A strong first automation candidate is recurring, costly when delayed or missed, governed by stable rules, supported by usable data, owned by a person, and safe to test or reverse. Do not start with the process that has the most impressive demo. Start with a process whose normal path and failure path the business can explain.
The scorecard in this guide helps compare candidates consistently. It is an Edex Growth decision tool, not a validated industry benchmark, and it does not predict savings or return on investment. A high score still cannot override privacy, safety, legal, financial, or professional-judgment concerns.
Define the process before scoring it
“Send an email” is a task. “Move a valid signed proposal into an owned onboarding path” is a process.
A process has:
- a legitimate event that starts it;
- information required before work can continue;
- a current state and a responsible owner;
- rules for routine decisions;
- a place where judgment or approval belongs;
- a visible success state;
- known exceptions and a recovery path.
If the team cannot agree on those elements, the first project is process definition. Software will repeat an unclear policy more consistently, but it will not make the policy correct.
This distinction also separates task automation from business process automation. IBM describes business process automation as software applied to complex, repetitive business processes, while its explanation of robotic process automation focuses on narrower, rules-based tasks. The practical implication is simple: score the whole handoff, not only the easiest click within it. See IBM’s current business process automation overview.
Apply three stop gates before using the score
Reject or pause a candidate before scoring when any of these gates fails.
| Gate | Question | If the answer is no |
|---|---|---|
| Definition | Can the team describe the correct normal outcome and the known exceptions? | Map or repair the process first. |
| Authority | Is there a named owner who can approve rules, access, tests, and changes? | Assign ownership before building. |
| Safety | Can the first release be tested, observed, stopped, and recovered without unacceptable harm? | Keep the action manual or design approval and rollback controls. |
The stop gates matter because a score can hide one fatal weakness. A high-frequency process with no authorized owner is not ready. An apparently simple financial decision with no correction path is not safe merely because the inputs are structured.
Use the seven-factor automation scorecard
Rate each factor from 1 to 3. Use evidence from the current process, not a vendor demonstration or a best-case future state.
| Factor | 1 point | 2 points | 3 points |
|---|---|---|---|
| Frequency | Occasional | Weekly | Daily or event-driven at meaningful volume |
| Friction | Minor inconvenience | Repeated coordination or retyping | Work regularly waits, repeats, or disappears |
| Rule clarity | Decisions vary case by case | Most rules are known, with exceptions | Inputs, rules, and outputs are stable |
| Cost of delay or miss | Delay has little operational effect | Delay creates rework or confusion | A delay or miss affects a customer, payment, booking, or delivery handoff |
| Exception visibility | Exceptions are hard to define | Common exceptions are known | Exceptions can be detected, classified, and assigned |
| Access and feasibility | Required access or data path is unknown | Some connections need verification | Supported access, permissions, and data paths are confirmed |
| Reversibility | Actions are hard to undo | Some actions require approval | Early actions are safe to test, repeat, stop, or reverse |
UiPath’s current Automation Hub documentation also separates automation suitability from readiness. Its factors include rule clarity, digital and structured inputs, expected process change, and documentation. That does not validate Edex Growth’s seven-factor score or its scale, but it supports the broader practice of assessing candidates before implementation. Review UiPath’s idea-score factors.
Use the total as a discussion aid after all three stop gates pass. The total compares priority and practical readiness for a bounded first release. It does not authorize autonomous action. In particular, a high score for cost of delay or miss means the problem deserves attention, not that a system should make the decision without human review.
- 17 to 21: consider a bounded first release after the stop gates and technical verification pass;
- 12 to 16: map, simplify, or reduce the release before reconsidering the candidate;
- 7 to 11: do not prioritize it as the first automation; keep the affected work manual or redesign the process first.
These bands are professional judgment, not research-backed performance thresholds. Record why each factor received its score. Two people giving the same process different scores is useful evidence that the rules or risks need discussion.
Compare two illustrative business examples
The examples below are hypothetical. They are not client results, and the scores are not promises about implementation effort or value.
Example 1: Route a valid website inquiry to an accountable owner
A service business receives a structured form submission. The current process requires someone to copy the details into a CRM, check whether the person already exists, select the right service queue, and notify a staff member.
| Factor | Score | Reason |
|---|---|---|
| Frequency | 3 | In this hypothetical example, form inquiries arrive daily through an event-driven intake. |
| Friction | 3 | Copying and routing create repeated coordination. |
| Rule clarity | 3 | Required fields and service-routing rules are documented. |
| Cost of delay or miss | 3 | A missed handoff can leave a real inquiry without ownership. |
| Exception visibility | 3 | Missing fields, duplicates, unsupported services, and failed assignment can be detected. |
| Access and feasibility | 2 | The CRM connection and permissions still need a controlled test. |
| Reversibility | 3 | The first release can create a review state before customer-facing action. |
| Total | 20 | Strong candidate if consent, duplicate handling, and fallback ownership pass testing. |
The first release should not automate every follow-up message. A safer scope may validate the submission, find or update the correct record, assign an owner or exception queue, and prove the handoff. Customer communication can follow only after eligibility, wording, exits, and responsibility are approved.
