Skip to content
Evangelize Consulting

AI economics & governance • EPF-Cost IQ

Token prices keep falling. Why do enterprise AI bills keep rising?

Moving from unattributed pilot spend to defensible unit economics. We connect every prompt, agent loop and inference call to accountable owners and verified business outcomes.

  • Attribution before optimisationTracing spend across models, agents and business units.
  • Architecture over prompt tricksControlling context bloat, reasoning tokens and retry loops.
  • Cost per outcome, not per tokenMeasuring verified value against every pound spent.

The anatomy of an AI bill

Where hidden token consumption multiplies

Focusing solely on headline API price lists masks how modern agentic and reasoning workflows consume compute.

  1. 01 / Standard input tokensSystem prompts, policy documents and conversation history ingested on every turn. Multi-turn sessions compound history rapidly without context pruning.
  2. 02 / Cached input tokensRepeated static instructions and knowledge bases served from provider memory, typically reducing input token rates by 50% to 80% when prompts are structured correctly.
  3. 03 / Output generation tokensResponses generated one token at a time, typically 3x to 4x more expensive than standard input, and amplified when unstructured prose replaces concise schemas.
  4. 04 / Reasoning & agent loopsInternal deliberation tokens and multi-step tool calls. A single user request can silently trigger dozens of background model calls and retries.

AI unit economics & savings simulator

Model what governed architecture is worth.

Set your volume and workflow size, then switch each lever on or off to see how the run-rate and cost per completed outcome change.

500,000
50,0005,000,000
12,000
2,00050,000
Current baseline monthly spend£42,000
Governance & architecture levers

Executive readout

Annualised run-rate, before£504,000
Annualised run-rate, after£171,253
Estimated first-year verified savings£291,15466% lower run-rate, assuming levers phase in over the first quarter.
Cost per 1,000 completed business outcomes£30.36£98.82Completion rate 85% uncontrolled, 94% governed.
Uncontrolled baseline£504,000
EPF-governed architecture£171,253

Illustrative unit-economics model at a blended £7 per million tokens. Actual baseline and savings are established during a 6-week EPF-Core diagnostic.

The 5-lever EPF governance framework for AI

Five controls that turn AI sprawl into a governed asset

Lever 01 of 05

Granular showback & ownership

Tagging every inference call by business unit, product feature, model provider and named executive owner.

What it changes

Every pound of AI spend has an owner who can explain it.

How we engage

Start with a baseline. Keep the control.

Recommended first step

6-week AI spend & baseline diagnostic

EPF-Core

See where the AI bill comes from before deciding what to cut.

Establish immediate visibility across cloud and API providers, identify runaway agent loops, and deliver a prioritised savings roadmap.

Best For

Organisations with growing AI spend across several providers and no single view of who is consuming it, or why.

Ongoing AI FinOps & accountability

EPF-Cost IQ

Keep AI unit costs defensible as usage scales.

Embed continuous token attribution, model routing governance and executive decision cadence into your operating model.

Best For

Organisations moving AI from pilot to production that need cost per outcome reported with the same rigour as any other service.

Scope, fees, responsibilities and acceptance conditions agreed before mobilisation.

Executive decision intake

Bring your AI bill, your architectural constraints, and your target decision date.

We will come back with what the evidence can establish, and how quickly.

0/436