Automation Maturity Score

Write the trigger, intended beneficiary, required inputs, accountable owner, permitted actions, terminal outcome, and review period. Examples include preparing a customer response, routing an inbound inquiry, reconciling a supplier invoice, or releasing a software change. Do not score a department or an entire company until its distinct workflows have been assessed separately.

Gather three recent examples, including one exception, and identify the authoritative source for each material fact. Record current volume, elapsed time, review burden, error or rework rate, cost, and outcome measure when available. If no baseline exists, mark it unknown. Maturity cannot be established by the sophistication of a model demonstration alone.

Each question uses a 0-to-4 scale: zero is absent or unknown, one is informal, two is defined but mainly manual, three is governed and repeatable for a bounded case, and four is measured, recoverable, and improved from observed outcomes. Cite an artifact, system record, test, policy, or representative example beside every score above one.

A score should fall when evidence contradicts the intended design. If a procedure says approval is required but recent work bypassed it, score the observed operating state. Unknowns should remain visible rather than being averaged away. This assessment supports prioritization; it is not a compliance opinion, security assessment, financial assurance, or product-availability claim.

Score objective clarity from 0 to 4. Four requires a specific beneficiary, measurable operating result, time boundary, and condition under which the workflow should not run. Score accountable ownership separately. Four requires a person with authority to accept, reject, stop, and change the workflow rather than a generic team or technical service account.

Score value hypothesis from 0 to 4. A four requires a stated baseline, expected change, value measure, guardrail, review cadence, and rule for stopping or scaling. Add the three answers for an intent subtotal from 0 to 12. A polished prompt with no owner or baseline should not score above one.

Score terminal definition from 0 to 4. Zero means "task completed" is the only signal. Two means a person can tell whether the output is usable. Four means the destination records acceptance, refusal, or a governed exception and links that state to the original objective. A draft, queue message, worker completion, or provider response is not necessarily the business outcome.

Score exception ownership from 0 to 4. Four requires classified exceptions, receiving owners, response expectations, evidence, and a decision that returns the item to a valid terminal state. Add both answers to the intent subtotal for a maximum of 20. Hidden exception queues are a common reason apparently automated work still depends on manual rescue.

Score source authority, source quality, and identity consistency from 0 to 4. Four requires named authoritative owners, validated fields or documents, current-enough data, and stable matching across the systems that participate. A general search result, copied note, or unverified model memory does not qualify as a governed source for a consequential decision.

Score access governance from 0 to 4. Four requires purpose-bound access, least privilege, approved custody, revocation, and an understood failure path. Add the four answers for a context subtotal from 0 to 16. A connector name or successful login does not prove that the correct account, scope, fields, retention, and provider policy are ready.

Score decision evidence from 0 to 4. A four lets a reviewer see the objective, relevant sources, uncertainty, recommendation, authority, expected effect, and approval. Score execution evidence separately. A four records what identity acted, what changed, the destination response, cost, exception, and final acceptance or refusal.

Score replay and lineage from 0 to 4. Four means a disputed or failed item can be reconstructed from retained evidence without relying on one person or an ephemeral transcript. Add the three answers to the context subtotal for a maximum of 28. Evidence retention must still follow privacy, security, contractual, and legal requirements; more retention is not automatically better.

Score action authority from 0 to 4. Four requires explicit limits by role or identity, purpose, data, system, action, amount or capacity, environment, and time. Score approval separation from 0 to 4. Four means preparation, recommendation, execution, publication, funds movement, and release are separated wherever consequence requires independent authority.

Score intervention from 0 to 4. Four requires a current owner who can observe, pause, revoke, redirect, or terminate the work without losing the evidence needed for resolution. Add the three answers for an authority subtotal from 0 to 12. An interface that says "human in the loop" does not score highly unless the decision and intervention actually work.

Score failure detection, safe stop, and recovery ownership from 0 to 4. Four for detection requires useful signals for missing context, provider error, policy conflict, budget breach, duplicate action, and destination rejection. Four for safe stop means the workflow fails closed when required authority or evidence is absent rather than improvising a wider action.

