See
Dashboards instrument the organisation and make stable, recurring questions visible.
A vision for enterprise intelligence
From dashboards
to decision
intelligence.
Enterprise AI will not be won by organisations that merely deploy assistants. It will be won by those that engineer the semantic, organisational and experiential conditions for AI to become a reliable partner in business understanding.
Our decisions deserve better than another dashboard.
For decades, dashboards, reports and applications promised that better instrumentation would create better understanding. They expanded access to data and disciplined operational conversations—but also produced a dense fog of apparent precision.
The sentence is an opening challenge, not the destination. The next stage will be defined by systems that understand business intent, reconstruct context, reason over governed definitions and generate the right representation at the right moment.
Explore
Conversational AI lets people investigate, reformulate questions and assemble provisional answers.
Understand
Analytical AI reconnects facts to assumptions, history, constraints, intentions and accountability.
Dashboards still matter.
When the question is known, repeated and well understood, a stable report remains an efficient instrument.
Specialised observationIntelligence must assemble the answer.
When the question is ambiguous or still forming, the system must help formulate hypotheses and connect evidence.
Intent-driven explorationRead the full executive argument
This vision consolidates three previously published essays—“No pAIn, No gAIn,” “AI is the new UI,” and “Our decisions deserve better than another dashboard.” Their strong professional reception gives credibility to a position that has already resonated with practitioners, technology leaders, data professionals and business decision-makers.
The ambition is not to replace existing analytical practices overnight, but to move progressively toward a richer operating model. Reports and dashboards will coexist with conversational, exploratory and generative environments designed for questions that are emerging, ambiguous or dependent on context.
Together, the three ideas form a causal progression: decisions expose the limits of predefined representations; analytical AI requires trusted business context; and once intent and context become computable, the interface itself can become adaptive.
Three essays.
One architecture of meaning.
The document’s themes are not sequential chapters. They are interdependent threads: generated interfaces need context; context needs domain skills; skills need stability; and all of them need governed facts and accountable human judgement.
Office automation reveals AI’s potential—and the business context it still lacks.
Generation without context produces elegant ambiguity.
Context without skills remains passive metadata.
Skills without stability create intelligence users cannot master.
Facts without narrative explain what happened, not why.
Human judgement is the accountability layer—not the fallback.
Office AI is essential.
It is not sufficient.
Copilots are a decisive milestone. They make the tools people already use faster, more fluent and more accessible—but productivity inside familiar work is not yet business transformation.
Writing, summarising, searching, meeting preparation and content creation improve immediately. The value is real. Yet these capabilities primarily optimise activities that already exist inside the office environment.
Visible milestone
Office automation
AI accelerates individual work inside tools whose documents already carry part of their own narrative.
Strategic destination
Business transformation
AI participates in how the organisation interprets operations, challenges assumptions and prepares consequential decisions.
Office AI changes how people perform existing work.
Analytical AI changes how the organisation understands and decides.
Deep read · why the first milestone is not the destination
Office copilots deserve to be recognised as an important adoption threshold. They familiarise organisations with conversational interaction, reduce friction and create visible productivity gains at individual scale.
But automation has always been one of the purposes of information systems. If AI only accelerates established tasks, it improves the existing operating model without fundamentally changing how the organisation understands its business.
Transformation begins when intelligence can work with governed operational facts, reconstruct business meaning and participate in the preparation of decisions—not merely in the preparation of documents.
AI can read the company.
Can it understand the business?
Documents describe the organisation. Structured data records its operations. Neither becomes reliable business understanding without semantics connecting facts, language, rules and intent.
Office AI sees emails, meetings, documents and presentations particularly well because they carry a narrative. Operational data is different: a number does not explain its definition, hierarchy, calendar, lineage, ownership or permitted interpretation.
Narrated surface
What the company says
EmailPresentationMeeting
DocumentTranscriptSpreadsheet narrative
The semantic threshold
Operational depth
How the business operates
TransactionsMeasuresHierarchies
CalendarsBusiness rulesLineage
Without governed semantics, AI can be fluent in the company’s language while remaining blind to its operating reality.
Deep read · structured data is not self-explanatory
Office tools can read tables and spreadsheets, but access is not the same as business grounding. Analytical reliability requires authoritative sources, certified measures, time logic, business entities, hierarchy rules, security boundaries and explicit ownership.
This is why enterprise AI cannot simply reproduce the consumer-chatbot model at greater scale. Consumer conversation often tolerates probabilistic context. Business decisions require shared definitions, provenance, repeatability and visible limits.
The blind spot is therefore not a lack of data. It is the absence of an operational layer of meaning through which structured facts and enterprise narratives can be interpreted together.
