Skip to content
Enterprise Architecture — one company, five architectures, one decision system: ESA, ETA, ECA, EPA and ERRA on one enterprise state, unified as UETA and optimized through GEOP

Enterprise Architecture

One company, five descriptions.

DCT describes a company in five ways: its wiring, its moves, its capital, its performance and its exposure. Then it insists that the five describe one company. The architectures stand on one skeleton and read one state, a register records what passes between them, and the decision problem sits on top of all five.

GEOPVol I, Ch 15

GEOP and the Enterprise Optimization Architecture (GEOP)

What is the best plan the architectures allow?

What it is

The General Enterprise Optimization Problem is the decision the architectures exist to support (Vol I, Ch 15; Def 15.1). It has six declarations, each with an owner: the horizon; the decisions and the set each may take; the policy class, from a fixed plan to full feedback; the environment, given as a reference, a model of chance or a set; the objective; and the commitments the answer must keep.

Each architecture supplies part of it (Table 1.4; Table 15.1). The state architecture supplies the admissible states, the scopes of the commitments and the structure of the dynamics. The transformation architecture supplies the decisions and their domains. The capital architecture supplies the laws of the stocks, the capital floors and the prices of stocks. The performance architecture supplies the objective, the targets and the frontier. The risk architecture supplies the risk limits and the stress sets. The coupling register supplies the buffer identities and funding couplings, and the unified view supplies the feasible region and the stability certificate.

The Enterprise Optimization Architecture is the estate that hosts these problems, in four layers (Def 15.9). The model layer holds the unified declaration and its data; the formulation layer holds the problem instances and their audits; the solver layer holds methods with the certificates they can return (Table 15.4); the governance layer says who may formulate, solve, approve and implement. A number without a certificate does not travel. Machine learning enters only at three declared points, and never replaces the statement of the problem.

The cheapest package touches the 5% contour (seed 8, the book's paths)
Ch. I.15 The cheapest package touches the 5% contour (seed 8, the book's paths)

The board questions it answers

  • What is the cheapest plan that keeps every commitment, and what does each commitment cost?
  • Is the plan feasible before it is called optimal, and can it be held?
  • What does running the company as silos cost?
  • Who formulated this number, who solved it, who approved it, and what does its certificate say?
  • Can the number be reproduced?
  • Where did machine learning touch it?

What the book proves

  • Prop 15.4 Feasibility precedes optimality: the audit runs in a fixed order, consistency, then viability, then optimization.
  • Thm 15.3 At the optimum the plan is held only by what binds, and each binding commitment carries a price; a slack commitment's price is zero.
  • Thm 15.5 With several objectives, best means Pareto-best, and the frontier's slope is the exchange rate between them.
  • Thm 15.6 Where the register is silent, decentralization is exact; where it is not, prices on the coupling commitments coordinate the offices; dividing shared resources by fixed shares never does better; and appending the stability certificate gives the holdable optimum.
  • Prop 15.7 A certainty-equivalent, a robust and a stochastic formulation make different promises, and the board should know which one it is buying.

Meridian, the book’s worked company

Vol I solves Meridian's funding decision as a GEOP in Worked Example 15.1.

  • The decisions are a standby line, owned by the CFO, and a delay to the build, owned by the transformation office; the objective is the line's fees plus the cost of delay; the commitment is a chance of at most 5% of breaching the $350m floor in any quarter. (Vol I, WE 15.1)
  • The certainty-equivalent formulation buys nothing, because the mean path's trough is $460m, while the true chance of breach is 24.6%. (Vol I, WE 15.1; Prop 15.7)
  • At 5% the optimum is a $170m line and a delay of 0.668 quarters, costing $10.44m. The appetite's price is $1.81m per point of breach probability, an implied value of about $180m on a breach. (Vol I, OA 15.B.Ex.4)
  • Silos: the line alone cannot get below 8.4% at any size, and phasing alone costs $40.76m; coordinating the two saves $30.3m. The earlier package of line and phasing costs $19.0m at 3.0%, against $14.7m on the frontier. (Vol I, OA 15.B.Ex.4)
  • The book gives no EOA table for Meridian; the example fills its four layers implicitly: the model (the Ch 8 paths), the formulation with its owners, the solver (the exact quantile on 100,000 paths) and the governance (the board owns the appetite). (Vol I, WE 15.1)

Explore it

What each architecture supplies

Each architecture feeding the parts of the decision problem (Vol I, Table 1.4; Table 15.1), and the four layers of the Enterprise Optimization Architecture that host it (Def 15.9). Select a node to read what it supplies and what the book proves about it.

ArchitecturessupplyDecision problemDef 15.1GEOPEOADef 15.9AS​ · StateAT​ · TransformationAC​ · CapitalAP​ · PerformanceAR​ · RiskK · CouplingregisterUE​ · UnifiedDynamicsDecisions · policy classEnvironmentObjectiveHorizonCommitmentsFeasible regionGEOPthe decisionModel layerFormulation layerSolver layerGovernance layer

KeySuppliesHosts

Architectures

The decision problem

The optimization architecture

How it goes wrong

  • The problem is optimized before it is checked for feasibility (Prop 15.4).
  • A paper optimum is adopted that the company cannot hold (Thm 15.6(iv)).
  • One readout silently plays two roles, objective and commitment (Table 15.3).
  • Offices optimize separately and pay for the coordination they skipped (Thm 15.6(iii)).
  • An uncertified number travels into a board paper (Def 15.9).
  • A learned model is allowed to stand in for the problem statement (Def 15.9).

What a company must declare

  • The six declarations of Def 15.1, each with its owner.
  • Every commitment by class (Table 15.2): its owner, level and whether it binds.
  • The certificate each result carries: method, grade, tolerance, audit status and the seeds that reproduce it.
  • Governance: who may formulate, solve, approve and implement, and above what spending threshold approval is needed.

How AXIOM uses it

AXIOM computes the appetite package of WE 15.1, with its frontier and the price of each binding commitment, checks Optimization Readiness R1–R4 before it optimizes, and records each result's method, certificate and approval. The full nine-class commitment table and the EOA view belong to AXIOM's architecture module (in development).

Best Practices

Practices from the Corporate Transformation Best Practices knowledgebase that bear on this architecture.

Cases

Further reading

Read it on the DCT MapIn the LexiconRobust optimization, in the LexiconRead the chapter: Vol I, Ch 15

Where to go next