Insights
What we have learned, written down
Three clusters: AI governance, systems integration, and sector notes. Each article names its claim and the evidence behind it.
Clusters
UNIT 01AI governance
Responsible AI management systems, ISO/IEC 42001, and what an auditable AI deployment actually requires.
7 published
Systems integration
OT and IT convergence, legacy estates, and the engineering decisions that determine whether an integration holds.
7 published
Sector notes
Maritime and port operations, manufacturing, and the general SME base.
6 published
Articles
UNIT 02
Systems integrationThe access control your unified data layer just lost
Connecting scattered data is step one. The moment that unified data feeds or trains a model, the permissions that used to separate finance from HR do not automatically come with it.
Systems integrationMost AI projects fail on the data, not the model
Gartner puts the share of enterprise AI projects failing to deliver real business value at 85 percent. Buying a capable model before the underlying data is connected does not shorten that path, it lengthens it.
Systems integrationThe productivity gap is a systems problem, not a technology one
McKinsey's November 2025 survey found 78 percent of companies using AI regularly, but only 33 percent scaling it across the organisation. MIT NANDA put the share of pilots reaching lasting financial gain at around 5 percent.
Sector notesSingapore SME AI use tripled. The gap with larger firms still widened
IMDA's data shows 95.1 percent of small businesses already using at least one digital tool, and AI use tripling between 2023 and 2024. The constraint is not appetite. It is picking the wrong first thing.
AI governanceAI assurance is cheaper before go-live than after
Assurance proves, with evidence rather than intentions, that a system does what it should and keeps doing it after handover. Retrofitting it onto something already in production costs materially more.
AI governanceAgentic AI went commercial. Traceability is what to buy on
Once an agent is billed for completing tasks rather than demonstrated against benchmarks, the question stops being how well it scores and becomes whether it finished, and whether it can prove it.
Systems integrationNot every problem needs a model. Some need a lookup table
The question has shifted from whether AI can do something to whether it can run at scale, on a budget, without falling over. Cost now scales with adoption rather than with the value being created.
Sector notesSeven funding instruments and still no structure
Singapore SMEs operate in a policy environment built to lower the cost of adoption. The binding constraint is not funding or tools, it is the absence of alignment between leadership intent, process definition, and execution discipline.
AI governanceWhat your AI vendor contract is probably missing
Most AI agreements are recycled SaaS templates. That structure assumes software which does what it is told, performs consistently, and does not learn from your data. None of the three holds.
AI governanceShadow AI is already inside your organisation
MIT's enterprise research puts personal AI tool use among employees above 90 percent. Prohibition does not remove that risk. It removes your visibility into it.
Sector notesThe first use case decides whether the programme survives
MIT's research is explicit that most pilot failures come from poor use case selection rather than model quality. The first two or three carry disproportionate weight, and the instinct about how to choose them is inverted.
Sector notesFive questions before you buy anything
Buying an AI tool before checking whether the business can use it well does not just waste the money. It creates operational debt that the next project has to work through.
Sector notesThe software AI market is crowded. The hardware one is not
Every enterprise platform has an AI layer and every SaaS product has a copilot. Intelligence embedded directly into equipment and physical infrastructure is a category that has not yet been reinvented.
Systems integrationLocal inference versus API: where the break-even actually sits
Running models on your own infrastructure is conditionally correct, not universally so. The break-even depends on sustained token volume, regulatory constraint, and whether you have the MLOps capability in-house.
AI governanceFrom human-in-the-loop to AI-on-the-loop
In several operational domains AI now checks work more reliably than human reviewers, particularly at scale and over long horizons. The governance regimes still assume a person is accountable, so the architecture has to change, not just the tooling.
Sector notesWhere AI is actually safe to deploy
Deployment safety is no longer a yes or no question. It varies by industry, application and regulatory tier, and the organisations getting a return are matching scope to data readiness and the right level of human oversight.
Systems integrationWhy the demo works and the production system does not
The gap is not the model. It is the operational architecture around it — context handling, evaluation, versioning, observability and cost control — none of which a demo is asked to have.
AI governanceSystems that know when they do not know
A softmax classifier cannot output 'unknown'. It is obliged to guess, and it will assign high confidence to inputs it has never seen. In embedded infrastructure that is not a statistical curiosity, it is a liability.
AI governanceA working pilot is not evidence of production readiness
AI systems fail silently rather than crashing, which is why a standard risk matrix mis-ranks them. A quiet hallucination in a citizen-facing service can outrank a visible outage that everyone noticed and fixed within the hour.
Systems integrationStop tuning prompts. Start cleaning data
Prompt tuning has a low ceiling because it cannot fix what a retrieval system is actually reading. On one project, cleaning the knowledge base moved the hallucination rate further than three weeks of prompt engineering had.