// What is Decision Engineering™
Decision Engineering™ examines how institutional purpose and policy become actual human and automated decisions, where that chain breaks, and how control can be rebuilt. It is a discipline of institutional decision integrity, still early in its development.
This page describes the institutional form of Decision Engineering™ developed by Deepak Aggarwal: how purpose and policy travel through strategy, intent, rules and judgment into the decisions an institution executes — and how institutions preserve control over that journey. The phrase “decision engineering” is also used in other fields; no claim is made over those uses.
A decision taken at the top rarely reaches execution in one step. It moves through policy, teams, data and technology, and then through whatever human or automated judgment remains at the end. The decision can change at every transition. What finally happens may comply with the rule in force and still be far removed from what the institution intended.
Institutions rarely lose control in one moment. A decision changes gradually as it moves through the institution. Every part can explain what it did — the board's direction was reasonable, the policy can be defended, technology built the approved specification, operations followed the process, risk reviewed the component it was asked to review — and yet the outcome is still different from what the institution intended.
That is what makes the problem hard to see. Governance may exist across the institution, yet the decision can still be lost between areas that are governed separately. Decision Engineering™ calls this gradual separation decision drift: the growing distance between what the institution decided and what its people and systems actually do. Automation does not create decision drift — it reduces the time available to see it.
The unit of analysis is the complete decision — not only the policy that authorised it, the model that processed it, or the person who acted. For any material decision, Decision Engineering™ asks five questions:
The discipline begins with diagnosis: make the journey visible first, then decide what has to change. If the break lies in an ungoverned translation, a new control or platform will only carry the same uncertainty faster.
// The map
The Decision Integrity Chain™ (DIC™) maps the eight layers a decision passes through inside an institution:
Purpose → Strategy → Intent → Rules → Judgment → Decision → Outcome → Feedback
It gives boards, executives, risk teams and engineers one shared map — to see where a decision sits, which layer is drifting and which join has broken. See the full Decision Integrity Chain™ explainer →
The Fiduciary Gap™ is the structural gap between what an autonomous system is optimised towards and what the institution is obliged to protect. The divergence emerges when optimisation logic is granted authority without incorporating those broader constraints. An agent optimises yield duration while the board mandate is liquidity resilience. A model is tuned for cost reduction and quietly concentrates model risk. Nothing has failed. The system is executing exactly what it was specified to execute.
The gap widens as authority moves. A tool enters the workflow as support, overrides become costly, and the default gradually becomes the decision. In human-led systems authority and accountability could usually be linked to identifiable roles; automated and agentic systems stretch that link, so the institution answering for a decision drifts further from the system actually making it. Giving someone an accountable title does not close the gap — accountability becomes real only when that person can produce the decision trail and intervene in the system they answer for.
Decision Engineering™ does not replace AI governance, risk management, decision science or decision intelligence. It looks across them, and asks whether institutional intent remained intact as it moved through every area.
| Discipline | Primary focus |
|---|---|
| AI governance | Design, use and oversight of AI systems. |
| Risk management | Exposures and controls. |
| Decision science | Using evidence and judgment to improve choices. |
| Decision intelligence | Typically uses data, analytics, models and decision methods to improve or automate choices and outcomes. |
| Model risk management | Validates that individual models are sound, correctly implemented and performing within approved limits. |
| Operational resilience | Sustains important business services through disruption and recovery. |
| Algorithmic accountability | Seeks transparency, responsibility and recourse for automated outcomes. |
| Decision Engineering™ | Whether institutional purpose, intent, authority and accountability remain intact as those choices travel through human and automated execution — and the joins between them. |
Each discipline is strong within its own area but less able to see what happens at its edges. That gap — where a decision crosses boundaries no single discipline holds as a complete object — is what Decision Engineering™ examines. Read the full comparison with decision management, decision intelligence and decision science →
A log shows that an action occurred. A decision record must reconstruct the authority, context and reasoning across the chain. That is the difference between explainability and replayability. If the chain can be replayed, control can be demonstrated. If it cannot, control rests on an assumption.
// Where it sits
Decision Engineering™ does not replace the frameworks institutions already run under. It begins where their responsibilities meet in operation.
Frameworks such as the NIST AI Risk Management Framework, the EU AI Act and the Monetary Authority of Singapore's FEAT principles organise responsibility around AI risk, transparency, oversight and accountability. Decision Engineering™ reconstructs the actual decision across models, rules, systems, human judgment and delegated authority — not only the AI component within it.
Model risk management asks whether a model is conceptually sound, correctly implemented and performing within approved limits. Decision Engineering™ asks a wider question: even if every model performed exactly as validated, did their combined outputs, the surrounding rules and the delegated actions produce the decision the institution intended?
Operational resilience regimes — MAS expectations, the Basel Committee's principles, the Hong Kong Monetary Authority's guidance — focus on sustaining critical services through disruption. Decision Engineering™ focuses on the integrity of the decisions occurring inside those services. A service can remain fully available while its decisions move quietly away from authorised intent.
Algorithmic accountability seeks transparency, responsibility and recourse for automated outcomes. Decision Engineering™ supplies the reconstruction method: trace the outcome back through the decision, judgment, rule, intent, strategy and purpose that authorised it — and forward through feedback and correction.
// Common questions
Decision Engineering™, in the institutional form practised here, is the practice of keeping an institution's intent intact as it travels into execution. A board sets a purpose. That purpose passes through strategy, rules, data, human judgment and automated systems before anything actually happens. At each handover it can change without anyone deciding to change it. Decision Engineering™ makes that journey visible, controllable and replayable — so an institution can show not only what it decided, but that what it executed was the same thing.
Decision science studies how a choice should be made — the reasoning, the probabilities, the biases that distort it. It largely stops at the moment of choice. Decision Engineering™ begins there. It assumes the choice may be sound and asks what happens next: whether authority is preserved as the decision is delegated, whether rules are applied as written, whether the executed outcome matches the intent. Decision science improves the decision. Decision Engineering™ protects it in transit.
Decision intelligence applies data, models and increasingly AI to improve the quality of decisions — better inputs, better predictions, better recommendations. It optimises the decision. Decision Engineering™ examines the chain the decision travels through afterwards: who held authority, which rule fired, what the model was told, what actually executed, and whether that sequence can be reconstructed. A better model inside a broken chain produces a better decision that still arrives at execution changed. The two are complementary, not competing.
The Decision Integrity Chain™ (DIC™) is an eight-layer model of how an institutional decision travels: Purpose, Strategy, Intent, Rules, Judgment, Decision, Outcome and Feedback. Each layer hands over to the next, and every handover is a join where intent can be lost, diluted or silently rewritten. The chain is a diagnostic instrument — it locates where a failure happened rather than declaring that one did. In case work it consistently identifies breaks at the joins rather than inside any single layer.
Explainability asks a model to describe its reasoning. Replayability asks something harder and more useful to an institution: can this decision be reconstructed after the event — the authority relied on, the data and rules in force at the time, the judgment applied, what was executed, and where any material divergence entered the chain? It does not require an identical model output to be reproduced. It requires enough evidence to establish whether the decision executed was the decision authorised. Explanations can be plausible and wrong. A replay is evidence. For boards and regulators, replayability is what turns an automated decision from something the institution hopes it understands into something it can actually account for.
Every term used across this work — replayability, the Decision Integrity Chain™, the Fiduciary Gap™, decision drift — is defined once in the glossary.