Between 70% and 85% of AI projects in financial services fail to reach production or deliver measurable business value, according to research from Gartner, FinTellect, and the RAND Corporation. The figure is consistent across geographies — European institutions fail at the same rate as their North American and APAC counterparts. What varies by region is not the failure rate but the cost of failure, and in Europe the cost is rising. With the EU AI Act’s August 2026 deadline now weeks away, and with PSD3 and DORA creating hard architectural requirements, an AI programme that fails to reach production is not simply a wasted investment. It is a missed compliance deadline, a widening capability gap, and a compounding disadvantage that grows with every quarter the institution is not accumulating outcome data.
The reason the failure rate persists despite this urgency is the same in Europe as it is everywhere else: institutions are making the same two mistakes in sequence. They are targeting the wrong decisions first. And when they do target the right ones, they are running programmes that are structurally set up to fail regardless of where they are pointed.
Volume and value are not the same thing
The instinct that drives most AI prioritisation is straightforward and almost entirely wrong. Teams identify decisions made at high volume, calculate the labour cost saving from automating them, and build a business case on that basis. The decisions that attract AI investment first are typically account administration, STP routing, low-value card authorisation, and other high-frequency, operationally repetitive processes. These are not bad decisions to automate. They are simply not the decisions where AI generates the most material return.
The decisions with the largest value at stake combine two characteristics: high volume and severe financial consequence per error. Payment fraud, AML screening, and large credit decisions all sit in this category. They are processed at scale, made under time pressure with incomplete information, and a single decision error — a missed fraud signal, a false positive that absorbs analyst capacity, a credit model that cannot produce an ECOA or EU AI Act compliant adverse action explanation — generates a consequence that is orders of magnitude larger than a misrouted payment or an incorrectly auto-populated form.
In EMEA, regulatory urgency is now functioning as a sequencing accelerant that changes how institutions should think about this prioritisation. The EU AI Act’s Annex III classification of credit scoring as high-risk means that credit decision AI is moving onto compliance roadmaps regardless of where it would sit in a pure value-prioritisation exercise. PSD3’s real-time transaction monitoring requirements mean that payment fraud AI cannot be deferred without creating structural regulatory non-compliance. Institutions that treat these regulatory deadlines as constraints on their AI roadmap — things to be managed around — are misreading the situation. The compliance requirement and the value opportunity point at the same decisions. The deadline accelerates action that was already warranted on financial grounds alone.
The reason European institutions sequence their investment backwards is partly organisational and partly structural. High-volume, lower-consequence decisions attract AI first because the efficiency case is legible, the business sponsor is easy to identify, and success can be measured in straightforward cost-per-transaction terms. The harder decisions — payment fraud, AML, credit — require cross-functional sponsorship, more complex outcome measurement, and data that is often fragmented across legacy systems that were not designed to support the feature construction modern AI models require. They are worth substantially more. They are also substantially harder to get started on. The EU AI Act has, in effect, resolved that sequencing debate by attaching a compliance deadline to the decisions that matter most.
The three reasons AI programmes fail — and none of them are technology problems
For institutions that correctly identify the high-value decisions and commit to targeting them, the failure patterns are consistent across every market we have assessed. They break into three categories in roughly the same proportions every time.
The first and most common is governance failure, accounting for around 42% of failed initiatives. The pattern is recognisable: technology was selected before the problem was defined, the proof of concept proved the model works, and then nobody was assigned to act on it. No defined decision, no named sponsor, no measurable outcome, and no production funding pathway established at the start. Vendor demonstrations get approved before a business sponsor has been identified. PoC success gets defined as model accuracy rather than business outcome. The model achieves its technical targets and then sits in a staging environment while the organisation debates who owns it. This failure mode has a specific additional cost in Europe right now: a model that is technically complete but ungoverned is almost certainly also non-compliant with EU AI Act Annex III requirements for human oversight, technical documentation, and audit trails. Governance failure and regulatory non-compliance are not separate problems. They are the same problem.