Begin with the decision.
Not the screen.
The No Dashboard Principle does not reject visualisation. It rejects the assumption that every future question must fit inside a representation designed in advance.
The recurring export from dashboard to spreadsheet is not merely an ergonomic failure. It is evidence that a predefined representation no longer matches the question the user is trying to answer.
Dashboards remain powerful instruments of observation. They are weaker instruments for reconstructing the reasons behind change—especially when the user does not yet know which question matters.
The representation is designed first.
- Predefined questions and navigation paths
- Charts, filters, alerts and exports accumulate
- Isolated signals dominate the field of view
- Users escape into spreadsheets to reframe
- Optimised for observing what changed
The question shapes the representation.
- Emerging questions and hypothesis formation
- Evidence assembled around business intent
- History, assumptions and constraints reconnected
- Interpretations compared and challenged
- Optimised for understanding why it changed
No fixed dashboard
The question chooses the evidence instead of inheriting a predefined field of view.
No fixed journey
The user’s intent determines which sequence, controls and representations are needed.
Interface on intent
The entire workspace is tailored to the decision at the moment it must be made.
Deep read · the higher-dimensional view
The traditional dashboard was born from a reasonable assumption: assemble the right indicators in one place and decision-makers will see more clearly. Over time, organisations multiplied visual surfaces until visibility itself became saturated.
Decision-making requires more than the derivative of events. It requires reconstructing the field from which those events emerge: history, assumptions, constraints, intentions and context. The future of analytics lies less in adding visual surfaces than in enabling data to be interpreted, questioned and reassembled into meaning.
Better visualisation libraries and more interactivity cannot solve an upstream limitation. A predefined dashboard offers only a limited set of paths when the user does not yet know the right question. Decision intelligence must help formulate that question, test hypotheses, compare interpretations and identify which evidence deserves attention.
The principle therefore expands beyond business intelligence. A dashboard is only one form of fixed interface. Once intent can shape the analytical surface, it can also shape the workflow, controls and temporary application assembled around a decision.
The interface becomes part of the answer.
A chart, a table, a narrative, a simulation, a recommendation or a temporary workspace can be generated when the need arises.
Software traditionally required users to learn stable screens, menus, workflows and vocabularies. Generative AI reverses the relationship: if intelligence understands intent and can orchestrate information and tools, the experience no longer needs to be fully designed in advance.
Then · screen-first
“There is an app for that.”
Next · intent-first
“There is an intelligence for that.”
Explain the margin deterioration and show me the evidence that matters.
narrativevariance treescenariotablerecommendationtemporary workspace
Schema-on-write
Structure is imposed before usage.
Interface-on-intent
Representation is assembled when meaning is requested.
The principle completed
AI is the new UI.
The interface is no longer the permanent container of intelligence. It becomes a temporary expression of intent, context and the decision in front of us.
Deep read · from predefined screens to generated experiences
This evolution mirrors the move from schema-on-write to schema-on-read. Modern data platforms separated storage from predefined structure; generative interfaces separate user intent from predefined screens.
The future analytical workspace may begin with an intent: “prepare the sales review,” “compare scenarios,” “identify risks in the forecast,” or “summarise what changed since the last committee.” The system then assembles the relevant data, documents, calculations, explanations and representations required to support that intent.
This does not eliminate deliberate design. It moves design from composing every future screen to defining the grammar through which appropriate screens can emerge.
Adaptive enough to help.
Stable enough to trust.
Traditional interfaces are limited but familiar. Generated experiences are powerful because they adapt—but can remove the cognitive anchors people need to learn and master a system.
Explore the paradox
92
Explainable adaptation
The experience changes, while its reasoning patterns and visual grammar remain recognisable.
Design systems become grammars
01Patterns of interaction
02Representation principles
03Domain-specific constraints
04Business vocabulary
05Conditions for trustworthy generation
Deep read · trust through explainable repetition
Stability will not come from returning to static dashboards. It will come from rules, constraints, skills and invariants that govern what a generated interface is allowed to become.
Users will not trust an analytical agent merely because it is fluent. They will trust it when they recognise its reasoning patterns, understand the origin of its answers, see how it handles uncertainty and observe that similar questions produce consistent forms of explanation.
Trust therefore emerges from explainable repetition within adaptive experiences: not sameness of screen, but consistency of reasoning, evidence and interaction.
From a universal assistant to specific intelligences.
The deeper transformation begins when AI understands the concepts, metrics, hierarchies, rules, exceptions, processes and decisions that define an organisation’s operating reality.
