71% of employees use AI weekly at the best-governed banks. The figure is identical at the worst-governed.
Deloitte Access Economics published "Banking on trust: AI governance for growth, resilience and scale" on 8 July 2026. The research draws on 24 leaders from G-SIBs and D-SIBs across 14 countries, plus 111 senior technology, AI and data employees across 16 countries, scored against a Trustworthy AI Governance Index built from five pillars and 41 indicators.
What it says
Average weekly AI user share by governance tier runs 71% at ad hoc, 54% at rudimentary, 64% at established, and 71% at optimised. The relationship is not linear. Deloitte attributes part of the ad hoc result to shadow AI, where usage is real but not visible to the institution.
The measure is a share of employees using AI for work at least weekly, reported by one senior respondent per bank. The report does not specify whether the tool is sanctioned, what kind of AI it is, or how the figure was derived.
Why it matters
Adoption rate has become a standard board metric for AI maturity. On this evidence it cannot separate a governed estate from an unmonitored one.
The two 71% figures are not the same population. At the optimised end, most of that usage is sanctioned and visible. At the ad hoc end, nobody can say what the composition is, including the executives who reported it. The report finds 72% of banks register less than half their AI use cases, mandatory controls fall to 55% at the monitoring stage, and to 44% for agentic systems.
The other finding sits in the middle of the curve, where rudimentary is lowest at 54%. Enough governance to obstruct, not enough to enable. Partial builds do not produce partial benefit; on this measure they produce less adoption than doing nothing.
Key takeaways
- Governance built halfway is worse than either end. The lowest adoption in the survey sits at the rudimentary tier.
- Employee share measures workforce uptake. It does not measure how much AI activity the institution can account for.
- Adoption metrics need a denominator. Without a use case register, there is none.
Source:
· Deloitte Access Economics — Banking on trust: AI governance for growth, resilience and scale — deloitte.com, July 2026
If not adoption rate, what should boards be measuring?