Saturday’s discussion established a truth that is not always acknowledged fully: focus is the most expensive unit of economics in your business. When focus is fractured, the "Open Tab Tax" accumulates. In mid-sized enterprises, research shows that this tax is approximately $15,000 per employee annually. The cost comes from a variety of factors, such as context switching, fragmented data, confusing coverage of the data. It comes from generating t he operational "glue" required to hold broken processes together.
Monday is for execution, following on from Saturday's strategy session. Today, we move from the theory of focus to the practice of tracking a ledger of AI spend.
Microsoft’s recent launch of seven MAI models, including MAI-Thinking-1, is a move from "AI at any cost" to "AI at the right cost." This new variety is useful, but it shifts the burden of responsibility. Routing decisions, evaluation criteria, and cost controls are now your problem, not your vendor's.
Model choice is a policy decision. It is not a developer preference. Before you switch models to chase lower invoices, you must audit your foundations.
The 5-Point Model Migration Checklist
Swapping models changes the invoice. It does not change the outcome if your data foundations are weak. Before you move, verify these five areas:
- Evaluation criteria: Are your metrics tied to specific business outcomes or just technical benchmarks?
- Cost-per-task benchmarks: Do you know the exact cost of a single successful output?
- Audit logs and access controls: Can you trace who is using what, where, and why?
- Data quality gates: Are you feeding your models "trash" that results in expensive, hallucinated "rubbish"?
- Exit plan: How quickly can you pivot if the new model’s safety controls or data residency policies change?
Data Fluency and the 8Cs of Quality
High-performance AI requires high-performance data. We call this data fluency: the ability to interpret, communicate, and act on data effectively. If your team lacks data fluency, they will likely use a premium model like GPT-5.5 for a task that a smaller, cheaper model could handle.
To prevent this, apply the 8Cs of Data Quality:
- Consistency: Data is uniform across all departments.
- Certainty: The source of truth is verified.
- Coverage: There are no gaps in the required data points.
- Completeness: All necessary fields are populated.
- Currency: The data is up-to-date.
- Commonality: Definitions are shared across the organization.
- Consent: Data usage complies with privacy permissions.
- Compliance: Governance standards are met.
How to Audit Your AI Financial Runway in 20 Minutes
I have built the CDQ Tracker (Cost/Data/Quality) to help you identify architectural leakages and reclaim your focus.
The tracker uses a Granular Log Simulator to highlight where optimization fails or succeeds. Deploying this audit takes less than twenty minutes.
Step 1: Clone the Tracker
The master document is locked to protect the core formulas. Click File in the top left menu of Google Sheets and select Make a copy. Save it directly into your secure corporate Google Drive.
Step 2: Establish Your Departmental Baselines
Navigate to Tab 1: Executive FinOps Dashboard. Input the number of active AI users in your departments (Engineering, Support, Marketing, Operations) along with their current monthly target budgets. The sheet calculates your real-time spend limits automatically.
Step 3: Input Token Volumes
In Tab 3: Granular Log, input data from your enterprise API developer console or model billing statements. Enter the user ID, the task performed, the input tokens, and the output tokens. If you use a programmatic API gateway, these token counts stream into this ledger automatically.
Step 4: Analyse Routing Anomalies
Look at the Audit Validation Flag column on the far right of the audit log. The sheet uses lookup formulas mapped to your rate card to identify waste:
- ✅ OPTIMISED: The task matched the cheapest appropriate model tier.
- 🚨 MISMATCH: An employee used an expensive, premium model for a basic utility task, burning capital unnecessarily.
- ⚠️ CONTEXT WINDOW BLOAT: An employee is keeping a single chat thread open too long. This forces the model to re-read thousands of historical tokens on every new turn, which is a primary driver of the "Open Tab Tax."
Step 5: Tighten Spending Gates
Review the Departmental Waste Audit Matrix weekly. If a department shows high mismatch rates or negative ROI, use these hard data metrics to configure automated spending caps and model-routing limits inside your enterprise console.
The Metrics That Matter
To manage AI as a hard asset, you must track these four metrics:
- Total AI Spend (MTD): Your gross investment to date.
- Blended Cost Per Query (CPQ): The average cost of an interaction across all models.
- Context Window Waste: The financial cost of "forgotten" tabs and over-long threads.
- Trivial Model Mismatch Rate: The percentage of tasks sent to a high-cost model that required low-cost logic.
Take Action This Morning
The bottleneck is rarely the model. The bottleneck is the data foundation and the lack of governance.
If you are feeling the pressure of vendor sprawl or the complexity of Microsoft’s new multi-model landscape, I can help. I offer Surgery Hours to help you set up a lightweight model evaluation and governance pack. This provides enough structure to make informed decisions without building a stifling bureaucracy.
Book a Surgery Hour session here to stop the leakage and reclaim your focus.