Situated intelligence crosses the threshold between fluent assistance and business understanding. It is embedded in domains, constrained by governance and shaped by the responsibilities of the people who use it.
A skill is not an improved prompt.
MethodsHow a decision is prepared and challenged.
ConventionsWhich comparisons and representations are valid.
ConstraintsWhat the intelligence may infer or execute.
ExamplesRecognisable patterns and accepted precedents.
Business objectsThe entities, metrics and relationships that matter.
GovernanceOwnership, official definitions and action boundaries.
Planning
Close budgets, compare scenarios, formalise assumptions.
Merchandising
Review assortment, interpret performance, prepare allocation.
Finance
Analyse variance, reconcile explanations, challenge forecasts.
Supply chain
Investigate anomalies, identify constraints, evaluate action.
High-value recurring decision patterns
close a budget
review assortment performance
prepare an allocation
investigate an anomaly
challenge a forecast
reconcile contradictory explanations
Deep read · formalise business skills first
A planning, merchandising, finance or supply-chain intelligence should not merely answer in natural language. It should know which objects matter, which comparisons are valid, which metrics are official, which representation is appropriate and which action is possible.
This reframes AI adoption. The strategic question is not only which model to use, but which business skills to formalise first. High-value candidates are recurring decision patterns where method, context and accountability already exist—though often tacitly.
The durable asset is not a conversation with a generic model. It is a governed, testable and reusable expression of how the organisation makes sense of a domain.
Metadata becomes operational context.
A missing definition now produces a weak answer immediately. An implicit business rule becomes dangerous once an agent can act upon it.
Documents carry context—but it may be ambiguous, contradictory, outdated or disconnected from authoritative definitions. Analytical AI must know what revenue, margin, inventory, budget, forecast or performance mean here, in this organisation, for this decision.
Context as found
Rich, dispersed, unreliable.
Presentationnarrative
Spreadsheetassumptions
Email threaddecision trace
Meeting transcriptdebate
Expert memoryexceptions
Context as engineered
Explicit, governed, testable.
official metricsauthoritative sourcescalculation rules
business calendarsownershipquality constraints
security boundarieshierarchiesexceptions
assumption registerslineagedecision rights
A metric definition is valuable when it…improves answers.
A business glossary is valuable when it…reduces ambiguity.
A lineage model is valuable when it…clarifies reliability and limits.
Deep read · context engineering as a product discipline
Context engineering is the deliberate construction of the semantic, procedural, organisational and behavioural context that allows AI systems to reason reliably.
It should be managed as a product, not as a documentation exercise. That means prioritisation, ownership, lifecycle management, testing and continuous improvement. Definitions are not complete because they have been written; they are complete enough when they improve the quality and consistency of reasoning.
AI makes the cost of weak metadata visible and immediate. Ambiguous hierarchies create observable reasoning errors. Missing ownership blurs accountability. Unrecorded exceptions turn plausible language into operational risk.
Connect what the company measures with what it means.
The most powerful enterprise AI will not choose between structured data and documents. It will connect them through governed business semantics.
A figure in a presentation can be linked to an official metric. A sentence in a transcript can be interpreted as an assumption. A spreadsheet can be recognised as a local deviation from an enterprise definition.
Structured truth
Measured facts
- Governed metrics
- Business entities
- Hierarchies and calendars
- Rules and quality controls
- Authoritative sources
Decision intelligence
Enterprise knowledge
Situated meaning
- Presentations and narratives
- Assumptions and scenarios
- Meeting notes and decisions
- Customer feedback
- Internal correspondence
01Measures
02Says
03Assumes
04Decides
Performance × explanation
Compare sales results with the explanations offered during the commercial review.
Variance × assumption
Relate a budget variance to spreadsheet assumptions and decisions captured in meeting notes.
Indicator × signal
Combine operational metrics with customer feedback, internal correspondence and strategic documents.
Deep read · a fused analysis of enterprise reality
The result is neither a summary of documents nor a query over databases. It is a fused analysis of what the company measures, what it says, what it assumes and what it decides.
This is where the boundaries between knowledge management, data governance and AI engineering begin to blur. Knowledge scattered across documents, conversations and expert memory cannot easily be mobilised by agents. Knowledge that is structured, governed, connected and continuously refined becomes an active component of the enterprise intelligence system.
Semantics provide the framework through which unstructured content becomes intelligible—and through which apparently precise numbers recover the narratives and assumptions that make them meaningful.
11 · Target architecture
The intelligence beneath
the interface.
Decision intelligence emerges from four interdependent layers. The upper experience is only as trustworthy as the governed facts, semantics and domain methods beneath it.
The visible interface is the final layer—not the foundation.
