Skip to content

Ontology

An ontology writes how your company decides into the system

The factory counts stock in quantities; accounting records it in money. It is the same material, yet inside the two systems the two are strangers. An ontology is the work of tying them together as one and the same, so that when someone asks why material costs rose, the system can answer.

An ontology is not an IT department’s project. It is management putting “this is how we decide” on record in the system.

Why the perfect system went unused

Plenty of companies build a system that captures every item, order and sale — and still stall at “why is revenue falling?”. The design started from data fields, not from the questions of the floor.

Start from data fields

  • Items — all captured
  • Orders — all captured
  • Sales — all captured
  • Equipment history · shifts · weather — not in the design

“Why is revenue falling?”

The system stops here

Start from the floor’s questions

“Why do defects rise at month-end on this line?”

Line supervisor

“Why does yield drop in the rainy season?”

Quality manager
Only then does this enter the dataEquipment historyShift patternsWeather

The system starts pointing at the real problem

This is why a DARVIS build starts with interviews on the floor before it connects any data.

An ontology designed without the people who know the floor best is the first one to be abandoned.

Common misreadings, and the redefinitions

The common viewThe redefinition
A technology for integrating dataA structure that fixes how an organisation thinks
A technology for collecting moreA technology for deciding first
A document, a deliverableThe result of an agreement
A store of knowledgeA digital twin of reality
A data modelThe constitution for the AI agents to come

Don’t build the dam — connect the relationships first

Start with company-wide data integration and it usually stalls. The book’s recommended starting points are five relationships, and most companies begin at ①.

Customer → behaviour → revenue

Marketing gets judged on contribution to revenue rather than volume of activity. Sales reporting shifts from describing results to analysing how conversion works.

Event → cost

Accounting records the outcome but doesn’t explain the cause. Separate immediate from deferred cost, and direct from opportunity cost, and cost management moves from after-the-fact reporting to judgement made in advance.

Decision → assumption → outcome

The most important relationship, and the one that most often disappears. Store why a decision was made as an object, and the system becomes the organisation’s memory.

Risk → probability → loss

One department sees risk as likelihood, another as cost. Structure it as probability and impact and risk management becomes arithmetic rather than instinct.

Current state → trend → future outcome

Where time-series data meets the ontology — equipment condition and failure probability, stock level and the odds of missing a delivery date.

An intelligent company is not the one with the most data — it is the one that fixed its most important relationship first.

Which side is your company on — a 14-question self-check

Across seven areas, pick whichever sounds more like your company. The result appears right here, and we never ask for your contact details.

0 / 14
  1. 01Spotting a problem

  2. 02Spotting a problem

  3. 03Data and reporting

  4. 04Data and reporting

  5. 05Executive meetings

  6. 06Executive meetings

  7. 07Root-cause analysis

  8. 08Root-cause analysis

  9. 09AI and systems

  10. 10AI and systems

  11. 11Crisis response and organisational learning

  12. 12Crisis response and organisational learning

  13. 13How it feels as CEO

  14. 14How it feels as CEO

Five beliefs that keep you from starting

The fifth is the most dangerous — it is the one that delays starting the longest.

  • “It’s an IT project”It’s a question of how the organisation thinks
  • “We don’t have enough data”What’s missing is the connections
  • “The ROI isn’t clear”Removing risk is the ROI
  • “It’s too complex for the people doing the work”It is nothing more than structuring the words those people already use
  • “It can wait”The gap compounds

The book answers the ROI objection like this: “The return on an ontology is proved not by what you earn, but by what you don’t lose.” The first number to move is investigation time — finding one cause used to take weeks; once the relationships are fixed, it is a query of minutes.

So where do you start — three rules for doing it

01

Fix one question at a time

What is the question your company asks over and over, every day?

02

Structure the judgement before the data

The numbers can be attached later.

03

Put the thing you argue about in meetings into the system first

The noisiest topic is the most valuable relationship.

This is what a fixed relationship looks like on screen

Everything above is the concept; below is what the concept becomes inside a tool. Click an entity and its attributes, its relations and the rules that read it as a condition all change together. The rules moving in the same place is what it means for an ontology to run as a system rather than sit in a document.

DARVIS Ontology ExplorerFictional food manufacturer · month-end cost · CLOSE-SYN-08
Read-OnlyExample — fictional data
100%
ASSESSMENT COREEVIDENCE · PLAN · REVIEWCLASSIFICATION · SUPPORTrelatedToevidencesgeneratestriggerstestedByproposesquantifiestracesscoresPGProductGroupProduct family and SKU schemeCECostElementTypeMaterial, labour, overheadARAllocationRuleTypeAllocation rule typesACAccountTypeAccount schemeIKInformationKindObserved, inferred, AI, simulatedCLConfidenceLevelLOW · MED · HIGHSCSecurityLevelMarking and export controlPRPerformerRoleFDE and reviewersRCRiskCategoryWatch, potential, highCPMonthlyClosePlanAugust close · CLOSE-SYN-08CMCostModel24-step, 3-tier allocationAPAllocationPolicyFreight allocation policy v12AVApprovalRuleCorrection and price approvalMTMetricSpecOperating margin definitionGSGoldenSetGolden set · validation specAFamily A12 frozen-food SKUsERERP ledgerVouchers and accountsMEMES actualsProduction and input actualsCCCost centre CC-095Logistics and freightROReview bodyFDE and customer ownerCACarrier TSynthetic vendor07Analyst-07FDE reviewerLLLedger load2,344 rows · read-onlyALAllocation runRUN-084 · doneREReconciliationSG&A gap zeroANAnalysis runRUN-118 · runningRVExpert reviewFirst and second · pendingAECorrection approvalNot approvedMGMarginDropSignalFamily A −2.1%pLELossExposureExposure · being estimatedCVCounterEvidencePrior-year policy changesCSCauseCandidateFreight allocation 0→400FTCorrectionTargetCorrection target · unapprovedSMScopeMatchQuestion inside scopeRPRealizedProfitRealised profit · unmeasuredPVProvenanceRecordSource, transform, model lineageDTDecisionTraceFull decision traceSXSecurityMarkingSYNTHETIC · DEMOCFConfidenceAssessment0.87 · HIGHTITimeInterval08-01 ~ 08-21OSOrgScopeFood division · Seoul plant
41 nodes46 relations6 layersONTOLOGY 0.4ObservedDARVIS inferencePlan / ruleRisk / unapprovedSimulated

Click a node and only its direct relations stay lit — you can see how evidence, counter-evidence, rules and review hang off a single assertion.

Pinch to zoom, drag to move.

See real cases analysed this way →

An example model for a fictional manufacturer. In a real build, both the entities and the rules follow from the question your company asks every day.

Curation and validation

Sources are mapped to the core, and the loss is verified

Terms that ERP, MES and documents each call something different are brought onto one core. Every mapping keeps its method, loss rate and approval state, and nothing enters the model without passing the gates.

What to read first, and what to do next

That is the concept. What to prepare and in what order, when you apply it to a real company, is set out in the kickoff guide. Leave a few details and it opens right away.

DARVIS is that relationship structure applied to cost and margin questions on a factory floor.See how the ontology gets built automatically →