Template 02 — AI Agent Intake Form

Purpose

A single mandatory front door for every new AI agent request. Filters bad ideas before any build effort. Captures the minimum data needed for triage and risk classification at M7 in the In Action roadmap.

• When you use it: Every time a Department Champion or employee wants to propose a new AI agent. No exceptions — exec-sponsored or routine, intake first.
• Who fills it: Department Champion (on behalf of the workflow's actual user).
• Where it lives: As a Notion form, Airtable form, ServiceNow request, Jira ticket type — wherever your registry lives.
• Time to fill: 10–15 minutes if Discovery (M6 in the roadmap) is complete.
• What happens next: Registry entry auto-created with Status = Intake. CoE Lead triages within 5 business days.

Worked example (AP Accountant invoice reconciliation)

"Finance Champion fills the intake on 2026-04-12 after sitting with the AP accountant for a 90-minute discovery session."

1. Request title

Finance — AI-assisted invoice reconciliation

2. Submitter info

• Submitting Champion: Mike Chen (Finance Champion, AP Manager)
• Date submitted: 2026-04-12
• Department: Finance / Accounts Payable
• Workflow's primary user: Sarah Patel (AP Accountant)
• Backup user: Tom Riley (AP Accountant)

3. Workflow being targeted

The accountant receives vendor invoice emails throughout the week. Each invoice arrives as a PDF attached to an email from accounts.payable@<vendor>.com. The accountant manually:

1. Opens the email + downloads the PDF
2. Extracts vendor name, total amount, line items, and PO reference from the PDF
3. Opens NetSuite, searches for the matching open PO
4. Compares invoice amount to PO amount
5. If variance ≤ 2%: marks ready-to-pay queue
6. If variance > 2%: emails the requesting department for clarification

Volume: ~120 invoices per week. Time per invoice: ~3 minutes (average), ~8 minutes (exceptions). Total: ~6 hours / week.

4. Current pain point

• Volume: 6 hours/week of repetitive low-judgment work that prevents the accountant from doing higher-value vendor-management work.
• Error rate: ~2% of weeks have a missed match that's caught downstream during month-end close.
• Speed: Bottleneck at month-end when invoice volume doubles.
• Skill: No skill development for the accountant; the work is rote.

5. Expected business value

• Time savings: 5 hours / week saved (~83% of current time)
• Error reduction: Target 0 missed matches at month-end
• Loaded cost saved: 5 hrs × $55/hr loaded × 52 weeks = ~$14,300 / year
• Reallocated effort: Accountant moves to vendor management, terms negotiation, exception handling

6. Data the agent will see

• From: AP accountant's Gmail mailbox (scoped to invoice-bearing emails)
• Data classifications touched:
  • ✅ Internal (vendor names, PO numbers)
  • ⚠️ Confidential (vendor contract terms within invoices)
  • ✅ Mild PII (vendor contact names/emails — not customer PII)
  • ❌ No PHI
  • ❌ No regulated financial data (no SSN, no bank account numbers in scope)

7. Decisions the agent will make

• Advisory: Yes — proposes the matched PO + match confidence score to the accountant.
• Automated: No — every NetSuite action requires accountant click-approval.
• Consequential about people: No.
• Customer-facing: No — internal-only.

8. Systems the agent will read from / write to

Explicitly NOT in scope: posting journal entries, modifying vendor records, sending to external parties, NetSuite write of any kind.

9. Estimated volume per month

• Normal months: 120 invoices × 4 weeks = ~480 / month
• Month-end weeks (3–5 days/month): spike to ~250 / week
• Peak: ~600 / month at quarter-end

10. Proposed autonomy level

Stage 1 (Assistive). Agent drafts reconciliation proposals; accountant approves each one before any NetSuite action.

Possible promotion to Stage 2 (Validated) after 30 days at ≥90% acceptance rate — but only with re-approval from the CoE Lead and the Head of Finance. Stage 3 (Autonomous) is out of scope at launch.

11. Known constraints

• Regulatory: EU AI Act not in scope (no EU vendors, no EU customers in scope for this agent). NIST AI RMF voluntary alignment desired.
• Contractual: No customer-facing exposure, so no contractual constraints.
• Operational: NetSuite API rate limits — must batch reads and handle retries.
• Timing: Department wants pilot by end of Q2 2026 (10 weeks from intake).

12. Risk red-flags (quick self-assessment by submitter)

Submitter's expected risk tier: Medium (financial data, writes to ERP via human approval, mild PII).

