CD Consulting R&D

CD Consulting R&D · KB-AI · entry 1 · report summary · 9 September 2026 · published 9 September 2026

The 2026 AI Barometer: where Belgian financial services actually stand on AI

If your AI strategy is ready but nothing ships, twenty Belgian institutions explain the execution gap.

Start here — if the subject is new to you

An industry barometer is a recurring temperature check: the same sector is surveyed year after year, so the interesting signal is not any single number but the movement between editions. The one summarised here measures how Belgian banks and insurers are adopting artificial intelligence — from everyday assistants that draft e-mails to agentic AI, systems that plan and execute multi-step tasks with little human intervention between the steps.

Why it matters beyond the sector: financial services are heavily regulated, risk-averse and data-rich — close to a worst case for careless AI deployment and a best case for observing how serious organisations actually absorb the technology. What blocks them, and what works for them, tends to preview what other industries will meet a year or two later.

What the Barometer is

The 2026 AI Barometer is an annual survey of AI adoption in the Belgian financial sector, run by the consultancy Sailpeak with FinTech Belgium and Vlerick Business School (Prof. Bjorn Cumps). This edition combines a quantitative survey of 20 institutions — covering, by the organisers' count, over 90 % of the Belgian retail banking market and roughly two thirds of private banking — with nine long-form interviews of senior AI executives at BNP Paribas Fortis, Belfius, ING Belgium, Ethias, Baloise, DKV/ERGO, Bank van Breda, Keytrade Bank and one bank that chose to stay anonymous. Respondents sit mostly in the functions doing the work — data & AI offices (45 %), transformation offices (25 %), IT (20 %) — which makes the answers operator-led rather than communications-led, a point the editors themselves make.

What it finds — ten takeaways, condensed

The panic is over. Only 20 % of leaders now say their organisation moves too slowly on AI, down from 54 % a year earlier; one in five plans to more than double AI investment, and the largest cohort now expects a 2–3× return. The editors' warning: "confidence is not embeddedness".
Two use cases have reached real production scale: coding assistance (the widest deployment in the sector) and customer-service automation. Around them cluster document processing and contact-centre transformation — one institution handles loan-origination documents autonomously in 85 % of cases; another pushes millions of customer contacts a year through an AI-enhanced channel.
The most discussed use cases are the least deployed. Wealth management, financial forecasting and credit scoring sit close to zero in production. The barrier is not the models: 60 % of organisations have not automated the preparation of the documents their AI depends on, and fully automated document readiness stands at 5 %.
The bottleneck is no longer the technology — it is legacy IT integration (top barrier at 60 %), data quality (55 %) and process redesign. A recurring practitioner lesson: do not build in a sandbox; work inside the production environment from the start.
Governance is centralising — nearly two thirds of institutions now run AI through a dedicated unit with C-suite representation — but the governance of third-party AI use (brokers, agents, advisers handling client data with their own tools) is a named blind spot.
The productivity dividend is real and badly counted. Copilot-class assistants are the most widely deployed category of all; the realised value exceeds what was modelled, yet organisations admit they have no full inventory of the time and cost actually saved.
AI is "10 % technology, 90 % people" — the interviewees' consensus. Employee adoption ranks with regulatory compliance as the top rollout blocker, while HR scores zero as an expected AI impact zone in the survey.
Agentic AI is the theme of 2026 — and mostly a slide, not a system. Half of organisations grant agentic AI zero decision-making authority; three quarters govern it without a dedicated framework. A small frontier group is already running multi-agent architectures in production, including one customer-facing "super-agent". No institution lets fully autonomous agents make consequential decisions without a human in the loop.
The workforce question is unanswered. 80 % of organisations have no formal plan for AI-driven workforce change although a clear majority expects headcount reductions within three years; 42 % have assigned accountability without producing a plan.
The gap is execution, not strategy. The report's closing thesis: the leaders will be the institutions that did the unglamorous work — process documentation, data governance, model-migration inventories — before the technology made it urgent.

A reading

Three threads deserve more attention than the headline numbers.

The data floor. The single most quotable statistic is the least dramatic one: 5 % fully automated document readiness. Every ambition higher up the stack — agents, autonomous decisions, hyper-personalisation — rests on that floor. The Barometer is, read closely, a report about plumbing.

The HR paradox. The function responsible for building workforce capability scores zero as an expected AI impact zone, while 80 % of organisations have no workforce plan. The report notes the coincidence without forcing the conclusion; the conclusion rather forces itself.

Universal oversight, unexamined oversight. Every institution keeps a human in the loop for consequential decisions — reassuring, until one asks how well that loop performs under volume. That question is precisely the subject of a paper summarised in the neighbouring chapter of this knowledge base ("AI Agents Push Humans Out of the Loop" — a critical reading note): human oversight degrades exactly where agentic deployment stresses it. The Barometer documents the sector building the loop; the paper explains why the loop wears out. Read together, they bracket the real 2026 question.

Limits of the exercise

It is a consultancy-led publication, not an academic study. Sailpeak organises the Barometer and sells adjacent services; the academic partner lends method, not independence. Percentages on a panel of 20, with multi-select questions summing over 100 %, should be read as directional, not statistical.
The panel self-selects toward the convinced. The organisers acknowledge the skew "toward institutions where AI investment is most material"; the interviewees are heads of AI talking about their own programmes. Absent are the institutions with nothing to show.
A third of the report is showcase material — profiles of FinTech vendors (debt-collection agents, stablecoin rails, document-fraud detection, credit-decision automation) written in the vendors' own voice, with their own metrics. Useful as a market map; not evidence in the survey's sense.

Verdict

Verdict: methodologically modest, directionally credible. The survey numbers are soft, but the qualitative picture — production scale in two mundane use cases, near-zero deployment in the glamorous ones, data readiness as the binding constraint, workforce planning as the open flank — is consistent across twenty institutions and nine interviews, and consistent with what international surveys report at larger scale. As a one-hour orientation on where a regulated European sector actually stands with AI in 2026, it is hard to beat — provided the showcase chapters are read as advertising.

Source

2026 AI Barometer — In the financial services industry, Sailpeak · FinTech Belgium · Vlerick Business School, 2026. This summary is based on the "Key Results" and "Full Interviews" sections (site captures of 26 June 2026; site re-checked live on 9 September 2026). Figures quoted are the report's own.