Effective date: 2 August 2026
SpendTensor uses automated logic and, in some features, AI models to turn raw provider usage into cost figures, forecasts, and savings recommendations. This page explains the limits of that output so you can use it responsibly.
1. Where automation and AI are used
Automated processing appears in the following places in the product.
- Cost calculation — token counts reported by each provider are multiplied by our maintained price table for that model, so figures are estimates, not invoices.
- Forecasts and trends — projections extrapolate recent usage and will be wrong when your workload changes.
- Savings recommendations — heuristics and model-assisted analysis suggest changes such as model swaps, caching, or batching, with an estimated impact.
- Anomaly and budget alerts — thresholds and statistical rules on daily aggregates.
2. Estimates are not your invoice of record
Provider pricing changes, discounts, committed-use agreements, enterprise rates, free tiers, credits, taxes, and reporting lags all cause differences between SpendTensor figures and your provider bill. Always reconcile against the invoice issued by OpenAI, Anthropic, Google, Microsoft Azure, AWS, or your reseller before making financial commitments or filings.
3. Recommendations require human judgement
Recommendations are suggestions, not instructions. Model swaps, prompt changes, caching, and routing changes can affect output quality, latency, safety behaviour, and compliance posture in ways we cannot see. Evaluate changes against your own quality benchmarks in a non-production environment before rolling them out, and keep a human owner accountable for the decision.
AI-assisted output can be incomplete or incorrect ("hallucinated"). Do not treat it as financial, tax, legal, security, or engineering advice.
4. Your data and model training
We do not use your prompt or completion content to train models — SpendTensor ingests usage and cost metadata (model, token counts, request counts, timestamps), not conversation content. Where a feature sends metadata to a third-party model provider to generate a recommendation, it is sent only for that request and is not used by us for training.
5. Availability of automated features
AI-assisted features depend on upstream model providers. They may be delayed, degraded, or temporarily unavailable, and we may change or retire a specific automated feature. Core visibility and reporting do not depend on AI-assisted analysis.
6. No warranty on automated output
To the maximum extent permitted by law, automated and AI-assisted output is provided "as is" without warranty of accuracy, completeness, or fitness for a particular purpose. Liability is limited as set out in the Terms of Service. Questions: cyberprosoftware@gmail.com.
This document is maintained by SpendTensor, Inc. and is provided for information only. It is not legal advice and is not an independent certification. Questions: cyberprosoftware@gmail.com.