One layer for human and artificial decision makers
Agents are registered as actors with an owner, a trust level and a discretionary range — not governed as software events.
Decision Intelligence · Governance for people, agents and applications
Govern every decision, whoever — or whatever — made it
A common decision intelligence and governance platform for people, AI agents and applications. Enterprise Decision Intelligence captures decisions wherever they are made, holds them in one register with the reasoning attached, applies your own policy before they take effect, and measures the result.
The business problem
Enterprises record their transactions in detail and their decisions almost not at all.
What was decided, by whom, on what basis, with what expected outcome, and whether it worked — none of it is held anywhere you can query. When a decision is questioned months later, the organisation reconstructs it from memory and from whoever is still employed.
Deferred decisions are worse. They exist in no system at all, so a choice can sit unmade for eleven weeks and nobody can see it.
And AI agents now make decisions with commercial and regulatory consequence while being governed as software, through change control, rather than as decision makers, through policy and accountability.
The platform
Five pillars, one register, one policy engine
Positioning
What makes it different
Records reasoning, not transactions
Rationale, alternatives, expected outcome and the exact policy version in force, held with the decision.
Verbal capture
The largest uncaptured category in any business. A decision stated on a call creates a provisional record and a prompt to its owner while the reasoning is still recoverable.
Discretion as a dial, not a switch
Policy sets the rule; a per-person trust level sets how far each actor may move inside it.
Override telemetry
A rule overridden two times in five is miscalibrated, not a misbehaving team. Governance spend turns into operating improvement.
01
Capture layer — The decisions you never knew were made
The routes by which a decision enters the platform: meetings and calls, a managed email inbox, connected business applications, a custom connector for anything with an API, the portal and mobile app, and an API for agents. Every route normalises into the same record.
The problem. Most consequential decisions leave behind a transaction, a thread or nothing at all. A commitment made on a call is gone when the call ends. An approval given by email sits in one person's mailbox. A discount applied in CRM records the amount, not the judgement.
| Functionality | What problem it solves | Impact on the business |
|---|---|---|
| Meeting and call captureTranscripts analysed for decision language; the owner is prompted to confirm and add reasoning. | Decisions agreed on a call leave no record, and the reasoning is gone by morning. | The largest uncaptured category of decisions reaches the register the same day. |
| Decision inboxA managed mailbox. Forward a thread and the decision is extracted, with the thread kept as evidence. | Approvals sit in individual mailboxes where nobody else can find or audit them. | Coverage in days, with no software to deploy and no behaviour to change. |
| Application connectorsSAP, Salesforce, Dynamics, Oracle Fusion and Sage, starting in read-and-advise mode. | Systems of record capture the amount but never the judgement behind it. | High-volume operational decisions become measurable without touching a live workflow. |
| Custom connectorA connector framework and SDK for any application that exposes an API. | The decisions that matter often sit in in-house systems nobody builds connectors for. | No system is out of scope, and integration does not wait for a product release. |
| Portal, mobile and agent APIStructured entry for people, and the same interface for agents and automated processes. | Agents decide with real consequence but are governed as software rather than as decision makers. | Agents are captured on the same footing as people. |
Business value
- Every channel and system lands in one register
- Rationale captured while it still exists, not at audit
- People, applications and agents captured the same way
How it connects
Capture feeds the register, which everything else reads from. Policy cannot evaluate a decision it never saw, so coverage constrains every other pillar.
02
Central decision register — One record of how the business actually decides
An append-only, versioned store holding what was decided, by whom, on what evidence, under which policy version, with what expected outcome. It holds decisions that were deferred as well as those that were taken.
The problem. When a decision is questioned by a regulator, a board, a customer or a court, the organisation reconstructs it from memory. Deferred decisions are invisible: they exist in no system, so stalled choices age silently.
| Functionality | What problem it solves | Impact on the business |
|---|---|---|
| One canonical modelEvery channel maps to the same record: owner, value, risk, rationale, policy version, outcome. | Decisions arrive in five formats and cannot be compared or reported on together. | One query answers what currently takes five separate investigations. |
| Append-only and attributedVersioned entries, never destructively updated. Every change carries an actor, a time and a reason. | A record that can be edited after the fact is worth very little to an auditor. | Evidence stands up to scrutiny, because the history cannot be quietly tidied. |
| Tabled decisions are recordedA deferral requires a review date and is aged against it, escalating when it passes. | Deferred decisions exist in no system, so a choice can sit unmade for months unnoticed. | Stalled decisions become visible and chaseable for the first time. |
| Scoped access and data sovereigntyRelationship-based access and tenant isolation, deployed inside the boundary you choose. | A register of sensitive commercial and personnel decisions is itself a risk if wrongly exposed. | Sensitive decisions can be held centrally without creating a jurisdictional problem. |
Business value
- Decision trails produced by query, not investigation
- Reasoning survives staff turnover
- Stalled decisions become measurable
How it connects
Capture writes to it, policy replays it to simulate rule changes, intervention records approvals into it, and learning measures outcomes from it.
