ARIIARun the free scorecard
Free operator guide · no account required

Audit an AI workflow before you automate it.

Before treating an automation problem as a model failure, inspect the handoff: its trigger, accountable owner, required information, exception path, and completion measure. Map those five boxes first.

Prepared by ARIIA, an ALMFC LLC product · Published July 17, 2026 · No client data, testimonials, or reported results were used.

Use one real workflowChoose work that repeats and crosses at least one person, app, or approval boundary.
Use observable factsMap what happens now, not what a policy says should happen.
Change one boundaryMeasure a baseline before connecting more tools or agents.

The five-box handoff audit.

Complete each box in plain language. If two people give different answers, that disagreement is evidence: the workflow is not ready for autonomous execution.

01

Trigger

Name the observable event that starts the workflow. Avoid phrases such as “when ready” or “as needed.”

  • What exact event creates the work?
  • Where is it recorded?
  • Can the same event create duplicates?
02

Owner and next

Name one current owner, the next owner, and the event that proves the next owner accepted the handoff.

  • Who is accountable right now?
  • What status means accepted?
  • What happens when nobody accepts?
03

Required information

Define the minimum complete packet and its source of truth. Do not make an agent guess a missing field.

  • Which fields are mandatory?
  • Which system is authoritative?
  • Where does incomplete work go?
04

Exception and approval

Separate routine work from judgment, money, legal commitments, publishing, deletion, or security changes.

  • What conditions stop automation?
  • Who can approve the exception?
  • What evidence must that person see?
05

One measure

Pick one outcome that the workflow itself can prove. Establish a baseline before claiming improvement.

  • Cycle time, completion, rework, or aging?
  • Where is the timestamp or count?
  • What comparison period is honest?

Copy this map.

Use one sentence per line. A useful map is short enough for everyone involved to challenge.

WORKFLOW:

1. TRIGGER — The workflow starts when…
2. OWNER + NEXT — The current owner is… The next owner accepts when…
3. REQUIRED INFO — The minimum complete packet is… The source of truth is…
4. EXCEPTION + APPROVAL — Stop and ask for approval when… The approver needs to see…
5. MEASURE — We will baseline and compare… using…

Illustrative example.

This example demonstrates the level of specificity. It is not a client engagement, case study, testimonial, or reported result.

Inbound audit request

  1. Trigger: a requester explicitly submits the fit form.
  2. Owner + next: the intake owner reviews it; acceptance is a recorded fit status.
  3. Required info: one workflow, friction, reply email, and contact consent. Authority and timing may be clarified in the reply.
  4. Exception: stop if consent, authority, scope, or safety is unclear.
  5. Measure: time from consented submission to fit decision.

Why this is automation-ready only in parts

Software can validate required fields, timestamp the request, remove duplicates, and prepare a review summary. A person still owns the fit decision, scope, official payment request, and any material promise.

The automation boundary is part of the workflow design. “Human approval” is not a failure state when the action creates a financial, legal, publishing, deletion, or security consequence.

Decide what to automate.

Use the map to choose a control, not to justify a tool you already wanted to buy.

Automate

The trigger, required data, owner, exception rules, and outcome evidence are deterministic and reversible.

Assist

The system can gather facts or prepare a draft, but a named person must decide or commit.

Do not automate yet

Ownership conflicts, missing data, unsafe exceptions, or an absent baseline would make automation amplify ambiguity.

A seven-day audit loop.

Do this before a major integration. It costs attention, not software.

Days 1–3: establish facts

  1. Map the five boxes with the people doing the work.
  2. Inspect five recent cases, including one exception.
  3. Mark every wait, copy-paste step, rejected handoff, and missing field.

Days 4–7: change one boundary

  1. Choose one measure and record the baseline.
  2. Clarify one owner, required packet, or acceptance signal.
  3. Observe the same measure without claiming causation from a tiny sample.

Score one workflow privately.

The free scorecard runs on your device and does not submit answers. If you later choose to request a direct fit review, the separate ARIIA AI Operations Audit is USD $497 for one workflow; scope and payment terms are confirmed first, implementation is separate, and no outcome is guaranteed.

Run the free scorecardInspect the illustrative samplePrepare a fit request

No account, cookie, advertising tracker, hidden checkout, purchase, invoice, subscription, or marketing opt-in is created by reading this guide or running the scorecard.