Billing Data Analytics

What looks simple is almost always the hardest to actually build. Usage billing looks simple: aggregate events, fetch a price plan, multiply usage by price. The reality is an order of magnitude more complex. Every AI company scaling API revenue -- OpenAI, Anthropic and others -- is building some version of this right now. Usage metering for LLM tokens, compute time, and API calls is a genuinely hard infrastructure problem, especially at the speed and scale these companies are growing. At Cloudflare, a single customer can be billed for 50+ different metrics: Workers requests, Workers AI inference, KV operations (read, write, delete, etc.), R2 storage, Durable Objects and more. Each metric has its own unit of measure, its own aggregation logic and its own free tier threshold. This is very different from a traditional SaaS product that counts API calls or seats —- where you have one unit and one aggregation method. Usage billing is a fundamentally different class of problem. As we've been rebuilding our metering and aggregation pipeline, a few things stuck with me: 1. Billing has to sift through a ton of data. Our edge network generates hundreds of millions of events per second. Each event has dozens of fields. But billing only needs a few: datetime, account id, metric name, value. The key insight here is to aggregate before storage. 2. Your analytics database is likely wrong. If you don't read from the billing source of truth, chances are you are using different aggregation logic, different timing windows, different free tier calculations. Very easy to get vastly different numbers. 3. Calendaring is hard. A customer who signs up January 14th has a billing cycle of Jan 14–Feb 13. Every customer has a different anniversary date. What happens when the anniversary is the 31st and the month is February? Aggregating usage by calendar month is trivial. What happens when you need to aggregate by per-customer billing cycles with month-boundary edge cases? 4. Plan changes are where billing systems break. A customer upgrades from Free to Pro on day 12. Their included Workers requests jump from 100,000 to 10 million. Do you apply the new allocation retroactively to day 1 or from day 12? Prorate the subscription to 18 days? What about a mid-year annual-to-monthly downgrade? 5. Two-stage aggregation is critical to get right. You can't partition by account id at the first stage —- some accounts generate 10,000x more events than others creating hot spots. So we aggregate first without partitioning for even load distribution and then aggregate again with partitioning for final accuracy. These are just some examples. Would love to hear what others in the space are actively thinking through!

💰 Half your Sentinel bill might be tables nobody queries — here's how to see them 💰 Microsoft Sentinel cost is an ingestion story: the tables you fill decide the bill. But most teams can't answer "which tables are growing, and does anyone actually query them?" A new table-insights view in Sentinel gives you that at a glance — and it's one of the highest-leverage things a security architect can look at. Why it matters: - See per-table volume and usage so you can spot the noisy, never-queried tables draining budget. - Make the ingestion-tier decision deliberately — analytics vs basic/auxiliary logs per table, instead of defaulting everything to premium ingestion. - Turn a vague "Sentinel is expensive" into a concrete, defensible data-management plan. The architecture habit: treat SIEM data like any other cost surface — measure what you ingest, tier it to how you query it, and revisit before the true-up. Security and FinOps meet right here. 📌 Plan Microsoft Sentinel costs & billing — https://lnkd.in/e98nYTpz 📌 What is Microsoft Sentinel? — https://lnkd.in/efMDUQ4y 📌 Sentinel data source schema reference — https://lnkd.in/eRS7u45S What % of your Sentinel bill do you reckon is tables nobody actually queries? Have you ever audited it — and what did you cut? #Azure #AzureTipOfTheDay #AzureMissionCritical #MicrosoftSentinel #Security #FinOps

Anthropic just released an Enterprise Analytics API. Nine endpoints. Per-user cost attribution. Full engagement tracking across Claude, Claude Code, and Cowork. Here's why this is a bigger deal than it sounds. Until now, you could track Anthropic spend by workspace or API key. Team-level attribution at best. But the question every CFO actually asks is: "Who on our team is driving this bill?" Now you can answer that. Per-user token usage. Per-user USD spend. Broken down by model, context window, inference region, and speed. Named users with email addresses, not anonymous API key IDs. The timing isn't accidental. Anthropic dropped the 2x pricing premium on their 1M context window three days before this. They removed the financial friction — and immediately gave you the visibility to track what happens next. This is the pattern we keep seeing at Finout: AI providers are building the billing infrastructure that cloud providers took a decade to get right. Usage APIs, cost APIs, per-user analytics. It's happening in months, not years. The enterprises that treat AI spend like any other infrastructure line item — with the same attribution, anomaly detection, and showback discipline — will be the ones that scale AI without the budget surprise. The ones that wait for month-end invoices will learn the hard way. Are you tracking AI spend per user today, or still looking at the total bill? Read the full article: https://lnkd.in/e3x7H4Zn #FinOps #AIInfrastructure #LLMOps #GenAI #CloudCost