03
Policy and trust control plane — Set the guardrail once, then widen it as it is earned
Directors author policy in business terms — decision type, value band, risk class, condition. Managers set a trust level for each person under their command. The two combine into an effective discretionary range, recalculated whenever either changes.
The problem. Delegation today is implicit, inconsistent and undocumented. Binary permission systems make it worse: they block sensible actions, so people find workarounds, and a bypassed control provides no assurance at all.
| Functionality | What problem it solves | Impact on the business |
|---|---|---|
| Policy in business termsDecision types, value bands, risk classes and conditions, bound to roles, teams or named people. | Delegation is implicit and undocumented, so nobody can say what anyone may decide alone. | The rule is written and owned by the people accountable for it. |
| The trust dialFour levels per person, set by their leader. Adjustments are recommended from evidence, never applied automatically. | Leaders have no instrument for extending more latitude to the people who have earned it. | Autonomy becomes a managed variable that can be widened and reversed. |
| Accountable overrideProceed outside policy with a mandatory justification; the manager and director are notified. | Rigid controls get bypassed, and a bypassed control provides no assurance at all. | The rule can be wrong without the exception being invisible. |
| Simulation before activationReplay a proposed policy against historical decisions before it takes effect. | Thresholds set by guesswork either flood approvers or catch nothing at all. | Alert volume is known before anyone is affected. |
Business value
- Fewer routine approvals reaching manager queues
- Delegation written down, versioned and reviewable
- Stepping outside policy is permitted, attributed and notified
How it connects
It evaluates what capture delivers, replays register history to test its own changes, and hands anything outside range to intervention.
04
Intervention and escalation — Stop the wrong decision without stopping the business
When a decision falls outside an actor's range, this pillar decides what happens next: hold it, flag it, or reject it outright where policy is absolute. Held decisions route to an available, authorised approver with the full context attached.
The problem. Approval workflows fail in two directions. They stall, because the right approver is on a plane. Or they degrade, because approvers receive a request with no context and approve it reflexively. Both teach the organisation to route around the control.
| Functionality | What problem it solves | Impact on the business |
|---|---|---|
| Context travels with the requestDecision, value, breach reason, the requester's reasoning and comparable past decisions. | Approvers receive a request with no context and approve it reflexively. | Approval quality holds up even when the decision is taken from a phone. |
| Return for more contextA first-class outcome alongside approve and reject. The request resumes with the same approver. | An approver with too little information can only guess, or start an argument. | Fewer decisions approved blind, and fewer reversed later. |
| Presence-aware routingRequests go to someone available at the right level, sideways rather than upwards where that is faster. | The right approver is on a plane, and the business waits days for ninety seconds of work. | Decision latency stops being a function of one person's diary. |
| Cover without driftDelegating downwards can temporarily reduce the deputy's range. The reduction expires automatically. | Holiday cover quietly widens authority and surfaces later as an audit finding. | Cover is bounded and time-limited by design rather than by trust. |
| Approval on any channelMobile, Teams, email, SMS and WhatsApp, with step-up authentication above a value threshold. | Controls that require a desk get routed around by the people they govern. | Senior approvers stay effective while travelling, so the control is actually used. |
Business value
- Commitments stopped before they bind
- Approvals reach an available human
- A complete chain of approvers, timings, returns and overrides
How it connects
Policy decides; intervention enacts. Outcomes return to the register as the override telemetry learning reports on.
Preventive interception requires a pre-commit extension point in the host application. Where none exists, the intervention is detective and is described as such.
05
Analysis and learning — Find out whether your decisions were any good
Three jobs: structuring the reasoning before a significant decision, watching for decisions that look wrong at the time, and measuring what happened afterwards. What it learns feeds policy calibration, trust recommendations and the three dashboards.
