Definition
Bandwidth Allocation Equity Index (BAEI) measures the ratio of bandwidth satisfaction between the most-served and least-served application within the same SLA tier.
The Measurement Gap
A logistics company's network operations team rolled out AI-driven bandwidth allocation across their wide area network six months ago. The platform observes per-application demand patterns, predicts upcoming demand, and dynamically allocates link capacity to applications based on the predictions. The dashboard reports average allocation efficiency up 23%. Tier-1 application latency is consistently within target. The CIO holds it up as a win.
Then in a quarterly business review, the finance team raises an issue with the month-end close. Their financial consolidation runs were taking longer than they used to. Not failing, just slower: about 40% longer on the heaviest day. The consolidation runs use a network-intensive ETL workflow that pulls data from regional warehouses, and historically the workflow completed within a 6-hour overnight window. Recent runs have been spilling into the next morning.
The network team investigates. The consolidation workflow is classified as Tier-2 in the application catalog. The AI bandwidth allocator has been observing tier-2 applications for months and has learned demand patterns based on what they typically consume. For the consolidation workflow, the typical pattern is bursty: heavy use during the close week, minimal use otherwise. The allocator has learned to give it bandwidth on demand. The problem is that during the close week, several Tier-1 applications (a customer-facing tracking portal, a route optimization service) are running their own demand peaks. The allocator is satisfying Tier-1 demand fully and only partially satisfying Tier-2 demand. The consolidation gets what is left.
Per the application catalog, this is the correct prioritization. Per the SLA contracts, no commitment is violated. Per the operational dashboards, everything is green. Per the finance team, their close is at risk.
The legacy metrics on the network operations dashboard report a healthy system.
Application latency by tier: Tier-1 within target, Tier-2 within target on average.
Bandwidth utilization: high (which is the optimization goal).
Allocation efficiency: up 23%.
SLA compliance: 100%.
QoS policy enforcement: 100%.
Demand satisfaction rate: tracked as an aggregate, reported as healthy.
These metrics report on whether the system is doing what the policy says. They do not report on whether the policy is producing equitable outcomes for applications within the same tier. The consolidation workflow and a tier-2 analytics dashboard are both classified as Tier-2. The dashboard runs steady traffic and gets steady allocation. The consolidation runs bursty traffic and gets allocated whatever is left. Same tier, very different experiences. The aggregate metrics smooth out the inequity.
These legacy metrics were designed for a world where bandwidth allocation was set by static QoS policies that gave applications a configured share of capacity. In that world, equity within a tier was a property of the policy design. If the design gave each Tier-2 application 10% of available bandwidth, equity was guaranteed. With AI-driven allocation, the equity property has to be measured because the policy is no longer setting it; the optimizer is.
There is no metric that tells you whether AI bandwidth allocation is starving applications with unpredictable demand patterns, which means historically quiet applications degrade silently while predicted-heavy applications hoard capacity.
Formula
Bandwidth Allocation Equity Index (BAEI) measures the ratio of bandwidth satisfaction between the most-served and least-served application within the same SLA tier. In plain language: for each application in a tier, calculate how much of its actual demand is being satisfied. Then take the worst-satisfied application's ratio divided by the best-satisfied application's ratio. A value near 1 means everyone in the tier is being treated roughly equally. A value far below 1 means some applications in the tier are starved while others have plenty.
The formula:
BAEI = Satisfaction_min / Satisfaction_max
Where, for each tier:
Satisfaction_tier = Bandwidth_allocated_tier / Bandwidth_needed_tier
And:
- Bandwidth_allocated_tier = the bandwidth allocated to application by the AI allocator during the measurement window (Mbps)
- Bandwidth_needed_tier = the bandwidth the application would consume if not throttled, derived from demand observation (Mbps)
Routing Concentration Index (RCI): both metrics arise from the same underlying pattern (optimizer rewards efficiency over diversity) and tend to move together. Downstream is
Cost Per Anomaly (CPA, GMI-IT-009): starved applications produce operational incidents (extended batch windows, missed business deadlines, user complaints) that drive CPA cost. The misinterpretation scenario: a team reports BAEI = 0.85 (above the equity threshold) and concludes their allocation is equitable. They miss that BAEI is computed across the application set as it existed during the measurement window. If a starved application was deprioritized below Tier-2 or moved off the network entirely, it disappears from the calculation. BAEI improvement that comes from removing the worst-served application rather than from improving allocation is not real improvement. Track the application population alongside BAEI.
Proposed formula, pending validation pilots.