Enterprise AI economics

Measuring
Context Engineering
ROI.

Most enterprises evaluate AI one use case at a time, then start again. Context Engineering changes the maths: the context engineered for the first decision becomes reusable infrastructure for every decision after it. Value compounds while the cost of each new decision falls.

Walk the seven views Download the white paper
Cost per trusted decisionthe economic unit Four layerscontext, decision, execution, value Read in 9 minutesor watch in 2
The compounding ROI gapIllustrative
Marginal cost per decision Cumulative value DECISION DOMAINS ENABLED ON SHARED CONTEXT D1D2D3D4D5D6D7D8 Crossover Foundational investment Integrations, entities, policy Compounding ROI gap
Watch

The economic argument, on video.

Context Engineering changes the economics of enterprise AI: value compounds while the cost of each new decision declines.

Context Engineering ROI Open the video page
Seven views on the economics

The ROI of enterprise AI is not a use case. It is a curve.

Move through the seven views with the arrows, the keyboard, or from the contents list. Three of them are live: drag the controls and watch the numbers change.

01 The executive proposition

Context is not consumed. It is reused.

A model is deployed, a workflow is automated, a business case is calculated. Then the next use case starts again from zero.

The first decision domain carries a foundational investment: connecting data, resolving entities, establishing meaning, applying policy. That context then becomes infrastructure. Every later domain inherits most of it, so the marginal cost of the next decision falls while the value keeps stacking.

The question changes. Not "how much value did this use case create?" but "how much more value can the context we already engineered create?"
Figure 1 Build vs reuseIllustrative
Decision domains enabled on shared context4 domains
Context reused in latest domain80%
Marginal cost index (D1 = 100)28
Cumulative value index128
02 Why traditional AI ROI is incomplete

AI business cases measure the application. Context Engineering measures the decision.

Compute, productivity and accuracy still matter. But they miss the defining economic trait of Context Engineering: reuse. When every copilot, agent, workflow and model retrieves data, interprets policy and rebuilds relationships on its own, the enterprise pays for the same intelligence four times over, and gets four different answers.

Why it matters. Enterprises create value by making and executing decisions, not by generating responses.
Figure 2 Two economic modelsConceptual
Copilot
EntitiesPolicyRelationsEvidence
Agent
EntitiesPolicyRelationsEvidence
Workflow
EntitiesPolicyRelationsEvidence
Model
EntitiesPolicyRelationsEvidence
Context rebuilt 4 timesDuplicated cost, inconsistent meaning
Copilot
Agent
Workflow
Model
Shared enterprise context
EntitiesPolicyRelationsEvidence
Built once, reused 4 timesPersistent, consistent, governed
03 The ROI chain

Context itself is not the ROI. What context enables is the ROI.

A larger graph, more sources and more tokens are enabling capabilities. Value appears only when better context changes the quality, speed or cost of an enterprise decision, and then only when that change reaches a line the CFO can bank.

The principle. Do not measure the context layer in isolation. Measure the economic consequence of the decisions it changes.
Figure 3 From context to outcomeRead left to right
Stage 1

Context capability

  • Is information current?
  • Are entities resolved?
  • Is evidence traceable?
Stage 2

Decision performance

  • Faster cycle time?
  • More consistent?
  • Fewer overrides?
Stage 3

Execution performance

  • Fewer handoffs?
  • More straight-through?
  • Faster exceptions?
Stage 4

Business outcome

  • The stage a CFO banks.
  • Every gain upstream must land here.
Where the economics land
↑Revenue created or protected↓Operating cost avoided↻Working capital improved↓Losses reduced ◑Risk avoided↑Productivity released↓Decision cost reduced»Time to value improved
04 The four layers of ROI

A practical framework, from understanding to economics.

Four connected layers, each with a defining question and measurable indicators. Read from the foundation up: if the system does not understand enough to decide, nothing above it can be attributed. Select a layer to see what it measures.

Two ways it fails. Strong context with no outcome is not ROI. An outcome with no traceability cannot be attributed.
Figure 4 The four layersSelect a layer
What is the actual economic consequence?
Revenue upliftCost reductionMarginAvoided penaltiesReduced leakageLower riskCost per decision
Failure mode AStrong context, no outcome: not ROI.
Failure mode BOutcome, no traceability: cannot be attributed.
05 The new economic unit

AI economics become clear at the level of cost per decision.

