EFF / MT–01Method

Drawing EFF / MT–01 · Evidence system

The Effority Method.

A production-grade AI engineering method that resolves uncertainty before the POC, validates the final implementation before release and controls every change after deployment.

01Evidence-gated workflow

Ten phases. One production standard.

Every phase produces the evidence required to continue, adjust or stop. The POC demonstrates an already-validated AI capability; it is not used to discover whether the AI works.

Engineering principles

Apply across every phase

01Outcome before technology
02Acceptance before experimentation
03Evidence before investment
04Feasibility before POC
05Production path before solution demonstration
06Verification before acceptance
07Acceptance before release
08Controlled evolution after deployment

Evidence-gated workflow

0009

Each phase must earn the next.

  1. 00discover

    Business & Process Discovery

    Identify the business decision or process where AI may add value, the users it will support, the required human judgement and the cost of getting it wrong.

  2. 01define

    Requirements, Risks & Success Criteria

    Translate the business need into measurable AI behaviour: target outcomes, error costs, human oversight, acceptance thresholds and go, adjust or stop rules.

  3. 02prepare

    Data & Evaluation Readiness

    Verify data access, quality, ground truth and representativeness, then lock the evaluation set and scoring method used to judge the AI.

  4. 03validate AI

    AI Feasibility & Statistical Validation

    Test the non-deterministic AI capability against the agreed statistical and business thresholds before designing the wider solution. This is not the POC.

  5. 04assess

    Solution & Production Readiness

    Define how the validated AI will be served, secured, integrated, observed and evolved, including data pipelines, infrastructure, CI/CD and MLOps.

  6. 05prove

    End-to-End AI POC

    Demonstrate the validated AI capability inside the intended interface, workflow, integrations, guardrails and human controls.

  7. 06build AI

    Production AI Build & Integration

    Turn the validated AI capability into a production-grade implementation with reproducible pipelines, versioned models, prompts and data, scalable inference, guardrails, human oversight and MLOps.

  8. 07test AI

    AI Testing & Technical Verification

    Test the final AI implementation against locked evaluation sets and failure scenarios, covering quality, robustness, regression, latency, cost, security, fallbacks and monitoring.

  9. 08accept AI

    Controlled AI Pilot & Final Acceptance

    Validate the complete AI system with representative data and users, measuring prediction quality, false positives and negatives, escalations, workflow fit and business value against the agreed acceptance thresholds.

  10. 09operate AI

    Production AI & Controlled Evolution

    Release progressively and monitor model quality, drift, latency, cost and operational impact. Every model, prompt or data change must pass versioned evaluation, regression and approval gates.

Evidence gate / after every phase

Continue · Adjust · Stop

No phase advances on optimism. It advances on evidence.

02 / Apply the method

Bring us the business decision, the available data and the production constraint. We will help establish the evidence required for the next investment.

Discuss an AI or data initiative.