Skip to content
Bizzi AI Governance Framework
Pillar I: AI Organization · § 02

Governance model (Hub-and-Spoke)

Software companies govern AI in three ways. Centralized routes every decision through one committee. Consistent but slow. Federated lets every squad decide for itself. Fast but fragmented. Hub-and-Spoke puts the standards in a central body and pushes execution to the squads. Pure federation fragments our security surface. Pure centralization stalls our product squads. We run Hub-and-Spoke.

Context

Section titled “Context”

Bizzi’s product organization runs on multiple squads. Each squad ships AI features on its own cadence. A purely federated model ships faster in the short term. Then every squad invents its own evaluation method, its own PII redaction rules, and its own kill-switch. The security surface becomes uneven and the next ISO audit becomes a full rewrite. A purely centralized model forces every prompt change through a single review queue. A market with monthly iteration leaves us behind. Hub-and-Spoke is the answer we land on.

Tier 1
AI Governance Board
CEO · CPTO · Legal Director
Tier 2
AI Center of Excellence
AI Architects · Data Scientists · Security
  1. Tier 3
    Steward
    Product squad
    Product engineers
  2. Tier 3
    Steward
    Data squad
    Data engineers
  3. Tier 3
    Steward
    Platform squad
    QA / Security
Hub-and-Spoke governance

The three tiers

Section titled “The three tiers”
  • AI Governance Board. The top body. CEO, CPTO, Legal Director. Meets quarterly. Approves high-impact initiatives. Sets risk appetite. Owns major changes to BAGF.
  • AI Center of Excellence (CoE). The central engineering team. AI Architects, Data Scientists, Security Engineers. Owns the shared MLOps platform, the evaluation suite, the AI Gateway, and the technical standards every squad applies.
  • Data/AI Stewards. A CoE representative embedded in each squad. Confirms BAGF is applied before release. The two-way channel between Hub and Spoke. Escalates squad realities up. Pushes new standards down.

Why not pure federation

Section titled “Why not pure federation”

Pure federation works when every squad already has deep AI expertise and self-assesses risk. Bizzi is still building that bench. Our AI depth sits in the CoE. Federate now and we end up with three different PII redaction implementations, three different evaluation thresholds, and three different kill-switch designs. An attacker only needs to find the weakest one.

Why not pure centralization

Section titled “Why not pure centralization”

Pure centralization assumes the center has time. It does not. Squads iterate on prompts, vector schemas, and RAG corpora every week. If every change needs a central approval ticket, the queue becomes the product roadmap. Hub-and-Spoke keeps the standards centralized and lets the squad iterate within them.

Standards mapping

Section titled “Standards mapping”