Learnings across our think tank | a Chatham House Summary
At the Executive Technology Board meeting in Singapore, cross-industry leaders compared AI transformation across very different starting points - from greenfield platform roll-ups to century plus-old, regulated incumbents. Despite variance in maturity, themes converged on: simplify and standardize first; keep the core thin; push differentiation to data products and AI at the edge; and treat culture as the hardest constraint.
Key takeaways
- Simplify → Standardize → Automate with AI. Consolidate platforms (campaign-to-cash, source-to-pay), kill Shadow BI and model sprawl, and make dashboards reconcile to systems of record.
- Lean core, rich edge. Delay or narrow “big-bang” core upgrades; invest in a strong data layer plus AI services on top. “Evolution, not revolution” with disciplined project sunsetting.
- Culture beats tooling. Break “boulders into pebbles,” explain the why, and sequence visible wins. Silent resistance is common; crises create permission to change.
- Board ↔ management alignment. Close the horizon and capital gap with a simple taxonomy: Deploy (foundations), Reshape (AI in value chains), Reinvent (new revenue).
- Multi-speed Asia. Country context, regulation, and company culture drive different cadences; run “favorite-child” pilots where velocity is highest, localize elsewhere.
What’s working
- Outcome-anchored portfolios (productivity, cost-to-serve, NPS/CSAT, loss ratio).
- Platform consolidation and data cleanliness eliminating Shadow BI.
- Hands-on adoption (copilots, job-specific training), reverse mentoring, and show-don’t-tell use cases.
- Explicit “kill criteria” and steady de-risking over quarters, not “big transformation” programs.
Sticking points
- Talent & the frozen middle. Juniors and seniors gain from AI; mid-levels often regress (risk aversion, unlearning cost). Results vary by domain.
- Agent risk & governance. No fit-for-purpose agent platform; gaps in observability, traceability, and explainability; regulated workflows can’t tolerate one-off failures.
- Architecture churn. Vendor/model changes trigger costly refactors; need model portability and guardrails to curb citizen-dev and agent sprawl.
- Sustaining change without a crisis. Quarterly budgeting fights multi-year transformation.
Architecture stance
- Prioritize enterprise platforms, integration rails, and a canonical data layer; keep model choice portable (LLMs ↔ SLMs; on-prem/edge where needed).
- Plan for drift monitoring, policy placement, identity/entitlements for agents, and evidence logging.
- Knowledge graphs pay only with multiple reuse cases; otherwise miss hurdle rates.
Talent for the age of AI
- High appetite for learning across roles when content is practical and local-language.
- Move org shape toward a “bow-tie” (thinner middle, empowered edge) where appropriate.
- Treat agents as first-class “workforce”: identity, join/move/leave, SLAs/SLOs, owner of record.
Practical steps
- Publish the Deploy/Reshape/Reinvent capital mix and a sanctioned “easy path” (model-agnostic abstraction, API gateway, event bus, secrets/PII policy, telemetry schema).
- Stand up Agent Observability MVP (plans, tool calls, IO logs, policy hits, human interventions, drift alerts).
- Replace Shadow BI with productized, system-reconciled exec dashboards.
- Target the frozen middle: incentives, role redesign, and hands-on wins on brownfield code.
Metrics that matter
- % processes on standardized platforms; % Shadow BI eliminated
- Throughput/FTE, avoided hires, cycle-time and unit-cost deltas
- Agent incidents per 1k actions, drift alerts, mean time to correction
- Lock-in/portability index; portfolio yield vs promised ranges
Open questions
- Minimum viable agent observability spec acceptable to boards/regulators
- Where to standardize (integration, portal, telemetry) vs. allow BU-specific freedom
- How to price maintenance to reflect AI productivity gains
- Workforce planning for explicit X humans + Y agents by function
Executive Technology Board (c)