Skip to main contentSkip to footer
+321 123 4567
[email protected]

Black Tyger Strategies

Get In Touch With Us

We bring a personal and effective approach to every project we work on.
AI Architecture & Cost Structure Assessment — Free Assessment
Free assessment | 24 questions across 12 domains
Contact Name
Is there a complete, accurate inventory of all AI tools, models, and subscriptions used across the organisation?
Has shadow AI been assessed — AI tools adopted by teams outside the formal technology approval process?
Does the organisation have full visibility into total AI spend — cloud AI services, foundation model APIs, ML infrastructure, and AI SaaS?
Is AI spend tracked in near-real-time — not just at monthly finance reconciliation?
Has the degree of dependency on each major AI vendor been formally assessed?
Is there a single AI vendor whose removal would make critical AI capabilities unavailable?
Is there a documented framework for deciding whether to build, fine-tune, or subscribe for each AI capability?
Is the build-vs-subscribe framework consistently applied to all new AI capability decisions?
Is the AI cost trajectory — month-on-month growth rate — tracked and reported to technology leadership?
Are AI cost efficiency targets set — cost per inference, cost per active user — and tracked?
Are AI models built on open standards — ONNX, open weights, or vendor-neutral formats — where possible?
Are critical AI models deployable in a self-hosted environment — not solely dependent on vendor API availability?
Is AI compute utilisation tracked — are GPU/TPU resources right-sized to actual workload requirements?
Are spot or preemptible instances used for AI training workloads where interruption is acceptable?
Are ML models and training datasets versioned — can any model be reproduced from its training artefacts?
Is CI/CD for ML models implemented — automated testing and deployment pipelines for model updates?
Are data pipelines feeding AI training and inference reliable — SLA-governed with alerting for failures?
Are training datasets versioned and reproducible — can any model be retrained from the same data snapshot?
Is access to AI models and training infrastructure controlled — only authorised users and systems can invoke models?
Is training data access controlled — no unauthorised access to sensitive training datasets?
Are AI capabilities exposed through well-designed APIs that abstract model implementation from consuming applications?
Are fallback mechanisms implemented for AI service unavailability — graceful degradation rather than failure?
Are AI model performance metrics — accuracy, precision, recall, F1 — monitored in production?
Is data drift and model drift monitored — are alerts triggered when input distributions shift significantly?