The problem. Organisations spend heavily on decision-making capability and measure none of it. Nobody knows which decision types reliably go wrong, whose judgement holds up, or which policies are so miscalibrated that everyone overrides them.
| Functionality | What problem it solves | Impact on the business |
|---|---|---|
| Cartesian analysisFour mandatory questions above a threshold the director sets, with every answer scored. | People justify a decision by describing the upside of acting, and nothing else. | The expensive quadrants get examined before money is committed. |
| Two failure modes, handled differentlyIrrelevant answers return to their writer; implausible ones go to a human reviewer. | Mandatory forms invite box-ticking, which poisons the data everything else depends on. | Analysis quality is protected without a model passing judgement on anyone. |
| Anomaly detection in contextEach decision is compared with the actor's history, their peer group and similar outcomes. | Transaction monitoring sees the amount but not who authorised it, or under what rule. | Threshold-avoidance and insider patterns surface at the point of decision. |
| Outcome trackingExpected outcome and review date recorded up front; actuals gathered automatically where possible. | Nobody finds out whether a decision worked, so nothing improves. | Decision quality becomes measurable at person, team and organisation level. |
| Three dashboardsDirector, manager and individual views, each showing only what that person can act on. | Reporting that points only upwards is experienced as surveillance and gets resisted. | People engage with the platform because it shows them their own record. |
Business value
- Outcomes tracked against the stated expectation
- Coaching aimed at decision types that actually fail
- Override telemetry becomes operating improvement
How it connects
It reads the register, tunes policy and recommends trust adjustments. It is what makes the platform improve with use rather than ossify.
What it looks like
The screens, drawn to the current specification
These are drawn from the specification rather than captured from the prototype, and each is labelled as such in its own frame. The prototype's trust dial and rationale panel are built to models this version of the specification replaces, so showing them would show you the wrong product.
Recording the decision
Decision registerConceptual interface
Class: allLast 30 daysValue ≥ AED 50,000Overridden only
| Ref | Class | Decided by | Band | State | Value |
|---|---|---|---|---|---|
| D-48812 | Pricing exception | G. Okonkwo | Delegated | Decided | AED 180,000 |
| D-48809 | Supplier award | Procurement agent | Supervised | Held — awaiting approver | AED 1.2m |
| D-48801 | Credit limit | M. Haddad | Delegated | Decided | AED 95,000 |
| D-48794 | Capital approval | Board | No discretion | Deferred — review 3 Nov | AED 4.0m |
| D-48790 | Pricing exception | K. Al Suwaidi | Supervised | Decided — policy overridden | AED 310,000 |
Decision D-48790 · Pricing exceptionConceptual interface
Chosen
Approve at 22% discount, 12-month term
Rejected
Hold at 15% · Match incumbent at 25% · Do nothing
Reasoning, recorded at the time
“Three-year account, renewal at risk against an incumbent. 22% holds the margin floor and the term protects next year.”
Policy in force
PRC-04 v8 · ceiling 20% — breached, overridden with reason
Outcome, attached later
Renewed. Margin 1.4 points below expectation. Reviewed 90 days after signature.
Setting how much room people and agents have
Control plane — Pricing exceptionsConceptual interface
Grace Okonkwo · Regional sales director7
0No discretion1–2Recommend3–5Supervised6–8Delegated9–10Autonomous
Delegated — acts within policy, owner informed.
Projection against the last 600 decisions in this class
Human touch falls from 38 in 100 to 19 in 100. No decision above AED 250,000 changes band.
Nine months at Supervised, two reversals, both recoverable.
Commit changeWidening autonomy — a second approver activates this.
Agent registryConceptual interface
| Agent | Accountable owner | Scope | Autonomy | State | |
|---|---|---|---|---|---|
| Pricing assistant | G. Okonkwo | Pricing exceptions | Delegated (7) | Live | Stop |
| Supplier screener | Procurement ops | Supplier shortlisting | Supervised (4) | Live | Stop |
| Credit reviewer | Risk | Credit limits | Recommend (2) | Shadow | Stop |
| Contract drafter | Legal | Standard renewals | No discretion (0) | Blocked | Stop |
No agent reaches the top band on a consequential class. Autonomy is extended one class at a time, on shadow evidence, and withdrawn the same way.
