ALEX · Work · Workbench · Publications · Field Notes · About

Decision Support · Reference Architecture

A decision system begins where reporting stops.

A practical architecture for turning evidence and uncertainty into options, accountable commitments, action and learning.

The unit of design is not the dashboard, report or model. It is the decision cycle: what must be decided, what is known, what remains uncertain, who can commit, what changes afterwards, and what reality teaches us next.

Reference architecture

Follow one decision all the way through.

This is not a ten-step management ritual. It is a trace. Start with the trigger and follow it through evidence, uncertainty, alternatives, authority, commitment, action and outcome. Governance, information integrity and traceability run across the whole path.

1. What needs deciding?

Objective, horizon, constraints, trigger and decision owner.

2. What do we know?

Facts, observations, source quality, provenance and relevance.

3. What might be wrong?

Unknowns, assumptions, confidence, ranges and reversal conditions.

4. What happens next?

Forecasts, scenarios and models appropriate to the decision.

5. What can we actually do?

Real alternatives, including defer, stage, experiment or stop.

6. What gives?

Value, exposure, timing, capacity, dependencies and reversibility.

7. Who can commit?

Who recommends, challenges, approves, escalates and can reopen the choice.

8. What are we committing to?

The accountable choice, resources, conditions and review point.

9. What changes on Monday?

Workflow, ownership, change control, due dates and follow-through.

10. What did reality say?

Outcome evidence, variance, unintended effects and model revision.

Design rule: a dashboard can reveal a signal. It becomes decision support only when that signal is connected to uncertainty, alternatives, authority and a consequential choice. If the trail disappears somewhere between the dashboard and the meeting, the architecture has found its first job.

Information hierarchy

More data is not the same as more decision support.

Reporting technology often outruns management logic. These distinctions are worth preserving.

Data

Recorded events, values and observations. Can we trust what was recorded, by whom, when and against which definition?

Information

Data placed in context so change, condition or relationship becomes visible. What does this tell us about the state of the system?

Management information

Information organized around a management responsibility, threshold or review question. What deserves attention, challenge, escalation or follow-up?

Decision support

Evidence plus uncertainty, alternatives and consequences structured around a decision. What can we do, and what could make the recommendation wrong?

Commitment

An accountable choice with authority, resources, conditions, action and a review point. Who committed to what, under which assumptions, and when will reality be checked?

When dynamics matter

Some decisions cannot be understood from a static dashboard.

System Dynamics earns its place when behavior is created by feedback, delays, accumulations or nonlinear responses. The point is not a more impressive forecast. It is to expose the structure behind behavior and test how policy may change it over time.

  • Feedback: actions alter the conditions shaping later actions.
  • Delays: effects arrive after decision makers have reacted again.
  • Accumulations: backlogs, capability, inventory, trust, maintenance debt or exposure build and deplete.
  • Nonlinearity: thresholds and saturation mean twice the input does not mean twice the result.
  • Policy test: which decision rule remains robust when the system responds to us?

When justified, causal-loop maps, stock-and-flow models, simulation, sensitivity testing and tools such as Vensim can support this layer.

Governance and decision rights

Analysis can inform a decision. It cannot own one.

Roles may be combined in a small organization, but the responsibilities should remain distinguishable.

  • Decision owner: owns the choice and the conditions under which it will be revisited.
  • Decision analyst: structures evidence, uncertainty, alternatives and consequences.
  • Risk / control owner: tests exposure, controls and conditions that could change the choice.
  • Data owner: protects definitions, provenance, quality and refresh logic.
  • Action owner: converts commitment into controlled execution and evidence of completion.
  • Review body: challenges, escalates and checks whether the decision remains justified as evidence changes.

AI-assisted layer: AI can summarize evidence, surface anomalies, compare scenarios, draft decision briefs and prepare reviews. It can compress the work. It should not blur accountability. Authority to commit remains human and governed.

Generalized demonstration · Synthetic

A portfolio recovery decision.

This example is synthetic and generalized. A portfolio dashboard looks acceptable, but forward-looking evidence suggests that current commitments cannot all be kept.

The decision

Preserve the current portfolio sequence, or reallocate capacity before forecast pressure becomes delivery failure?

Evidence and uncertainty

Current milestones, CPI/SPI, forecast completion dates, resource loading, dependencies, benefits and open risk exposure are visible. Supplier lead time, scope stability, specialist availability, approval timing and remaining-effort confidence could still reverse the conclusion.

Alternatives and trade-offs

Resequence, add capacity, defer work, or protect one strategic commitment at the expense of another. The exchange is between near-term delivery, strategic benefit, risk exposure, cash timing, management attention and reversibility.

Decision record

Choice: protect the two highest-consequence commitments and defer lower-value work competing for the same specialist capacity.

Why: the forward forecast shows a capacity collision that current status reporting does not. The decision accepts a visible delay in one area to avoid hidden failure across several.

Conditions: reopen the decision if supplier timing or specialist availability moves beyond the agreed tolerance.

Next evidence: reallocate named capacity, baseline revised dates and review forecast error at the next portfolio checkpoint.

Practical diagnostic

If these answers are fuzzy, the system is too.

  1. What specific decision is this information supposed to change?
  2. Which uncertainty or assumption could reverse the recommendation?
  3. What alternatives are genuinely available, including delay, experiment, stage or stop?
  4. Who has authority to commit resources or change the operating course?
  5. What is recorded when the decision is made: rationale, assumptions, conditions and action owners?
  6. When will outcome evidence be compared with what the decision assumed?

What informs this architecture

This is an ALEX synthesis rather than a reproduction of any one standard. Established disciplines provide boundary conditions and tested concepts; the architecture reorganizes them around operational decision flow, accountability and learning.

ALEX · Make complexity workable.