Token, infrastructure and licence cost are the visible tip. A decision that looks cheap in compute can still be expensive if ten people spend hours assembling evidence around it, reconciling sources, chasing approvals and handling exceptions.

The target is not a lower cost per token. It is a lower cost per trusted decision, a yardstick that works across sourcing, customer retention, workforce allocation, service assurance and capital deployment.

Figure 5 The true cost of a decisionConceptual
WATERLINE Visible cost Compute, tokens, licences HIDDEN COST Data accessHuman researchReconciliationApprovalsEscalationExecutionExceptions
A useful executive metric
Enterprise
Decision
Economics
=
Business value created
Total cost of decisioning

Context Engineering improves the ratio from both sides: more value per decision and lower cost to enable it.

Decision costwhat it takes to make and execute Decision timeevent to enterprise response Decision qualityconsistency, overrides, explainability Decision outcomethe economic consequence
06 Decision velocity

The largest value is not automating tasks. It is deciding faster.

Hours saved understate the value. The bigger opportunity is shrinking the time between an enterprise event and the enterprise response. With fragmented context, most of that time goes on assembling evidence, reconciling it and waiting for approvals. With engineered context, the evidence is already standing when the event arrives.

Productivity measures how efficiently work is performed. Decision velocity measures how fast understanding becomes action.
Figure 6 Event-to-action latencyIllustrative
Fragmented context
EventEvidence assemblyReconcile, interpretApprovalsDecideAct
Engineered context
EventContext readyDecideGoverned actTime returned to the business
Time from event to enterprise response →
RevenueAct on opportunities while they exist.
OperationsResolve exceptions before disruption.
CapitalDeploy resources where value is highest.
RiskApply policy at the point of decision.
CustomerAct on the full relationship.
WorkforceKeep supply and demand aligned.
07 Governed autonomy as a multiplier

Value multiplies when decisions move safely from recommendation to action.

A recommendation still needs review, validation, approval and follow-up. Governed autonomy removes that friction without giving up control: the system knows what is true, which policy applies, what action is permitted, when escalation is mandatory and how it will be traced.

And every action creates new context, so the next decision starts richer than the last.

Remove unnecessary intervention, not humans. The goal is economically optimal governed autonomy, set decision by decision.
Figure 7 The autonomy spectrumDrag to explore
Human-ledNovel, high-stakes. People decide, AI informs.
AI-assistedAI recommends, people approve within thresholds.
Governed autonomousWell-understood decisions run straight through.
The multiplier logic, worked example
Value per decision$120
×
Decisions per year50,000
×
Straight through62%
=
Value executed without manual friction$3.72M
Transactions update stateOutcomes change relationshipsOverrides reveal exceptionsResults inform thresholdsEvents alter priorities
1 / 7 Reuse is the ROI
How to measure ROI credibly

The business case must survive CFO-level scrutiny.

Measurable, attributable and conservative. Six disciplines turn ROI from a claim into an operating practice, and separate the one-time cost that becomes reusable from the recurring cost that does not.

Baseline first

Capture cycle time, cost, effort, override and automation levels before you start. Without a baseline there is nothing to attribute against.

Measure by domain

Prove value decision by decision, then aggregate. Portfolio numbers earn trust only when each domain can stand on its own.

Split cost types

One-time: integration, ontology, entities, policy models. Recurring: infrastructure, compute, operations, governance. Much of the first becomes reusable.

Attribute conservatively

Claim only the share reasonably due to Context Engineering. A smaller number that holds beats a larger one that is challenged.

Avoid double counting

Productivity, cost and margin overlap. Count each once, and say which line it lands on.

Projected versus realized

Report plan, actual, variance and reason every period, by domain and for the portfolio, where the reuse effect shows up.

Go deeper

The full framework, in two formats.

The executive white paper sets out the measurement framework end to end, from context to decision to execution to outcome, with the cost structure and value realization tracker you can put in front of a CFO. The video covers the economic argument in two minutes.

Executive white paper, 2026 edition

A point of view for CEOs, CFOs, CIOs and CDAOs: the ROI chain, the four layers, cost per trusted decision, decision velocity, governed autonomy and the six disciplines of credible measurement.

Download the white paper PDF · 15 pages · 3.9 MB