Automation Scope Triage (Onboarding)

Automation Scope Triage (Onboarding)

1. 01

   Inventory every manual back-office task the client's team touches in a normal week

   Capture hours per task, who owns it, and which system it lives in. Contractor onboarding, invoice entry, and compliance checks usually surface first because they repeat on a fixed cadence.

2. 02

   Label each task by judgment density before any tool is shortlisted

   Split the list into rule-following work (reconciliation, data entry, payment triggers) and work that needs a human call (disputed invoices, contractor classification edge cases). Only the first bucket is a candidate for agent execution.

3. 03

   Confirm the underlying data is clean enough to automate against

   Most failed AI projects trace back to poor data preparation before model selection, so run a deduplication and required-field pass on the client's ERP, payroll, and vendor records first.

4. 04

   Map which systems expose APIs and which only expose a screen

   API-connected stacks suit finance agents that write entries directly into ERPs, while legacy portals with no integration path point toward computer-use agents that read the screen instead.

5. 05

   Assign a stage label to each candidate workflow before scoping the retainer

   A proof of concept, prototype, pilot, and MVP answer different questions. Skipping the pilot stage is the most common agency mistake when deploying automation into a client environment.

6. 06

   Document who controls pricing, feature access, and dispute resolution for each vendor in the stack

   Vendors that set usage limits, gate features, and arbitrate their own disputes concentrate risk. Flag any workflow where a single vendor holds all three roles.

7. 07

   Set a human-in-the-loop threshold per workflow and get it signed off

   Define the dollar value or exception type that forces escalation to a person. Over-automating client-specific processes is the failure mode that erodes trust fastest.