Beyond the Trough: The Foundations That Determine Whether AI Scales in Jersey FS
GenAI hype is over. Disciplined foundations are the differentiator.
Most AI programmes are stuck. Not because the AI does not work. Because the foundations were not built before the capability was deployed.
The pattern is recognisable: a promising pilot, early enthusiasm, an expansion plan — and then the rate of incidents rises, exceptions multiply, evidence gaps appear, and the programme quietly stops scaling. The tools are still in place. The problem is everything underneath them.
Gartner's hype cycle positions GenAI as having entered the Trough of Disillusionment (Gartner, 2026b). The data behind that diagnosis is specific: organisations spent an average of $1.9 million on GenAI initiatives in 2024, yet fewer than 30% of AI leaders report that their CEOs are satisfied with the investment return (Gartner, 2026b). That is a significant commitment producing a disappointing signal — and it is the evidence that explains why foundational discipline has become more commercially important than the capability itself.
For Jersey FS leaders, this is not bad news — but only for those who treat it as a diagnostic rather than a disappointment.
Why a Regulated Environment Is Actually an Advantage Here
The trough is brutal for organisations that deployed AI without governance, data discipline, or a clear investment framework. They are now paying the remediation cost. A regulated firm that has maintained governance discipline throughout — that built the control baseline before scaling, that trained for behaviour rather than prompting, that established evidence standards before deployment — is in a fundamentally different position. The trough does not punish discipline. It punishes the shortcut.
Jersey's regulatory environment turns out to be a structural advantage at this moment. The JFSC's expectations around governance, evidence, and responsible adoption mean that the most compliant firms have been doing the foundational work that less-regulated environments are now being forced to retrofit. The Jersey AI Playbook calls this "ethical differentiation" — using responsible AI adoption as a competitive advantage (Government of Jersey et al., 2026).
The Five Foundations Scorecard
Gartner identifies a consistent set of foundational enablers that separate organisations that scale from those that stall (Gartner, 2026b). The scorecard below maps each foundation to observable signs of low and high maturity. Score your programme honestly before committing to the next phase of scaling.
Table 6 — The Five Foundations Scorecard: Self-Assess Your AI Programme
| Foundation | Signs of low maturity (score 1–2) | Signs of high maturity (score 4–5) |
|---|---|---|
| AI-ready data | No named ownership for key datasets. Permissions not reviewed. Retention rules exist as policy documents only. Quality unmonitored. | Named owners per workflow dataset. Permissions reviewed and remediated. Retention operationalised. Quality monitored with anomaly triggers. |
| AI engineering | Each deployment is bespoke. No standard patterns for guardrails or validation. Hallucination incidents occur without systematic prevention. | Standard patterns for input guardrails, grounding, and validation. Deployment checklist applied consistently. |
| ModelOps and lifecycle | No change control for prompt or model updates. Drift not monitored. Regression testing absent. Evidence log degrades over time. | Change control applied before any update. Regression test plan in place. Drift monitoring with named reviewer. Evidence log maintained. |
| Governance and assurance | Risk tiering informal or absent. Exception process exists on paper but not used. Evidence pack cannot be assembled within 48 hours. | Risk tiering applied to all use cases. Exception process active and monitored. Evidence pack assembled on demand. |
| People readiness | Training completion tracked; behaviour change not measured. Manager reinforcement assumed not confirmed. Escalation route unused. | Training connects to behaviour signals (bypass rates, exception patterns). Manager cascade confirmed by name. Escalation route actively used. |
If two or more foundations score below three, the programme is not ready to scale. The investment case for the next phase should be a foundations investment, not a capability expansion.
ModelOps: The Foundation Nobody Funds Until Something Breaks
When a prompt is changed — even slightly — to improve output quality, does anyone test whether that change has degraded performance in a downstream workflow? When a model is updated by a vendor, does anyone confirm that the behaviour of the deployed capability remains within approved bounds?
In most organisations, the answer is: not systematically. The consequence is drift — gradual, hard to attribute, and expensive to remediate retrospectively. Gartner identifies AI engineering and ModelOps as critical disciplines for secure, repeatable, and scalable enterprise AI delivery — not optional engineering refinements, but the infrastructure that keeps deployed capabilities reliable after their launch date (Gartner, 2026b). The prevention is comparatively cheap: change control protocols, regression test plans, drift monitoring dashboards, and a clear process for approving updates before they go live.
Why Process Redesign Is the Value Unlock, Not the Tool
Gartner's research on AI and productivity makes a point that is directly relevant to the foundations argument: organisations that amplify individual productivity without redesigning the surrounding process capture a fraction of the value available to those that recalibrate or reset the workflow entirely (Gartner, 2026h). The five foundations exist not primarily to protect against risk — though they do that — but to create the architecture in which deeper process change becomes possible and evidenceable.
A programme with weak data ownership, no ModelOps discipline, and untrained people cannot safely pursue the process redesign that generates green money. It can only pursue the marginal productivity gains that produce blue money stories. The foundations are what convert an AI programme from a productivity narrative into a value programme.
The Specific Challenge of AI-Ready Data in a Jersey FS Context
Jersey FS firms often hold client data across multiple systems built at different times, under different data models, with different retention assumptions. Trust structures, beneficial ownership data, suitability assessments, AML documentation — these sit in different repositories with different access models and often without clear ownership at the workflow level.
When AI is deployed into these environments, it makes content from across all of these repositories more discoverable. That is the efficiency benefit. It is also the risk. The AI-ready data standard is a direct response: before AI is deployed against any repository, permissions are reviewed, retention rules are confirmed and operationalised, ownership is assigned, and quality is assessed.
Privacy-by-Design Is Architecture, Not Compliance
The JOIC has been explicit: responsible AI and robust data protection must evolve together, and organisations must address lawfulness, data minimisation, purpose limitation, and retention from the outset (Government of Jersey et al., 2026). For regulated firms, this is a core design requirement, not a compliance annotation. Every AI capability deployed in a workflow involving personal data should have explicit answers to: What is the lawful basis? What is the minimum data required? What retention rule applies? Who is accountable as the capability evolves?
Closing question: If GenAI is in the trough, are you investing in the foundations — AI-ready data, ModelOps, governance architecture, people readiness — that will let you scale when the market moves? Or are you still funding pilots on top of foundations that will not hold?
Next in the series — Part 5: Agentic AI in regulated firms. Why the arrival of AI agents changes the governance problem entirely.
Sources: Gartner Hype Cycle for Artificial Intelligence (2026b); Gartner AI Is Coming for Inefficiency, Not People (2026h); Jersey AI Playbook, March 2026.
Professional disclaimer: The views expressed are my own and are shared for general informational purposes. This content is intentionally non-confidential and does not include proprietary, client-specific, or employer-sensitive information. It is not legal, regulatory, investment, or professional advice; readers should seek appropriate advice for their circumstances.
Financial services leader (20+ years) | Lean Six Sigma Green Belt | London School of Economics (LSE) AI Leadership Accelerator
See where your firm stands
The free diagnostic scores your firm across the twelve operating-model instruments in about eight minutes.
Take the diagnostic →A version of this article was first published on LinkedIn as part of the Sequence body of work. This is the canonical version.
The views expressed are the author's own and shared for general information. This content is intentionally non-confidential and does not include proprietary, client-specific or employer-sensitive information. It is not legal, regulatory, investment or professional advice. © 2026 Paul Soares. Sequence is the trading name of Sequence Advisory Limited (in formation) · hello@sequenceadvisory.com