Written by:
Future Processing’s AI Scaling Paradox report examines why many organisations invest in AI without turning that activity into measurable business value. The report argues that successful AI scaling depends less on access to models or tools, and more on validating the use case, business case, readiness and governance before major build decisions are made.
Key takeaways
AI activity is not the same as AI value: only 6% of organisations draw more than 5% of EBIT directly from AI, while many others remain stuck between tool adoption, pilots and limited production impact.
Most AI failures are sequencing problems: organisations often commit budget before they know where AI should be applied, whether the business case is measurable, and whether data, infrastructure and governance are ready to scale.
Readiness protects investment decisions: an early AI readiness assessment can create a clear GO, NO-GO or redesign point before major implementation spend begins, reducing the risk of expensive late-stage failure.
Executive summary: why AI value starts before the build
Executive summary: why AI value starts before the build
AI Scaling Paradox explains why widespread AI activity has not yet translated into widespread measurable business value. Its core thesis is that most organisations do not fail with AI because the technology is unavailable or because the models cannot work. They fail because they make decisions in the wrong order.
Every CEO has seen how quickly an AI discussion can turn into a discussion about tools. The real executive question is different: which business decision are we trying to improve, what would make the investment pay back, and what has to be true inside the organisation for that to happen? The companies that win with AI bring this discipline to the table early, before enthusiasm, momentum and budget turn an idea into a project.
Michał Sztanga
CEO
The cost of checking whether an organisation is ready to build is small compared with the cost of discovering the problem after a project has already started. This is why the report frames AI readiness not as a technical formality, but as an investment-control mechanism: a way to create a clear GO, NO-GO or redesign decision before major budget is committed.
The report begins with the gap between AI ambition and AI value.
Many organisations report that they are using AI, but that often means access to tools, scattered pilots or isolated functional use. The report separates those input indicators from output indicators such as measurable EBIT impact, redesigned processes, reduced cost lines, production deployment and accountable governance.
The report calls this gap the Adoption Illusion. The headline adoption numbers can make boards believe the market is further along than it is. Yet only 6% of organisations draw more than 5% of EBIT directly from AI. The report argues that the difference between this group and the rest is not primarily access to technology. It is the discipline of validating the use case, business case and readiness conditions before committing to build.
For the UK market, the report highlights a use-case identification problem. The report cites DSIT research showing that 71% of UK firms have not identified a concrete AI use case. For many organisations, the practical question is not whether AI is available, but which operational problem it should address and what “better” means in measurable business terms.
The report identifies three cycles that explain why AI programmes stall. The strategic cycle begins with board pressure and turns into a mandate without a validated use case. The operational cycle is a triple lock between legacy infrastructure, data readiness and talent constraints. The regulatory cycle appears when governance is treated as something to add after deployment, rather than a design requirement from the beginning.
The scaling paradox is visible in sectors that adopted AI early as well as in sectors with heavier legacy and regulatory constraints. In media and advertising, the report points to the difference between widespread experimentation and real business value. AI can create advantage only when it is tied to proprietary data, creative workflows, governance and process redesign.
In financial services, the report frames AI adoption as an engineering and integration problem as much as a strategy problem. AI investment has to be assessed against legacy architecture, compliance obligations, data controls and the full cost of making systems production-ready, not against standalone productivity claims.
The CFO perspective is central. The report compares the cost of early diagnosis with the cost of late failure. S&P Global Market Intelligence puts the average sunk cost of an abandoned AI project at $7.2 million. Future Processing’s report argues that a diagnostic before build can create a controlled decision point: GO, NO-GO or redesign. A NO-GO at this stage is not failure; it is avoided waste.
This is where the report becomes directly relevant for finance leaders: AI readiness is not only a technical assessment, but a way to preserve control over investment decisions before major implementation spend begins.
The cost asymmetry between a diagnostic and an abandoned build is not only about the numbers, but mainly about retaining control over the decision. A diagnostic creates a point where you can decide whether to proceed or stop before you commit real money, while without it you start spending before fully understanding the risks, and stopping later becomes much harder. So, this ultimately becomes a question of governance and investment discipline.
Mikołaj Gwóźdź, CFO, Future Processing
The cost asymmetry between a diagnostic and an abandoned build is not only about the numbers, but mainly about retaining control over the decision. A diagnostic creates a point where you can decide whether to proceed or stop before you commit real money, while without it you start spending before fully understanding the risks, and stopping later becomes much harder. So, this ultimately becomes a question of governance and investment discipline.
Mikołaj Gwóźdź, CFO, Future Processing
Mikołaj Gwóźdź
CFO Future Processing
The report recommends two diagnostic paths.
- Track A applies where an organisation has a board mandate but no validated project; it begins with AI Opportunity Mapping and ends with a GO/NO-GO gate on the business case.
- Track B applies where an organisation already has a pilot or candidate list but has not assessed readiness; it begins with an AI Readiness Assessment across data and infrastructure, security and governance, legal and compliance, and organisational readiness.
Both tracks lead to the same decision artefact: an AI Readiness Score, two independent GO/NO-GO gates and a prioritised roadmap. The report’s practical conclusion is that diagnosis should precede delivery. AI value starts before the build, at the moment an organisation asks not only whether it can build something, but whether this specific use case should be built now, on these foundations, for this measurable result.