Getting a decision to somebody who can settle it
My queueConceptual interface
Pricing exception · AED 180,000
Raised by Grace Okonkwo
4h2h target
Past targetSupplier award · AED 1.2m
Raised by Procurement agent
1d 3h3d target
Within targetCredit limit · AED 95,000
Raised by M. Haddad
6h8h target
ApproachingCapital approval · AED 4.0m
Raised by Board sponsor
2d5d target
Within targetEscalation · Credit limit D-48809Conceptual interface
Tier 2 of 4 · 40 minutes before the director is notified
Holder · G. Okonkwo
Notified 10:02 · no acknowledgement
Cover · M. Haddad
Notified 10:32 · acknowledgement pending
Class owner · Risk
Not yet notified
Accountable director
Not yet notified · then class default applies
Finding out whether it was any good
Decision quality · Pricing exceptions, last 12 monthsConceptual interface
14Bad luckSound method, poor result
118Deserved successSound method, good result
9Deserved failurePoor method, poor result
23Lucky escapePoor method, good result
Select a quadrant to list the decisions in it. Nothing here is scored or ranked — the shape is the finding.
Impact · Decision D-48790Conceptual interface
Account margin
−3.1 points
High confidence
2 other decisions in window
Renewal rate
+4 points
Medium confidence
5 other decisions in window
Support volume
+11%
Cannot attribute
11 other decisions in window
Where confidence is too low the product refuses the question: “eleven other decisions touched this account in the window; movement here cannot be attributed to this decision.”
StallsConceptual interface
Signals with no decision
7
Raised, never classified. Oldest 11 days.
Owner: class owners, unassigned
Ageing past target
14
Credit queue. Oldest 6 days against a 2-day target.
Owner: Elena Duarte
Windows that closed
2
Renewal dates passed while the decision was open.
Owner: K. Al Suwaidi
Cartesian decision making
Four questions that surface what a decision really costs
Cartesian decision making examines a proposed decision from four directions instead of one. Asked to justify a decision, people describe the upside of acting — the one view least likely to contain anything they had not already thought of.
Evidence
Four metrics, measured from your own register
Enterprise Decision Intelligence has no published customer results. These are the measurements a first engagement produces, each taken from your own decision register against a baseline back-filled at onboarding.
| Measure | How it is produced |
|---|---|
| Decision coverage | Proportion of in-scope decisions captured during the pilot, measured against a back-filled baseline |
| Decision velocity | Reduction in the average age of open decisions, available from the register from week one |
| Value intercepted | Aggregate value of decisions held, rejected or amended at approval |
| Policy calibration | Fall in approval volume between go-live and a tuned steady state |
No values are quoted here because none have been earned yet. We publish no benchmark we cannot reproduce from a live engagement.
Read further
19 documents on this product
Every one is a working draft, and says so. No form, no email address — the link downloads the PDF.
- White paper: Cartesian Decision MakingEDI-WP-01 · v2.0
- White paper: The Cost of Organisational IndecisionEDI-WP-02 · v2.0
- White paper: Trust and VerifyEDI-WP-03 · v2.0
- White paper: Delegation Without Losing GovernanceEDI-WP-04 · v2.0
- White paper: Building a Decision-Making CultureEDI-WP-05 · v2.0
- White paper: Decision Impact AnalysisEDI-WP-06 · v2.0
- White paper: The Decision-Aware EnterpriseEDI-WP-07 · v2.0
- White paper: Decision Making AnywhereEDI-WP-08 · v2.0
- White paper: When Micromanagement Is the Right ManagementEDI-WP-09 · v2.0
- White paper: The Customer JourneyEDI-WP-21 · v2.0
- Data sheet: Suite overviewEDI-DS-00 · v2.0
- Data sheet: CaptureEDI-DS-01 · v2.0
- Data sheet: Record and evidenceEDI-DS-02 · v2.0
- Data sheet: Decision support and contextEDI-DS-03 · v2.0
- Data sheet: Delegation, policy and agent authorityEDI-DS-04 · v2.0
- Data sheet: Routing, intervention and escalationEDI-DS-05 · v2.0
- Data sheet: Reach: presence, delegation and notificationEDI-DS-06 · v2.0
- Data sheet: Outcomes, learning and stallsEDI-DS-07 · v2.0
- Data sheet: Platform and serviceEDI-DS-08 · v2.0
Next step
A 90-day fixed-scope pilot on one decision class
Fixed scope and fixed fee, on one connected system. It delivers a working enforcement control, a populated decision register, a policy calibration report and a baseline you can take to an audit committee or a board — before any commitment to breadth.
Baseline first
Historical decisions are back-filled at onboarding so the change is measured rather than argued.
Four metrics, from your own data
Every figure is produced from your register against that baseline. We publish no benchmark we cannot reproduce from a live engagement.
One decision class
Scope is deliberately narrow: one class, one connected system, plus meeting and email capture.
Read and advise first
Connectors observe, record and flag before anything is held. Interception is a later, scoped phase.