Two analysts. Same hospital. Same question: "How many of our patients have diabetes?" One pulls from claims. One pulls from the EHR. The numbers don't match. Billing data and clinical data answer different questions. They were never designed to agree. Claims data exists to get someone paid. Clinical data exists to treat a patient. Those are not the same job. When two reports show different numbers, the first question isn't: "Which one is wrong?" It's: "Which one is right for the question we're trying to answer?" Here's why Not all healthcare data is collected for the same purpose. For example: • If a piece of data affects billing or reimbursement, it's usually checked very carefully. • If it doesn't affect payment, there may be more room for missing, incomplete, or inconsistent information. Another important thing to remember: A diagnosis code doesn't always mean a patient truly has that condition. It simply means someone coded it during a visit. That's why experienced analysts often look for additional evidence, such as: • Multiple visits with the same diagnosis • Relevant lab results • Prescribed medications • Supporting clinical records The same applies to procedures. What gets billed isn't always a perfect reflection of everything that happened clinically. Some services may be grouped together under a single billing code, which can hide important details. Junior analyst assume the data is the truth. Experienced analysts know that data is more like a story. A story written by people, for a specific purpose, and often for a specific audience. That's why analytics isn't just about finding numbers. It's about understanding where those numbers came from, what they represent, and when they should be used. Your job isn't always to find the "one correct number." Your job is to identify the number that best answers your stakeholder's question—and be able to explain why. That's what separates reporting from real analytics. And that's what makes great analysts so valuable.

Struggling to make sense of Google Cloud Platform (GCP) billing? I’ve put together a step-by-step video tutorial on building an AI-powered Cost Agent that lets you chat with your BigQuery billing data. Using Google's Agent Development Kit (ADK) and MCP Toolbox for Databases, you can ask questions like What’s my total spend for September? and get clear, instant answers—no complex SQL needed. In this guide, I cover: ✅ Setting up GCP billing exports to BigQuery ✅ Defining cost-querying tools with MCP Toolbox for Databases ✅ Building a conversational AI agent with ADK ✅ Live demo of querying costs by month, projects, services This is a transforms how developers and FinOps teams look at cloud cost analysis. Watch the full tutorial and share your thoughts—how are you managing cloud costs today? 🎙️ Youtube Tutorial: https://lnkd.in/dxtQfbKu 📄 Blog: https://lnkd.in/dFPFzqhw 📁 Code: https://lnkd.in/dNqhjuma #GCP #BigQuery #ADK #MCPToolboxforDatabases #FinOps

Could your team answer this today, and how long would it take? "How many of the customers who... - Signed up through our holiday marketing campaign in Q4 - On the annual family plan - paying with a Chase-issued card Had their first renewal attempt declined for ‘insufficient funds’, ? And of those... - How many were successfully recovered within 7 days - Versus how many ended up churning within 30 days?" An organization that can answer these types of questions quickly and accurately can react quickly, which results in happier customers and satisfied executives. At its core, payments optimization is only as good as the join between internal business data and transaction data. PSPs, networks, and banks give you a rich stream of authorization, settlement, and dispute data, but merchants must bring their own customer and product context in order to produce actionable insights that are relevant to their business. To make this data usable, merchants need to normalize a few key dimensions: A. Customer-level data - Unique customer identifier that persists across systems (CRM, billing, payments, support). - Cohort tags: acquisition channel, geography, subscription tier, customer tenure, annual vs monthly - Payment Event stage: Card verification, free-to-paid trial conversion, non-trial conversion, renewal (1 to N) B. Subscription contract data - Plan/tier ID (standardized across billing and payments). - Start and end dates, renewal frequency, billing currency. - Status flags: active, paused, canceled (with standardized cancel reasons). C. Invoice/order-level data - Invoice ID (must be consistently mapped to payment transaction IDs). - Line items (tier, add-ons, discounts, tax). - Net and gross amounts, including refund/credit adjustments. D. Payment transaction data - Transaction ID (gateway/PSP ID). - Decline reason codes (normalized across acquirers/networks). - Payment method type, issuer BIN, network tokenization status. - Success/settlement status, fraud signals, retries, dispute lifecycle. E. Linking logic - A clean key structure: customer_id → subscription_id → invoice_id → transaction_id. - Consistent timestamping (UTC, ISO8601, normalized across systems). - Master data management (e.g., ensuring “Invoice 12345” in billing = “Payment 12345” in PSP). When structured this way, you can slice performance by cohort (e.g., “Trial-to-paid in LatAm declines more often on first attempt”), payment method (e.g., “BIN ranges in Southeast Asia fail more often at renewal”), or customer lifecycle (e.g., “Annual renewals have higher closed account declines than monthly”). Most merchants I meet with can't answer even a fraction of that first question. Not because the data doesn't exist, but because it isn't structured. If this sounds familiar, let's talk.

