Runbooks and reference material for reducing AI spend, investigating spikes, and making model changes safely.
Topic overview
Start with the most useful StackSpend guides for this problem area, then move into the product workflow that fits what you are trying to do.
AI coding tools, model APIs, cloud GPUs, and AI SaaS add-ons now behave like cloud costs: usage-based, distributed, variable, and hard to explain from invoices alone.
Read articleA practical guide to AI cost anomaly detection for teams using OpenAI, Anthropic, Bedrock, Vertex AI, and Azure OpenAI. Learn which signals matter, how to set thresholds, and how to investigate anomalies without noise.
Read articleA compromised Google Cloud service account ran $12k of Gemini API calls in two hours — and was accelerating toward $50k a day. This is where security and FinOps collide, and why a real-time spend layer is a security control, not just a finance one.
Read articleA practical guide to AI cost observability for teams using OpenAI, Anthropic, Bedrock, Vertex AI, and Azure OpenAI. Learn what to measure, how to structure ownership, and how to turn raw usage data into useful cost decisions.
Read articleLLMOps and LLM FinOps overlap, but they are not the same job. Learn where tracing, prompts, evaluation, spend tracking, and cost controls fit in a modern AI operations stack.
Read articleA practical guide to where StackSpend, PostHog, Langfuse, Helicone, and Lunary fit across LLM FinOps, LLM observability, analytics, and multi-provider AI cost control.
Read articleA practical guide for product teams that need LLM spend tracking by feature, experiment, team, and customer. Learn what to instrument, what to review weekly, and how to connect model decisions to spend.
Read articleA step-by-step runbook for developers and product teams investigating a sudden OpenAI, Anthropic, Bedrock, Vertex AI, or Azure OpenAI spend increase.
Read articleA practical guide to making AI costs explainable. How developers and product teams should structure projects, workspaces, API keys, tags, and metadata to track spend by feature, team, and customer.
Read articleA practical guide for developers, product teams, and engineering leaders who need to track LLM API spend by provider, model, feature, team, and customer before the invoice arrives.
Read articleLower-cost models now handle most production AI tasks reliably. But switching without a process is how products break. Here is a task taxonomy, current 2026 pricing context, and a five-step evaluation framework.
Read articleAI and LLM spend can spike without warning. Learn the five most common causes of AI cost overruns and the monitoring approach that prevents them.
Read articleWhy AI-assisted development bills spiral — runaway agents, retry loops, cache-busting context bloat, MCP overhead, the rework tax, pricing rug-pulls, silent model degradation, and shadow-AI sprawl. A research-grounded map of every failure mode, and how engineering teams get control.
Read articleWhen your LLM spend spans OpenAI, Anthropic, and Cursor, visibility fragments. Learn how to consolidate LLM cost tracking across providers and avoid budget surprises.
Read articleYour OpenAI bill isn't high because OpenAI is expensive. It's high because you're paying for usage you didn't see coming—and you're finding out a month too late. Here's what usually causes it and how to fix it.
Read articleA practical OpenAI bill checklist for prompt size, retries, background jobs, embeddings, and model tier changes when spend jumps unexpectedly.
Read articleYour Anthropic bill usually isn't high because Claude is expensive. It's high because output tokens, long contexts, and un-cached prompts compound quietly. Here's how to find the cause and cut it without losing quality.
Read articleVertex AI and Gemini token usage is sitting in your Google Cloud billing export already — no SDK, no proxy, no code change. Here's how StackSpend reads tokens straight from the bill, and why that beats instrumentation.
Read articleCursor reports usage per event — model, user, and for token-based calls the full input/output/cache split. Here's what you can break Cursor spend down by, and the one limitation to know about request-based calls.
Read articleLearn how to monitor AI infrastructure in production with daily spend signals, alert thresholds, cross-provider visibility, and simple review loops.
Read articleA fast health check for AI cost tracking, monitoring, attribution, and optimization hygiene. Identify the first five fixes without a long read.
Read articlePrioritize the engineering tactics that lower AI spend fastest. Prompt compression, caching, smaller models, batching, and retrieval optimization with a clear savings vs effort ranking.
Read articleChoose the right architecture for knowledge and behavior. RAG, fine-tuning, and full-context each win in different scenarios—and hybrids are now the default.