13. Workflow documentation attached

✅ Linked: https://[internal-wiki]/finance/ap-reconciliation-asis-workflow ✅ Loom recording: https://[internal-share]/recordings/2026-04-09-ap-recon-shadow ✅ Sample anonymized invoice PDF (golden-set candidate)

14. Signed by Champion

Blank template (copy below for your company)

# AI Agent Intake Form — [Request Title]

## 1. Request title
[Short descriptive name: "Department — what the agent does"]

## 2. Submitter info
- **Submitting Champion:** [Name + role]
- **Date submitted:** [YYYY-MM-DD]
- **Department:** [Department name]
- **Workflow's primary user:** [Name + role]
- **Backup user:** [Name + role, or "none"]

## 3. Workflow being targeted
[Describe how the work is done today. Step by step. Be specific about systems and decisions. Include current volume and time per task.]

## 4. Current pain point
[What specifically is wrong with the workflow today? Volume? Errors? Speed? Cost? Skill bottleneck?]

## 5. Expected business value
- **Time savings:** [hrs/week]
- **Error reduction:** [target]
- **Loaded cost saved:** [$/year]
- **Reallocated effort:** [what the human will do instead]

## 6. Data the agent will see
- **From:** [Sources]
- **Data classifications touched:**
  - [Public / Internal / Confidential / PII / PHI / regulated financial] — check each

## 7. Decisions the agent will make
- **Advisory:** [Yes / No]
- **Automated:** [Yes / No]
- **Consequential about people:** [Yes / No]
- **Customer-facing:** [Yes / No]

## 8. Systems the agent will read from / write to

| System | Action | Scope |
|---|---|---|
| [System 1] | [Read / Write] | [Scope description] |
| [System 2] | [Read / Write] | [Scope description] |

Explicitly NOT in scope: [List things the agent will NOT do.]

## 9. Estimated volume per month
- **Normal months:** [N]
- **Peak periods:** [N]

## 10. Proposed autonomy level
[Stage 1 / Stage 2 / Stage 3 — and reasoning.]

## 11. Known constraints
- **Regulatory:** [EU AI Act exposure? Sector regulations?]
- **Contractual:** [Customer contracts that restrict AI?]
- **Operational:** [API rate limits, system constraints]
- **Timing:** [Deadline if any]

## 12. Risk red-flags (quick self-assessment)

| 3 risk drivers | Self-assessment |
|---|---|
| Processes PII / customer / employee data? | [No / Mild / Yes — what kind] |
| Makes consequential decisions about people? | [No / Yes — what decisions] |
| Acts autonomously without HITL? | [No / Yes — at what threshold] |

**Submitter's expected risk tier:** [Low / Medium / High]

## 13. Workflow documentation attached
- [Link to workflow doc]
- [Link to recording, if any]
- [Sample data / files attached]

## 14. Signed by Champion

| Role | Name | Date |
|---|---|---|
| Department Champion (submitter) | | |

What happens after submission

1. Auto-create registry entry (Status = Intake). Slack notification to #ai-coe.
2. CoE Lead triages within 5 business days at M7 in the In Action roadmap:
   • Confirms intake completeness
   • Tags the 3 risk drivers
   • Assigns Low / Medium / High tier
   • Checks EU AI Act Annex III exposure (if any)
   • Checks alignment with the signed risk appetite (template 01)
   • Scores on Value-vs-Risk matrix
   • Sets priority
3. Approval decision (M8) — approvers per tier sign off:
   • Low: CoE Lead
   • Medium: CoE Lead + Department Head
   • High: CoE Lead + Department Head + Security + Legal
4. If approved: Status moves to Approved. Move to M9 — Agent Card written (template 03).
5. If rejected: Submitter notified with rationale. Registry entry closed with reason.
6. If deferred: Re-review date set. Champion notified.

SLA target: from intake submission to approval decision = 10 business days for Low/Medium, 20 business days for High.

Usage notes

• Don't skip Discovery (M6). This form is meant to be filled out after the Champion has done the 90-minute observation session. A form filled without Discovery looks like a wish-list, not a request.
• The volume check is a real filter. If the workflow happens fewer than 10 times per month, the answer is usually "not an AI problem — automate with a rule or improve the process."
• Be honest about the data classification. Underestimating it just means you'll be re-classified at triage. Wasted cycle.
• The "explicitly NOT in scope" line in Section 8 is load-bearing. It defines the perimeter the agent must stay inside. Reviewers will check it. The build will be measured against it.
• Champion attaches workflow documentation in Section 13. Without it, triage often punts back to "please document the workflow first."

Common pitfalls

Framework cross-references

• framework.md §12 (intake process specification)
• framework.md §15.1 (registry schema — fields map to intake fields)
• framework.md §10 (risk classification — informs Section 12)
• framework.md §1.1 (workflows-first principle — Section 3 enforces this)
• workflows.md Steps 1–2 (Discovery → Intake)
• workflows.html → In Action view → node M7 (Intake + triage + risk)