"Why am I paying for all these dashboards?" (Fair question, CFO. No one uses them.) Last quarter, a $150M healthcare organization had: • 47 Power BI dashboards • 12 monthly reports • Countless Excel exports Leadership spent $200K building them. But when asked "What decision did you make based on this dashboard last month?" Silence. 𝗛𝗲𝗿𝗲'𝘀 𝘁𝗵𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺: Most healthcare companies confuse reporting with analytics. Reporting tells you what happened. Analytics tells you what to do about it. 𝗧𝗵𝗲 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝗰𝗲: 𝗥𝗲𝗽𝗼𝗿𝘁𝗶𝗻𝗴: "Revenue is down 8% this quarter." 𝗔𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀: "Revenue is down 8% because denial rates increased in two specific payer categories. Here are the three billing codes causing it and the expected revenue recovery if we fix them." One is a scoreboard. The other is a playbook. 𝗥𝗲𝗮𝗹 𝗲𝘅𝗮𝗺𝗽𝗹𝗲: We had a client with beautiful dashboards showing patient volume trends. Leadership looked at them weekly. Nodded. Moved on. Then we asked: "Which service lines are most profitable per patient? Which ones are losing money?" Nobody knew. We built ONE analysis that answered that question. Result? They reallocated resources away from two low-margin service lines. Increased capacity in their highest-margin services. Added $1.2M in net margin within 6 months. 𝗧𝗵𝗮𝘁'𝘀 𝗮𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀 𝘁𝗵𝗮𝘁 𝗱𝗿𝗶𝘃𝗲𝘀 𝗽𝗿𝗼𝗳𝗶𝘁. It's not about having more dashboards. It's about answering the questions that unlock revenue. 𝗛𝗼𝘄 𝘁𝗼 𝘀𝗵𝗶𝗳𝘁 𝗳𝗿𝗼𝗺 𝗿𝗲𝗽𝗼𝗿𝘁𝗶𝗻𝗴 𝘁𝗼 𝗮𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀: 𝗦𝘁𝗼𝗽 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗱𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱𝘀 𝗻𝗼𝗯𝗼𝗱𝘆 𝗮𝘀𝗸𝗲𝗱 𝗳𝗼𝗿 Start by asking: "What decision is stuck right now because we don't have the right data?" 𝗙𝗼𝗰𝘂𝘀 𝗼𝗻 𝗽𝗿𝗼𝗳𝗶𝘁 𝗱𝗿𝗶𝘃𝗲𝗿𝘀 Which service lines are most profitable? Where are we losing margin? What's driving revenue growth or decline? 𝗠𝗮𝗸𝗲 𝗶𝘁 𝗮𝗰𝘁𝗶𝗼𝗻𝗮𝗯𝗹𝗲 Every insight should answer: "What should we do differently tomorrow?" 𝗠𝗲𝗮𝘀𝘂𝗿𝗲 𝘁𝗵𝗲 𝗶𝗺𝗽𝗮𝗰𝘁 Did the analysis lead to a decision? Did the decision improve the business? Track it. 𝗧𝗟;𝗗𝗥: Dashboards don't drive profit. Decisions do. Stop measuring everything. Start answering the questions that matter. Analytics isn't about pretty visualizations. It's about identifying where you're leaving money on the table—and fixing it. ♻️ Share this if you've ever built a dashboard that looked great but changed nothing. Follow me for real talk on turning analytics into actual profit in healthcare.

