The primary economic unit is the business process or unit of work, not the programme, application or model. Current Process TCO brings together internal labour, outsourced and managed-service labour, applications and platforms, infrastructure and cloud, licences, data and integration, operating overhead, controls and compliance, and attributable change and maintenance burden.

In the simulator: volume, people, cost per person, outsourced cost, applications, infrastructure and licences, data and integration, overhead and controls.

Concentration is broader than model or token cost. The question is where cost or dependency is disproportionately concentrated: in processes, labour pools, managed-service providers, platforms, software vendors, cloud providers, exception paths, AI models and providers, high-cost execution loops, scarce skills or key people.

The simulator shows the largest cost pool and the share held by external providers. The retained run-cost method also measures concentration twice, by cost and by load, because the process that consumes the most capacity is not necessarily the one that consumes the most money.

AI addressability is never translated straight into headcount. Effort is split into automatable execution, augmented work, exception handling, supervision and assurance, accountable judgement, and the new AI-operations labour the future model needs.

Which work AI can change economically, not merely technically. A process is not automatable because someone named it. The simulator asks how much of the automatable work AI actually takes, and how much effort augmentation saves.

Current TCO is never compared with Future TCO while the cost of getting there is hidden. Transformation investment may include process redesign, engineering and integration, data remediation, AI and agent development, change management, workforce transition, training and reskilling, parallel operation, assurance and validation, migration and decommissioning.

What remains after change: residual human execution, exception handling and supervision, accountable judgement, and the applications, infrastructure and licences the process still needs. Technology does not disappear by default; decommissioning is a decision with its own cost and timing.

One component of Future Process TCO, never the whole proposition. See the mechanics section below for what it contains and how a cost per unit is built up.

Residual human cost, AI-operations staff, remaining outsourced cost, retained overhead and technology, plus the AI Cost Layer. The objective is not to prove that AI is cheap. It is to test whether the whole process becomes economically better once every material cost is included.

Cashable cost reduction: expenditure that can actually leave the P&L. Cost avoidance: future spend or hiring that no longer needs to happen. Capacity release: effort that is freed but remains an organisational cost unless it is redeployed or removed.

The three are never added together and called savings. Payback is calculated on the cashable change only.

Cost per unit of work today and after change, and how that cost behaves as volume grows. AI variable cost scales with volume and does not fall per unit; fixed platform cost does. The simulator shows cost per unit from half to double today’s volume.

A single figure hides which assumption carries it. The simulator moves one assumption at a time by a stated amount and ranks the result, so the discussion is about the assumption that carries the number rather than the number itself.

ROI at approval is not the end. What was modelled becomes what is measured, so the economics that justified the change are compared with the economics the running process actually produces. See the last section of this page.