B
Blue Edgewater
← AI Workforce Enablement
Artifact 04 · AI Workforce Enablement
AI Workflow Automation Starter Kit
A practical guide to converting repetitive office work into safe, reviewable AI-assisted workflows — without writing a single line of code.
01Workflow Map
02SOP
03Checklist
04Before / After
This starter kit uses a real office workflow: employee intake requests. An employee submits a request form, AI summarizes and categorizes it, a manager reviews and approves, the task gets logged, and a follow-up email is drafted. The same pattern applies to dozens of office processes.
5 steps
3 AI touchpoints
2 human gates
No code required
End-to-end workflow — intake to follow-up
1
Trigger
Employee submits intake form
An employee completes a structured intake form — a request for a resource, support, approval, or information. The form captures name, department, request type, priority, and description.
Tools
Microsoft Forms, Google Forms, Jotform, or any form tool your org uses
2
AI Step
AI summarizes and categorizes the request
The form response is passed to an approved AI tool with a standard prompt. The AI produces a 3-5 sentence summary, assigns a category, flags urgency, and identifies any missing information.
Sample prompt template
“You are an intake coordinator assistant. Summarize this employee request in 3-5 sentences. Assign one category from: [IT, HR, Facilities, Finance, Other]. Flag urgency as High/Medium/Low. Note any missing information needed to process the request. Request: [PASTE FORM RESPONSE]”
3
Human Gate
Manager reviews AI summary and approves routing
A human reviews the AI-generated summary alongside the original form response. They confirm the category, adjust urgency if needed, and approve routing to the correct team. Nothing moves forward without this step.
What the manager checks
Category is correct • Urgency matches context • Summary is accurate • No sensitive info was mishandled • Request is complete enough to act on
4
AI Step
Task logged and tracked
Once approved, the request details — including the AI summary — are added to a shared task tracker. AI can assist by formatting the entry, suggesting a due date based on urgency, and tagging the correct assignee queue.
Tools
Microsoft Planner, Asana, Trello, SharePoint list, or a shared spreadsheet
5
AI Step + Human Gate
AI drafts confirmation email, human sends
AI generates a confirmation email to the employee: acknowledging receipt, stating the category and timeline, and naming the next step. A human reviews and sends — the AI never contacts the employee directly.
Sample prompt template
“Draft a professional confirmation email to [NAME] acknowledging their [CATEGORY] request submitted on [DATE]. Include: request received, assigned to [TEAM], expected response by [DATE]. Tone: warm and clear. Length: 3-4 sentences.”
Standard Operating Procedure
AI-Assisted Intake Processing
Step-by-step instructions for the coordinator role
Follow these steps in order each time an intake request is received. Do not skip the human review steps. If anything in the AI output seems wrong, correct it before proceeding.
1
Receive and open the form response
Check your intake inbox or form dashboard for new submissions. Open the response and read it fully before doing anything else. Do not skim.
Tip: Set up a notification rule so new form submissions alert you immediately rather than batching.
2
Copy the form response into the AI summary prompt
Open your approved AI tool. Use the standard intake summary prompt (stored in your team’s prompt library). Paste the form response into the [PASTE FORM RESPONSE] field. Do not modify the prompt structure.
Risk: Never paste responses that include employee PII, health information, or financial data into a public AI tool. Use only your organization’s approved AI tool for this step.
3
Review and validate the AI summary
Read the AI output carefully. Verify that the summary accurately represents the original request, the category is correct, and the urgency flag matches the context. Correct any errors directly in the output before proceeding.
Human gate: You are responsible for this summary. If you pass it forward, you are confirming it is accurate. The AI does not own this — you do.
4
Route to the correct manager for approval
Send the validated summary plus the original form response to the appropriate manager. Include the category, urgency flag, and any notes you added. Wait for approval before logging or communicating.
Tip: Use a standard routing email template so managers receive consistent, easy-to-scan information every time.
5
Log the approved request in the task tracker
Once manager approval is received, create a task entry using the approved AI summary as the description. Add the category tag, urgency level, requestor name, submission date, and target response date.
Human gate: Confirm the entry is complete and correctly tagged before marking it as logged. Incomplete tracker entries create follow-up problems.
6
Generate and review the confirmation email draft
Use the confirmation email prompt template with the requestor’s name, category, assigned team, and expected response date. Read the full draft before sending. Edit any language that does not sound like your organization’s voice.
