency depends on energy having a real cost. When energy becomes cheap, the economic incentive to invest in saving it weakens. Consumption continues, but the logic for improving it quietly disappears. Tomasz Słupik, CEO of „Energopomiar” Sp. z o.o., places this in a broader context: the geopolitical instability shaping today's energy markets is unlikely to resolve within a decade. In that environment, electrification and genuine efficiency improvement are not just environmental goals. They are resilience strategies for organisations and countries that depend on stable energy supply. Energopomiar has been working across the full energy lifecycle for 76 years, and this question sits at the centre of much of what they do. We spoke with Tomasz Słupik as part of IT Insights EnergyTalks. Watch the full episode here: https://lnkd.in/gaXCHbVC

Thank you to everyone who joined us in Munich for our event: AI in Claims - Efficiency Miracle or Trust Killer? this Wednesday.   The conversation turned out to be exactly what we hoped for: direct, grounded and full of perspectives that do not often end up in the same room. The questions the panel raised, around trust, governance and what AI deployment actually requires in practice, are ones the industry will be working through for some time.   Our panellists Axel Kotulla (.msg), Dr. Stefanie v.d.Bergh (Allianz), Anton Horn (Envoy Security) and Pascal Béchaz (Yarowa) brought exactly the kind of honest and practical perspectives this conversation needed.   This evening would not have been possible without joined efforts of the Future Processing and Ventum Consulting teams. Special thanks to Piotr Piękoś, Justyna Szymańska-Laskowska, and Hajo Boerste who hosted the event.   More conversations like this to come - stay tuned!

72% of CEOs are now personally accountable for AI, double the figure from a year ago. 64% admit investing in the technology before they understood its value. 61% acknowledge the board is accelerating transformation without sufficient justification.   The pressure is real, and what it produces is projects that are launched before anyone has checked whether the business case, data, infrastructure, and governance are ready to support them.   Krzysztof Szabelski, our AI Innovation Partner, describes what this looks like from the inside, and what changes when organisations approach it differently.   The report maps all three cycles that keep organisations investing without return, and sets out what a structured decision looks like before the build budget is approved: 🔶 the strategic, where board pressure produces mandates without a validated business case 🔶 the operational barriers that surface mid-project 🔶 the regulatory requirements that arrive too late   Read The AI Scaling Paradox Report: https://lnkd.in/dAakqMKV  

Another Sailing Day on the Solent behind us. Every year we bring together a group of clients and partners from across the specialty insurance market for a day on board Lutine. This year, Dawid Glawdzin, Director of Partnership Development, and Eva Rodzik, Delivery Manager, were on board alongside a great group of people from across the market. Good company, good conversation, and a few hours away from the desk. Thank you to everyone who joined: Matt Wood, Nathan Lownds, Tim Scott-Simmons, Julie Serle, Julie Blowers, Samantha Haynes and Sarah Carpenter.

Most reports about AI open by telling you that AI matters. The AI Scaling Paradox opens by pointing out that almost everyone already agrees, and that the agreement is where the trouble starts. The easy explanation for why most organisations are not generating return is that the technology is not ready, or that they lack the budget. Neither holds. The models work and they keep getting cheaper. The separation between firms that capture value and those that stall is more mundane and more within their control: the order in which they decide. That is the argument Tomasz Hanke, our Chief Market Analyst, makes in his introduction to the report, and the premise on which the rest of it is built. The report maps what follows from that premise: three self-reinforcing cycles that keep organisations investing without return, four industries where the same structural gap takes different forms, and the economics of inaction in numbers a CFO can take to the board. Tomasz argues that the window to act is narrowing. On current trajectories, it is measured in months. Read The AI Scaling Paradox Report: https://lnkd.in/dAakqMKV

In a controlled PoC, agentic AI for claims management tends to perform well. Production is a different environment. Piotr Piękoś, our Head of Insurance Practice, argues the core problem is architectural. He published "Insurers are building agentic AI on architectures that are not agent-capable" in IT Finanzmagazin. Most core insurance systems were not designed for autonomous agents. The gap shows up in specifics: 🔶 APIs layered as REST wrappers over SOAP, with polling intervals of 5-15 minutes 🔶 No consistent event streams for real-time state changes 🔶 No auditable decision logs required by regulators 🔶 Agents acting on outdated or conflicting data Before reaching for more sophisticated models, most insurers need to address integration architecture first. The article is in German. If you are attending AI in Claims: Efficiency Miracle or Trust Killer?, we would be happy to continue the conversation there. Read the full article: https://lnkd.in/g37uhpBD

There is a point in most AI projects where better prompting stops being the answer. The model changes, the context drifts, and manual adjustments start creating new problems faster than they solve old ones. That is where context engineering begins to matter, and why having a systematic approach makes a difference at scale. The 33rd Data Community Poland Trójmiasto meetup is taking place in Gdańsk on Wednesday, hosted by Grzegorz Brodny, our Senior Cloud Data Engineer. Hubert Jegierski, our Senior Machine Learning Engineer, will speak about what that shift looks like in practice: prompt debt, how context shapes model output, and how to automate prompt improvement before the manual process becomes unmanageable. He will also introduce DSPy as a framework that reduces model dependency and makes the search for better prompts more measurable. If this is the kind of problem you are working on, it is worth joining. https://lnkd.in/dCrrew56