Select the Platform and Finish the Capstone
Choose tools from requirements, sequence proof stages, assign owners, and grade the complete implementation plan.
What you will complete
Complete the platform decision, architecture, rollout, owners, measures, limitations, and capstone acceptance record.
Select by layer, not product label
Evaluate intake and storage, classification and extraction, deterministic validation, duplicate persistence, review experience, orchestration, destination behavior, audit evidence, security, retention, cost, and maintenance. One platform may cover several layers. Do not assume its document-processing label means it owns all of them.
Power Automate can orchestrate while AI Builder or Azure Document Intelligence extracts and Outlook or SharePoint supplies intake. SharePoint can own library metadata, versions, permissions, and approval surfaces without becoming the extraction engine. n8n can orchestrate adapters, validation, review, and handoffs, but the current Flowgrammer starter’s default accepts labelled text and does not prove live OCR or production execution.
Use evidence cells. Mark unsupported requirements unknown. A hard disqualifier outweighs a high weighted score. If required security, residency, review, export, or destination behavior remains unknown, mark the option incomplete rather than declaring a winner.
Sequence proof before permissions
- Approve the business boundary, schema, exceptions, and safety gates.
- Run deterministic rules against labelled fixtures without a vendor account.
- Add classification and extraction for representative files.
- Calibrate confidence and test the human queue.
- Use a mock destination to prove approval and retries.
- Connect one sandbox draft object with least privilege.
- Pilot with review on every non-duplicate and monitor corrections and failures.
Pin platform, model, and API versions in each test. Recheck mutable vendor limits, licensing, residency, retention, and pricing when implementation begins. Course sources dated September 2026 are not a permanent product contract.
Worked example: Cedar & Quay decision
Fictional example: Cedar & Quay compares a manual tracker, n8n, Power Automate with a Microsoft extraction service, and a separate document-AI adapter. It records whether each option can accept its files, map the owned schema, show source context, enforce review, persist duplicates, create only a draft, retry safely, and produce audit evidence.
The team does not award a winner because it already owns Microsoft 365 or because a vendor advertises high accuracy. It first runs the shared corpus and resolves security and destination questions. If the intake remains small and one stable template covers the work, the manual tracker and deterministic checks may remain the best first release.
No course evidence establishes vendor superiority, Canadian data residency, regulatory compliance, production OCR performance, or a business outcome.
Complete Workbook Section 7 and the Capstone
- Map each requirement to a proposed layer, owner, evidence, and gap.
- Score options only where a current source or dated test exists.
- Write proof stages, rollback conditions, monitoring, and change control.
- Attach Sections 1–6 and grade every critical safety gate.
- Choose proceed to discovery, run more tests, repair, narrow, keep manual, or stop.
Capstone acceptance
The critical gates all pass: one bounded family; explicit unknown handling; no invalid approval write; duplicates stop; decisions bind to versions; only the allowed non-final destination exists; ambiguous retries cannot create a second record; and evidence claims remain proportional.
Then score scope, schema, controls, review, destination recovery, tests, platform plan, and usability from zero to four. Each dimension must earn at least three. A passing capstone supports the next business decision. It does not authorize production deployment, legal or compliance acceptance, or a performance claim.
Choose the next step
If several document workflows still compete or the boundary remains unclear, use an AI Success Audit to compare them. If one workflow, reviewer, test pack, and destination contract are defined, use the capstone to discuss a scoped AI Automation System. Keep the approved manual process until the replacement passes its real environment tests.
Check your application
1. A platform meets most requirements but has no evidence for a mandatory review path. What is its status?
2. What does a passing capstone authorize?
3. When should mutable platform limits be checked?