Risk: Do not send AI-generated emails without reading them fully. AI can produce technically correct but tonally wrong emails. A confused or cold confirmation email damages trust.
7
Send the confirmation and document AI use
Send the reviewed email from your own account. In the task tracker, note that AI was used for summarization and email drafting, and that human review was completed at steps 3 and 6. This is your audit trail.
Documentation: Log: tool used, steps AI assisted with, human review completed Y/N. This takes 30 seconds and protects you and your organization.
8
Troubleshooting
If the AI summary is clearly wrong: correct it manually and note the error. If the AI refuses to process the request: check whether the content violates data policy and handle manually if needed. If the email draft is off-tone: rewrite it yourself. Never send something you would not be comfortable signing your name to.
Tip: Keep a running log of prompt failures or AI errors. Patterns in failures help you improve your prompt templates over time.
Automation checklist
AI Workflow Readiness Checklist
Complete before deploying any AI-assisted workflow
Use this checklist before launching any AI-assisted workflow in your team or organization. Each item represents a failure point that has caused real problems in real deployments.
Workflow design
The workflow has a clearly defined trigger (what starts it) and a clearly defined endpoint (what done looks like)
Workflows without defined endpoints expand indefinitely and create confusion about who owns the final output
Every AI step has a corresponding human review step before the output is used or passed forward
AI outputs that are never reviewed become a source of undetected errors
The workflow has been mapped on paper or a whiteboard before any tool is configured
Build the logic before you build the automation
There is a manual fallback for every automated step in case a tool is unavailable
Workflows that cannot be run manually create single points of failure
Data and compliance
The data flowing through this workflow has been reviewed and contains no PII, PHI, financial data, or confidential client information
If it does, confirm your AI tool is approved for that data classification before proceeding
All AI tools used in this workflow are on the organization’s approved tool list
Unapproved tools invalidate your compliance posture even if the workflow is otherwise sound
IT or your compliance team has been informed of the workflow and has not raised objections
Prompt quality
Each AI prompt has been tested at least 5 times with real or realistic inputs before being used in production
A prompt that works once is not a reliable prompt
Prompt templates are stored in a shared, version-controlled location accessible to the whole team
Prompts that live only in one person’s head or browser history are not organizational assets
Edge cases have been tested: incomplete inputs, unusual requests, ambiguous data
Launch and measurement
A baseline task-time measurement has been recorded before the workflow goes live
You cannot measure improvement without a starting point
A 30-day review is scheduled to assess accuracy, time savings, and any emerging issues
Everyone who will use this workflow has been trained on the SOP and knows the escalation path if something goes wrong
Efficiency comparison
Before and After: Intake Processing
These estimates are based on a coordinator processing 10 intake requests per week. Actual results will vary by organization, volume, and workflow complexity.
Before AI workflow
Time per request
18-22 min
Weekly time (10 requests)
3.5 hrs
Tasks done manually
- Read and re-read full request
- Write summary from scratch
- Decide category manually
- Email manager for approval
- Manually log to tracker
- Write confirmation email
- Send and file
After AI workflow
Time per request
6-9 min
Weekly time (10 requests)
1.2 hrs
Tasks assisted by AI
- Read request (still human)
- AI drafts summary → human verifies
- AI assigns category → human confirms
- Routing template used
- AI formats tracker entry
- AI drafts email → human reviews
- Human sends and documents
Risk points and mitigations
AI misclassifies urgency
AI may flag a routine request as high-urgency or downgrade a genuinely urgent one based on word choice rather than context.
Fix: Manager review step at Step 3 is mandatory. Never skip it for speed.
Sensitive data entered into unapproved tool
If a form response contains employee health information or financial data and is pasted into a public AI tool, it is a compliance violation.
Fix: Screen every form response before Step 2. Handle sensitive requests manually.
AI email draft sent unreviewed
A coordinator under time pressure sends the AI draft without reading it. The email contains a commitment or timeline the organization cannot honor.
Fix: Build a 60-second read-aloud habit before sending any AI-drafted email.
Prompt template drift
Someone edits the shared prompt template without team awareness. Output quality degrades inconsistently and the cause is hard to trace.
Fix: Version-control your prompt templates. Treat prompt changes like policy changes.
Ready to design an AI-assisted workflow for your team? Jason Boursier builds practical AI enablement programs that start with your actual processes — not generic slide decks.
Work with Jason →