The second category is sequencing failure, accounting for around 31% of failed initiatives. Data was not ready. Nobody admitted it until the proof of value was already running. Teams proceeded with insufficient labelled outcome data, fragmented source systems, or features that could not be constructed from available inputs, and produced models too weak to deploy. The pattern is that data readiness gets assumed rather than assessed, and discovery of the real data posture gets deferred until commitment has already been made politically and budgetarily. At that point, the incentive to admit the problem has inverted. The EU AI Act’s Article 10 data governance requirements — which mandate that training, validation, and testing data be relevant, sufficiently representative, and documented — add a compliance dimension to data readiness that did not previously exist as a formal obligation. Institutions that have not assessed their data posture against these requirements face a double failure: models too weak to deploy and documentation insufficient to pass conformity assessment.
The third category is measurement failure, accounting for the remaining 27%. The model achieved its technical targets. Nobody had defined what a good business outcome looked like in dollar terms, or assigned accountability for achieving it. Model performance metrics become a proxy for business value, no economic baseline is established at the outset, and when the programme is reviewed there is no credible answer to what it actually delivered. This failure mode is the hardest to detect from the outside because the model is in production and reports are being generated. It is only when someone asks what the false positive rate did to analyst capacity, or what the credit default rate looked like before and after deployment, that the absence of measurement becomes visible.
| Failure mode | Share of failed initiatives | Pattern |
|---|---|---|
| Governance failure | ~42% | Technology selected before problem defined. No named sponsor. No production funding pathway. PoC defined by model accuracy, not business outcome. In EMEA: ungoverned model is typically also EU AI Act non-compliant. |
| Sequencing failure | ~31% | Data readiness assumed, not assessed. Real data posture discovered after commitment made. Models too weak to deploy. In EMEA: Art. 10 data governance requirements add compliance dimension. |
| Measurement failure | ~27% | No economic baseline established. Model performance metrics used as proxy for business value. No accountability assigned for outcome delivery. |
Source: European Banking Practice analysis, consistent with findings across North America, APAC, and ANZ markets.
What is notable is that none of these failure modes are technology problems. They are programme design, organisational, and measurement problems. A more accurate model does not fix a programme with no named sponsor. A better architecture does not compensate for data that was not ready. And a higher AUC score is not a business outcome. The EMEA-specific implication is direct: the same governance, sequencing, and measurement failures that kill AI programmes in general also produce EU AI Act non-compliance specifically. Building the programme infrastructure to avoid them is not just a question of getting better returns on AI investment. In EMEA in 2026, it is also a compliance requirement.
What this means for the sequencing question
The decision portfolio argument and the failure diagnostic are related rather than sequential. Correctly identifying the high-value decisions is necessary but not sufficient. Most institutions that fail do not fail because they targeted the wrong quadrant — they fail because they ran a programme that was structurally set up not to deliver, regardless of what it was pointed at.
The EMEA implication is that the question of where to invest in AI cannot be separated from the question of how, and both cannot be separated from the question of when. The EU AI Act compliance deadline has collapsed the timeline in a way that makes the sequencing question existential rather than strategic. Institutions planning to address their AI governance in 2027 are planning to be non-compliant in August 2026.
The third piece in this series addresses what a programme that avoids both failure modes looks like in practice in Europe, and what the governance architecture required by the EU AI Act shares with the governance architecture that produces competitive advantage — which is more than most institutions expect.
Part 2 of 3.
Sources
Gartner. AI Project Failure Rates. Multiple editions, 2023–2025. FinTellect AI. Why 80% of AI Projects in Finance Fail. 2024. RAND Corporation. AI Project Success and Failure Rates. Referenced in industry literature. EU Regulation 2024/1689 (EU AI Act). Articles 9, 10, 13, 14. August 2024. European Banking Practice. Decision Portfolio and Failure Mode Analysis. Internal analysis, consistent across North America, APAC, and ANZ markets.
