How to Build an AI Council & Intake Process (GTM L3)

Why this matters

In the rush to capture AI’s promised efficiency gains, most GTM organizations are inadvertently creating a "Shadow AI" crisis. When there is no clear path to request a tool like Clay for outbound prospecting or Granola for meeting notes, high-performing reps don't wait for permission—they use their personal credit cards.

The cost of this decentralized buying is three-fold:

Security & Privacy Leaks: Sensitive CRM data or customer call transcripts (via Fathom or Otter) being fed into models where the vendor retains training rights.
Redundant Spend: Paying for four different LLM wrappers that all do the same thing, while already paying for HubSpot AI or Salesforce Einstein seats that go unused.
Technical Debt: A fragmented tech stack where data lives in silos, making it impossible to build a cohesive "Golden Path" for your sales process.

Organizations that implement a Level 3 AI Council typically see a 15–20% reduction in SaaS waste within the first six months and a significant decrease in the risk of data breaches associated with unvetted third-party LLMs.

How it works

Moving to a proactive governance model requires shifting from "No" to "How." Here is the tactical breakdown of building your AI intake engine.

Step 1: Assemble the AI Governance Council

The goal is agility, not bureaucracy. Centralize decision-making into a core group of five stakeholders: RevOps, IT/Security, Legal, Finance, and a GTM Leader (VP level).

RevOps acts as the facilitator. You need a recurring 45-minute weekly session. This isn’t a brainstorming meeting; it’s a "Go/No-Go" tribunal. Legal reviews the "Data Training" clauses to ensure your company data isn't being used to train the vendor's models, while Security checks for SOC2 compliance.

Step 2: Launch the Public AI Intake Form

Stop accepting tool requests via Slack or email. Create a standardized form (using Jira Service Management or Typeform) that acts as the single point of entry.

Crucially, the form must ask for the Business Case (ROI) and Data Types Processed. If a requester wants Lindy to automate executive assistant tasks, they must define the manual hours saved. If they can’t explain the specific manual task being automated, the request is flagged and returned. Include a dropdown for "Existing Stack Capabilities"—forcing the requester to explain why Claude or ChatGPT Team isn't sufficient for their needs.

Step 3: Execute Weekly Review & SLA Tracking

To maintain trust, you must move fast. Set a 10-business-day SLA for an initial decision.

Use a Trello or Asana Kanban board to track requests through four stages:

Submitted
Security/Legal Review
Finance/RevOps Check (ROI vs. Integration)
Final Decision

For low-risk tools (e.g., under $5k/year, no PII access), create an "Express Lane" where the RevOps Lead can sign off without a full council vote. This prevents the council from getting bogged down in $20/month seat approvals for a tool like Momentum.io.

Step 4: Build a Centralized AI Tool Inventory

This is your "Source of Truth." Use Airtable or a shared sheet to list every sanctioned tool. This list should be read-only for the entire company.

When a rep thinks they need a new AI transcriber, they check the inventory first. If they see the company already pays for Gong or Fathom, they don't submit a request. This step alone prevents the "duplicate tool" bloat that plagues $50M+ ARR companies.

Step 5: Perform Quarterly Sunset Audits

AI moves too fast for annual reviews. Every quarter, export usage logs from your SSO (like Okta) or the tool's admin panel. Apply the 30% Rule: If a tool has less than 30% Monthly Active Users (MAU) or isn't hitting the ROI targets set in the intake form, it gets sunset. This "ruthless pruning" frees up budget for the next wave of innovation, such as Claude Code or advanced agents.

Tools you need

Intake: Jira Service Management, Typeform, or Google Forms.
Tracking: Trello, Asana, or Monday.com (Kanban view).
Inventory: Airtable or a centralized Notion database.
Connectivity: Zapier/Make to alert the council when a new form is submitted.

KPIs to track

# of Sanctioned Tools: Total count of vetted AI applications.
Decision Velocity: Average days from intake form submission to "Go/No-Go." (Target: <10 days).
SaaS ROI: Total cost of sunsetted/rejected tools versus the efficiency gains of approved tools.

Common pitfalls

The Room is Too Big: Keep the council to the core 5 stakeholders. Inviting every manager leads to "decision by committee" where nothing gets approved.
Ignoring "Implicit AI": Failing to track AI features already bundled in your current stack (e.g., Zoom’s AI Companion). You may already have the solution you're about to pay for.
Lack of an "Express Lane": Treating a $20/mo Chrome extension with the same scrutiny as an enterprise-wide Clay deployment.

When to graduate to the next level

Once your intake process is stable, you’ll graduate to Level 4: AI Sandbox. This is where you move from merely "approving" tools to providing a secure, internal environment where GTM teams can experiment with new LLMs and agents against a mirror of your production data before they ever hit the "Intake" phase.