From Copy-Pasting Old Resolutions to Instant Precedent Matching: How We Automated Board Resolution Drafting

A side-by-side look at what corporate secretarial work looked like before and after deploying an AI agent that finds, adapts, and enforces legal precedents automatically.

The Before: A Process Held Together by Institutional Memory

Monday, 9:15 AM. Your inbox:

"Hi, can you draft a board resolution for the appointment of a new company auditor? We need it by end of day."

Here's what happens next — the old way:

1. Open OneDrive. Navigate to the "Historical Resolutions" folder. It has 347 files across 6 years.
2. Search. Type "auditor appointment." Get 12 results. None of them are named clearly. Open 5 to find the right one.
3. Copy-paste the closest match. Change the date, the names, the company details.
4. Worry. Did the Companies Act clauses change since this precedent was written? Is Section 205 still the right reference? Was there a newer template approved last year that you missed?
5. Email Legal. "Can you just double-check this before I send it to the Board?" Wait 3 hours.
6. Revise. Legal caught a missing mandatory clause. Add it. Re-format. Send.

Total time: 4–6 hours, including wait time.

Total confidence: Moderate at best. You're relying on your memory of which resolution is "the good one" and hoping the legal framework hasn't shifted since it was last used.

The After: What Happens When You Deploy the Resolution Drafting Precedent Finder

Monday, 9:15 AM. Same inbox. Same request.

Here's what happens now:

1. Type a prompt into Copilot Studio: "Draft a board resolution for the appointment of a new company auditor."
2. The agent runs. In under 60 seconds, it:
3. Semantically searches your entire OneDrive for the 3 most relevant past resolutions — not by filename matching, but by meaning
4. Selects the best precedent and adapts it to the current request
5. Passes the draft through a Boilerplate Enforcer that verifies all mandatory legal clauses are present
6. You receive a formatted resolution — complete, verified, ready for the Board Chair's signature.

Total time: Under 5 minutes.

Total confidence: Maximum. The agent used real, previously approved precedents — not generated text. And the Boilerplate Enforcer caught anything the precedent might have been missing.

Get immediate access to the full JSON schema for this workflow.

Unlock the schema →

The Side-by-Side Comparison

Inside the T.A.C.T. Architecture

Trigger: Manual — Prompt-Based Request

The workflow fires when a team member types a natural-language request. Examples:

• "Draft a resolution for appointing a new director"
• "Create a board resolution approving the FY2025 financial statements"
• "I need a resolution for changing the company's registered address"

The manual trigger is intentional here. Resolution drafting is an on-demand activity — you need it when you need it, not on a schedule.

Agent: Resolution Drafting Precedent Finder (Knowledge Type)

The agent is structured as a three-stage pipeline:

1. Precedent Retriever — Semantically searches OneDrive for the 3 most similar past resolutions
2. Drafting Agent — Adapts the best-matched precedent to the current request
3. Boilerplate Enforcer — Ensures mandatory legal clauses are intact and unaltered

Why this architecture matters: The agent doesn't hallucinate legal text. It doesn't generate boilerplate from a language model's probabilistic understanding of "what a board resolution sounds like." It retrieves actual, previously approved corporate resolutions as its ground truth. The drafting agent adapts — it doesn't invent.

Full System Prompt:

You are a Resolution Drafting Precedent Finder. Your workflow:

1. Collect Input Data: Gather all relevant source data, documents, and information.
2. Consolidate & Structure: Organize and standardize the collected data.
3. Analyze & Process: Ensure strict procedural adherence. The AI doesn't hallucinate legal text; it relies entirely on past approved corporate precedents as its ground truth.
4. Validate Results: Review the processed output for accuracy.
5. Distribute Output: Format the final results and share with stakeholders.

Connector: OneDrive for Business

The connector gives the agent direct read access to your entire resolution library. This is the critical difference between "asking AI to write a resolution" (prompting) and "building an AI that drafts from your approved precedents" (orchestrating).

Tool: OneDrive Content Retrieval

Sample Output: The Finished Resolution

BOARD RESOLUTION OF [COMPANY NAME] PTE. LTD.

(Written Resolution pursuant to Section 184A of the Companies Act 1967)

---

APPOINTMENT OF COMPANY AUDITOR

RESOLVED that pursuant to Section 205 of the Companies Act 1967, Messrs. [Audit Firm Name] be and are hereby appointed as Auditors of the Company in place of the retiring Auditors, to hold office until the conclusion of the next Annual General Meeting, at a remuneration to be determined by the Directors.

RESOLVED FURTHER that the Directors be and are hereby authorized to do all acts and things necessary to give effect to the above resolution.

---

Precedent Source: Resolution dated 14 March 2023 (File: Board_Res_Auditor_Appointment_2023.docx)

Boilerplate Verification: ✅ Section 184A reference — Present. ✅ Section 205 reference — Present. ✅ AGM continuation clause — Present.

Generated by: TACT Resolution Precedent Finder | 15 November 2025

What the Transformation Really Means

This isn't just about saving 4 hours. It's about de-risking corporate governance.

In the "Before" world, your resolution quality depended on who was drafting it. A senior secretary with 15 years of institutional memory would find the right precedent instantly. A new hire would struggle for hours and might use the wrong one.

In the "After" world, institutional knowledge is encoded in the system. Every team member — regardless of experience level — gets the same quality output because the agent searches semantically, not by guessing which filename to click.

The Boilerplate Enforcer ensures that whether it's your best day or your worst, mandatory clauses are never missing. The agent is your institutional memory, externalized and automated.