CLOVER DIGITAL — ONE-WORKFLOW PILOT TEST
Version 1.0 · September 29, 2026

Use the same task and test inputs when comparing a human assistant, a DIY
automation platform, or a managed service. This is a proposed test plan,
not a benchmark result or a claim that Clover or another provider has passed.
Copy this file into a text editor. No account or email address is required.

AGREE THE TASK BEFORE TESTING
Workflow:
Start condition:
Authoritative record and required information:
Permitted action (draft / send / update):
Actions that require approval, and who approves:
Stop condition:
Provider/operator responsible for maintenance:
Exception owner and backup:
Agreed response time for an exception:
Setup cost / recurring cost / usage or change charges:

Use fictional data in a test inbox or sandbox. Separate a draft preview from
an action actually performed in a connected tool. Agree the expected result
before each run, then inspect the destination record as well as the reply.

RUN THESE SEVEN CASES

1. NORMAL REQUEST
Input: A complete, eligible request with the required information.
Expected: Exactly the agreed action, using the current source record.
Evidence: Destination record or draft, timestamp, and final state.

2. DUPLICATE INPUT
Input: Deliver the same request twice, using the same request reference.
Expected: No duplicate customer message or second booking/update.
Evidence: Request reference and action count in the destination system.

3. CHANGED OR CANCELLED REQUEST
Input: Queue a follow-up, then add a reply changing or cancelling the request.
Expected: The obsolete follow-up stops; the changed request reaches its owner.
Evidence: Latest thread, queue state, and handoff record.

4. MISSING OR CONFLICTING INFORMATION
Input: Omit the time zone, or provide two different appointment times.
Expected: Ask for clarification or hold for review; do not guess or confirm.
Evidence: Missing information identified and actual confirmation state.

5. APPROVAL REQUIRED
Input: Request a discount, refund, or commitment outside the agreed scope.
Expected: Hold the action for the named approver. If approval never arrives,
follow the agreed escalation rule without treating silence as approval.
Evidence: Pending action, approval state, and named owner/backup.

6. CONNECTION FAILURE
Input: In a test environment, make the required tool unavailable.
Expected: No false success message. The operator gets an exception and any
retry avoids duplicating an action that already succeeded.
Evidence: Error, exception notification, destination state, and retry outcome.

7. HANDOFF AND STOP
Input: Ask the backup to take over an open task, then stop the pilot.
Expected: The backup can see the context and next step. Scheduled actions
stop, and access removal is verified under the agreed offboarding process.
Evidence: Handoff record, remaining queue, and access/action state after stop.

RESULT RECORD — COPY FOR EACH CASE
Case / request reference:
Date, time, and time zone:
Operator / provider / product and configuration version:
Expected action and stop condition:
Observed action (preview or actually performed):
Evidence reference in the approved test system:
Result (pass / fail / not tested):
Review, correction, and troubleshooting minutes:
Failure owner / correction / retest result:

ILLUSTRATIVE CASE 3 — EXPECTED BEHAVIOR, NOT AN EXECUTED RESULT
Task: Remind a fictional customer about an appointment.
Source record: TEST-001, October 6, 2026, 10 a.m. America/Chicago.
Initial state: Reminder queued for October 5 at 10 a.m. America/Chicago.
Changed input: Customer replies, "Please cancel; I'll contact you to rebook."
Expected result: Cancel the old reminder and route the cancellation to the
named appointment owner. Do not invent a replacement booking or report that
the calendar was updated unless that action is permitted and actually verified.

DECIDE WHETHER TO EXPAND
Retest failures before live use. A successful preview does not prove inbox
delivery or a working integration. Keep "not tested" separate from "pass."
For an agreed live pilot, record completed tasks, corrections, exception volume,
operator time, your review time, and the full cost over the same period.
Recovered time = prior handling time minus current review, correction,
maintenance, and troubleshooting time. Do not assume a positive result.

Published explanation and version:
https://cloverdigital.com/blog/lindy-alternative-for-small-business.html#workflow-pilot-test

Controlled Clover example, with its test limits disclosed:
https://cloverdigital.com/guides/small-business-automation-service.html#inquiry-demo