Read articleChoose when to retrieve vs stuff more context. Embeddings and retrieval have different cost shapes than long-context prompting—this guide shows which wins for your workload.
Read articleUse schema-constrained outputs to make extraction, classification, and scoring workflows more reliable than free-form prompting.
Read articleUse LLMs for yes-no and limited-choice decisions only when the decision space, fallback path, and review thresholds are explicit.
Read articleSummaries only work in production when they are shaped for a real decision. Choose operational, executive, or structured summaries on purpose.
Read articleBuild tool-using LLM systems with the lightest orchestration that works: fixed workflows first, planner and executor loops only when the task truly requires them.
Read articleHuman review is not a fallback for bad AI design. Use it deliberately to control risk, protect quality, and keep automation economically sensible.
Read articleBuild a practical evaluation loop for LLM systems using task-specific metrics, regression sets, and release gates instead of ad hoc spot checks.
Read articleProduction chat is a workflow system, not just a prompt. Design memory, retrieval, containment, and human handoff together.
Read articleKeep LLM workflows inside acceptable risk boundaries with layered policy checks, review routes, and confidence-based gating.
Read articleThe highest-leverage AI architecture choice is often not using an LLM at all. Use this guide to reject bad LLM candidates early.
Read articleAI unit economics only matter when AI cost is a direct input to revenue. Internal tooling? Skip the complexity. Charge for AI? You need it. Here's when to bother, what to measure, and how to start.
Read articleAgentic systems create cost patterns that fixed budgets never anticipate — loops, retries, and multi-step tool calls. How to monitor and control the financial side of AI agents.
Read articleAI spend spikes fast. Static budgets and monthly reconciliation are too slow. Learn how daily alerts and anomaly detection catch AI cost problems early—before they become expensive.
Read articleClaude Sonnet sits between Haiku and Opus on price — and most teams' spend rides on which model their workflows actually use. How to track Claude Sonnet cost by model, context, and feature.
Read articleTotal AI spend hides your unit economics. Cost per LLM request — and cost per customer and per feature — is what tells you whether an AI product makes money. How to measure it.
Read articleCopilot bills per seat, but the spend story is active vs paid seats and how it sits alongside Actions and Codespaces. How to monitor GitHub Copilot cost across your org.
Read articleGPT-4o is cheaper per token than GPT-4 — but volume hides the real cost. How to track GPT-4o spend by model, project, and feature, and catch the changes that move your OpenAI bill.
Read articleToken-based and usage-based AI pricing is unpredictable. Learn why budgets fail and why forecasts and daily monitoring matter more than static limits.
Read articleThe native OpenAI dashboard shows usage slowly and separately from cost. Here's how to actually monitor OpenAI usage — requests, tokens, and model mix — tied to spend, with alerts.
Read articleEmbeddings look cheap per call — until a backfill or per-event pipeline reprocesses your whole corpus. How embedding cost spikes happen and how to monitor them before the invoice.
Read articleOpenAI bills per token, but tokens are invisible until they're dollars. How to track token cost — input vs output, by model and feature — and catch the changes that drive an OpenAI bill.
Read articleAI spend rarely starts with procurement. It starts with a credit card, an API key, and a trial. Here's how shadow AI spend accumulates — and how to bring it into one governed view.
Read articleThe difference between being "over budget" vs "at risk". Learn about pace, forecast, and days-to-risk in plain language.
Read articleFinOps is mature for cloud and new for AI. What AI FinOps means, how it differs from cloud FinOps, and the inform-optimize-operate loop applied to token-based spend.
Read articleWhen a cost anomaly fires, the next question is always 'what deployed?'. Here's a practical method to correlate a spend spike to the PR or deployment that most likely caused it.
Read articleA spike in your AWS, GCP, or OpenAI bill almost always has a cause you can find. Here's how to root-cause it deploy by deploy and separate real regressions from expected growth.
Read articleDeployment cost correlation connects a cost anomaly to the deployment and pull request that most likely caused it. Here's how source-control cost attribution works and why it matters.
Read articleA complete operating loop for cloud and AI cost incidents: detect the anomaly, correlate it to the deployment that caused it, assign the fix in Jira or Linear, and confirm it stays fixed.
Read articleConnect providers in minutes. Get 90 days of visibility and start receiving daily cost updates before the invoice lands.