Four for recovery requires a tested path to reconcile state, reverse what can be reversed, notify affected owners, address customer or operational impact, and decide whether the workflow may resume. Add the three answers for a recovery subtotal from 0 to 12. High-risk or irreversible actions may remain human-approved even with a high score.

Score predicted cost, actual cost attribution, and budget control from 0 to 4. Cost includes model or provider usage, infrastructure, storage, tool calls, retries, human review, implementation, support, and recovery where material. Four for attribution requires costs to be associated with the workflow and outcome rather than only a monthly vendor total.

Four for budget control requires a defined limit, responsible owner, alert or stop behavior, and a process for variance review. Add the three answers for an economics subtotal from 0 to 12. Do not infer margin, labor savings, or provider-cost reduction from a usage dashboard without reconciling the relevant commercial and financial evidence.

Score outcome attribution from 0 to 4. Four requires a defensible link from the workflow to an accepted operating, customer, revenue, cost, quality, or risk result, including material exclusions and uncertainty. Score decision use from 0 to 4. Four means the evidence actually informs whether the workflow is stopped, repaired, retained, or expanded.

Add both answers for an economics maximum of 20. Engagement, clicks, model responses, generated assets, completed tasks, and queued work can be useful leading indicators, but they are not automatically qualified demand, recognized revenue, released capacity, or accepted customer value. Keep each measure at the level its evidence supports.

Score pre-action prediction from 0 to 4. Four requires an expected outcome, confidence or uncertainty, expected cost, risk posture, and stop or scale rule before execution. Score observation from 0 to 4. Four requires timely measures of the workflow, destination, exceptions, cost, and accepted outcome rather than only the model output.

Score comparison from 0 to 4. Four means predicted and actual results are compared at a defined cadence with meaningful variance and error analysis. Add the three answers for a learning subtotal from 0 to 12. Retrospective storytelling after success is not a prediction, and collecting telemetry without a decision owner is not a learning loop.

Score adaptation from 0 to 4. Four means approved evidence can change routing, instructions, source requirements, thresholds, budgets, review intensity, or autonomy within explicit limits. Score durable memory from 0 to 4. Four requires retained decisions, outcomes, exceptions, and changes with lineage, ownership, access control, and a way to correct stale or wrong material.

Add both answers for a learning maximum of 20. A workflow must not rewrite its own purpose, permissions, public claims, financial authority, or risk appetite merely because a local metric improved. Material changes require the appropriate human or governance review, and privacy or retention obligations continue to apply to learning data.

Add intent out of 20, context out of 28, authority out of 12, recovery out of 12, economics out of 20, and learning out of 20 for a maximum of 112. As a directional guide: 0-25 aligns with A1 assistance, 26-50 with A2 governed preparation, 51-75 with A3 bounded execution, 76-95 with A4 adaptive closed loop, and 96-112 may support an A5 assessment for the applicable terminal outcome.

Apply three overrides. A workflow cannot exceed A2 when consequential release authority is unclear. It cannot exceed A3 when predicted-versus-actual learning and governed adaptation are absent. It cannot be called A5 unless the applicable business outcome closes with accepted downstream evidence, not merely a completed schedule or worker run. Legal, privacy, security, financial, people, customer, and production controls can impose stricter boundaries.

Example: an invoice-preparation workflow totals 63, which falls in the A3 range, but actual supplier cost is not attributed and exceptions have no response expectation. The company should not jump to autonomous approval or payment. It can retain bounded preparation, name the exception owner, connect cost evidence, and test recovery before considering a wider authority boundary.

Choose the three lowest evidence-backed questions that block the next useful operating outcome. For each, assign an owner, artifact or change, due date, success measure, stop condition, and reassessment date. Verify product, package, integration, and deployment availability separately before describing OmegaOS as the active solution. Request a Company Audit when the team needs a facilitated operating map rather than a self-score.