Read all four strategic roles
Governed data foundation. Provides trusted data, authoritative sources, quality rules, lineage and access control. It ensures AI reasons on reliable business facts rather than fragmented extracts through certified datasets, data contracts, access policies, observability and quality monitoring.
Semantic and context layer. Defines metrics, entities, hierarchies, calendars, business rules, assumptions and ownership. It transforms metadata into operational context through business glossaries, metric catalogues, semantic models, rule repositories and assumption registers.
Skills and domain intelligence. Encapsulates business methods, decision rules, preferred representations and execution patterns. It turns generic AI into situated intelligence through planning, variance-analysis, forecast-challenge and decision-preparation skills.
Generative experience layer. Generates narratives, tables, charts, simulations, recommendations and temporary interfaces on demand. It creates intent-driven decision spaces through conversational analytics, generated workspaces, scenario views, executive summaries and explainable recommendations.
Organisational implications
Technology changes.
Responsibilities move with it.
The shift to decision intelligence redraws the boundary between business, data, IT, design and governance.
Make decision models explicit.
Formalise questions, assumptions, rules, exceptions and accountability.
Deliver meaning, not only datasets.
Provide lineage, quality, semantics, context and reliable business definitions.
Orchestrate intelligences.
Move from delivering isolated tools to governing connected capabilities.
Define generative invariants.
Create the stable grammar through which adaptive experiences remain human.
Reconcile speed and trust.
Keep autonomy, security, consistency and human accountability compatible.
The competitive advantage
Not deploying AI fastest—but engineering the most reliable business context.
The winning organisation transforms dispersed knowledge into governed intelligence and shortens the feedback loop between definition, usage, correction and improvement.
The decision atelier · interactive model
Begin with an intent.
Assemble the decision space.
Choose a recurring business intent to see how governed facts, context, domain skills and an adaptive experience work as one system.
Business intent
Governed facts
Context
Domain skill
Generated experience
Explain the margin deterioration.
Margin bridge
Certified revenue, cost and mix metrics by period, market and product hierarchy.
Definitions & events
Official margin rules, calendar, pricing events, ownership and known exceptions.
Variance analysis
Decompose price, volume, mix and cost effects; test materiality and competing explanations.
Executive narrative
A causal story, evidence tree, uncertainty flags and scenario-ready recommendations.
Human role: choose which trade-offs are acceptable and own the decision.
Narrative · bridge · evidence
12 · From vision to roadmap
Progressive by design.
Anchored in real decisions.
Do not launch an abstract transformation programme. Start where trusted context, governed data and AI-enabled reasoning can create visible value quickly.
Short term
Map decisions
Focus
Identify high-value decision journeys and map the data, documents, definitions and recurring questions each requires.
Expected outcome
A prioritised portfolio of decision-intelligence use cases anchored in real business needs.
Medium term
Build context
Focus
Develop semantic foundations, domain skills and governed analytical agents for selected business domains.
Expected outcome
Reusable context assets and trusted agents supporting recurrent decision processes.
Long term
Scale intelligence
Focus
Scale generative experiences across domains while preserving governance, consistency, security and accountability.
Expected outcome
An enterprise intelligence fabric connecting data, documents, processes and decisions.
Guiding principles · house codes
Seven invariants for a generative future.
Start from decisions, not dashboards.
The unit of design is the business decision and the questions surrounding it—not the report format.
Make context explicit.
Metrics, assumptions, ownership, rules and exceptions must be usable by machines, not only readable by humans.
Preserve human judgement.
AI reconstructs possibilities; meaning remains attached to intention, perspective and responsibility.
Design stability within generation.
Generated experiences follow consistent rules so users can build trust and mastery.
Unify data and documents.
Connect measured facts with the narratives, assumptions and decisions that surround them.
Build domain skills.
Sustainable value comes from situated intelligence embedded in business practices—not generic conversation.
Measure decision quality.
Track better alignment, faster interpretation, fewer contradictory definitions and more explicit trade-offs—not usage alone.
Conclusion · beyond the interface
The interface was never the destination.The decision is.
The strategic measure
Not the number of assistants deployed or interfaces generated—but the quality of decisions organisations can make because their systems help them understand, simulate and act with greater confidence.
AI may be the new UI. But the UI is only the visible edge of a deeper capability: governed facts, operational meaning, situated methods and human accountability working as one.
Decision intelligence turns that capability into a continuous operating loop—observe, understand, simulate, act and learn.
The interface adapts to the moment. Context persists beyond it. Human judgement remains responsible for the choice.
Decision intelligence becomes the operating capability.
A business able to understand its context, simulate its choices, act with confidence and learn from every decision.