Example 2: Approve unusual customer refunds
The business wants to automate refunds because managers spend time reviewing them. The cases vary, policy is incomplete, evidence arrives in free text, and a mistaken approval is difficult to reverse.
| Factor | Score | Reason |
|---|---|---|
| Frequency | 2 | Requests occur weekly. |
| Friction | 2 | Review requires coordination. |
| Rule clarity | 1 | Managers use judgment that is not documented. |
| Cost of delay or miss | 3 | A delayed or incorrect decision affects money and the customer relationship. |
| Exception visibility | 1 | Unusual cases are not consistently classified. |
| Access and feasibility | 2 | Records are available, but authority boundaries are unclear. |
| Reversibility | 1 | A completed refund is difficult to undo. |
| Total | 12 | Illustrates risk and prioritization only; it does not qualify this process for implementation. |
The incomplete policy means the Definition gate fails, while the unclear authority boundary means the Authority gate cannot yet be verified. Either result overrides the numerical score. The total of 12 is shown to demonstrate how the scorecard surfaces risk and helps prioritize further process work, not to qualify the refund process for implementation.
After those gates are resolved, automation may still prepare the case: collect the approved evidence, flag missing information, calculate a policy-based limit, and place the request in a manager’s queue. The consequential decision remains with the authorized person.
Choose rules, AI assistance, or manual work
Use normal workflow automation when a known event and structured input lead to a known rule. Examples include moving a validated record to a review queue, notifying the assigned owner, or creating a task when a required field is missing.
Consider bounded AI assistance when the input varies but the output can be constrained and checked. Examples include proposing a category for a free-text request, extracting candidate fields from an approved document, or drafting an internal summary. AI is not needed merely because the process contains text.
Keep the consequential decision human-owned when it is sensitive, rare, ambiguous, hard to reverse, or dependent on context the system cannot reliably access. Automation or AI can prepare information for a person without owning the decision.
For AI-assisted steps, define allowed outputs, an unknown path, validation rules, human-review conditions, and what happens when the model or provider fails. NIST’s voluntary AI Risk Management Framework emphasizes context, accountability, transparency, privacy, reliability, and human oversight. Those are design requirements, not a reason to describe an AI output as an autonomous employee.
Once a process has passed the selection gates, use the AI-or-normal-automation decision framework to choose rules, bounded AI, an agent, or human approval for each step.
Check the implementation prerequisites
Before choosing a tool, confirm:
- Process owner: one person can approve the definition and accept the result.
- Source of truth: the system that owns identity, state, and outcome is explicit.
- Trigger: the exact event is distinguishable from a duplicate, retry, or test.
- Required input: fields and evidence needed for the next decision are known.
- Permissions: access is authorized, least-privilege where practical, and testable.
- Rules and exceptions: the normal path, exclusions, and human decisions are written down.
- Recovery: a failed or partial run has a visible state, a retry rule, and an owner.
- Acceptance evidence: the team knows which record, status, log, or notification proves success.
If a prerequisite is unknown, pause the affected step or reduce scope until it is verified. A small process with explicit boundaries is a better first release than an end-to-end diagram that nobody can test.
Avoid five common selection mistakes
Choosing the loudest complaint
Annoyance is not the same as suitability. Compare several candidates with the same scorecard.
Automating an unnecessary step
Remove a redundant approval or duplicated field before building software around it.
Treating the happy path as the process
Missing data, duplicate events, unavailable owners, failed connections, and manual takeover are part of the design.
Starting with customer-facing action
Begin with observable internal state when the business has not yet proved eligibility, content, timing, and stop rules.
Calling a successful software run a successful business outcome
The workflow can finish while updating the wrong record or leaving the next action unowned. Acceptance should check the business state, not only the connector status.
Define a first release that can fail safely
A dependable first release includes normal and exception test data, visible states for completed and failed work, an owner for every exception, duplicate-safe retry behavior, and a recovery procedure.
Write expected results before testing. The platform-specific GoHighLevel workflow testing guide shows how Edex Growth separates expected state, actual evidence, and defects. If the process is still being mapped, use the lead-to-booking workflow mapping method as an example of defining state and ownership before configuration.
The goal of a first automation is not maximum scope. It is one process the team can understand, operate, test, and improve.
If you need help comparing candidates, Business Process Automation and AI Integration explains how Edex Growth maps the current path, defines the bounded future state, and verifies the implementation. A focused systems audit and repair may be the safer starting point when the existing automation is already live but its rules or evidence are unclear.
Sources and editorial scope
External sources were checked on November 1, 2026: IBM’s business process automation overview, UiPath’s current Automation Hub idea-score documentation, and NIST’s AI Risk Management Framework. The scorecard, stop gates, scoring bands, examples, and decision rules in this article are original Edex Growth frameworks. They are not industry benchmarks, legal advice, or forecasts of savings, revenue, or ROI.