Most BIN owners don't do this one thing and it's costing them thousands, more often tens of thousands or more depending on their processing volume: Routine analysis of the numbers. Many just track net income. Maybe they also break it down by profitability at the merchant level. As processing volume grows, net income grows, and they leave it at that. But if this is your approach, I promise you're leaving money on the table. Much of this lost money is recoverable when you have access to good data and good analytics in place. I am going to break it down by sharing some of the analytical methods I use. First, I start with a high-level analytical review. Track these as a percentage of processing volume and as a percentage of total revenue: ➡️ Total merchant fees billed (this is generally the same as total revenue when you're the BIN owner, but still analyze it as a % of processing volume) ➡️ Interchange cost ➡️ Dues/assessments ➡️ Processing cost ➡️ Sponsorship cost ➡️ Third party costs like gateways, PCI, etc. ➡️ Residual expense The numbers from this vertical analysis will vary slightly based on a variety of factors like card mix, acceptance methods, weight of MCC types within your portfolio, etc. But in general, you shouldn't see big swings from month to month. If you do, it's your signal to dig deeper to find the reason. What do I do when I see swings that seem abnormal? ➡️ If it's in a category like interchange or dues/assessments, I'll break it down by card brand to see if I can narrow it down to which one is causing the swing. Then, I'll compare costs by item code or interchange category within that card brand over several months. ➡️ If it's in the residual expense category, I'll do a profitability analysis at the sales partner level to narrow down which relationship experienced the biggest variance. Perhaps it's explained by a legitimate reason, but it's possible it could have been caused by human error, which we can address once we identify it. ➡️ If it's at the processing level, I break it down for each processor and then compare costs by item code over the past several months. Sometimes we get billed for things that we shouldn't have. Or the billing rate is wrong. Here's what I want you to know: 💡 if something doesn't look right, don't be afraid to ask the party at the other end of what seems like a mistake-- whether it's a card brand, a bank, a processor, etc. I've seen billing errors happen at every level, and you need to have the confidence to ask about them. This is the highest level and perhaps the most simple review. You're looking for patterns in the data to tell you a story. But you shouldn't stop here. Tomorrow, we'll dig into a more detailed level of review. Stay tuned. 🙌

It's been about two years since law firms started adopting AI tools. Even though the results should be resoundingly positive, the feedback, quite frankly, is mixed based on all of the surveys I have read. Firms report concerns around inconsistent internal adoption, varied client permissiveness regarding AI usage, varying accuracy levels, lack of training, lack of clarity around workflow improvement, and little positive change to firm economics. The firms that are winning the AI-enabled law firm race share a common trait: they studied their data before implementing AI. Those that are struggling took the "ready-fire-aim" approach, jumping on the bandwagon after believing the promises of AI salespeople that GenAI is an out-of-the-box panacea. Make no mistake—I may be the BIGGEST fan of AI adoption in the legal industry, but AI must be adopted in a smart and strategic manner. Smart people don't start taking medicine without understanding their underlying condition; the same should be true of how firms adopt AI. Billing data continues to be the key component for guiding an AI strategy. The Critical Challenges: -Transitioning to Value-Based Pricing. Your billing data reveals which matters are actually profitable under the billable hour. This baseline is essential for setting competitive fixed fees that protect margins while meeting client demands for predictability. -Eliminating Training Inefficiencies. Historic matter data shows exactly where junior lawyer time added value versus where it inflated costs. As AI eliminates routine junior tasks, this analysis helps you redesign training programs and staffing models that work. -Shifting from Input to Output Pricing. Years of time entries tell you what it actually costs to deliver specific outcomes. Without this intelligence, you're guessing at project-based pricing—and likely leaving money on the table or overpricing yourself out of opportunities. -Investing in the Right AI Tools. Your billing patterns reveal which practice areas have the highest volume of repetitive, high-margin work—the sweet spot for AI automation. This data drives ROI-focused technology decisions, not vendor promises. Your billing data is a strategic goldmine. It shows you: • Where you're most efficient (and why) • Which clients and matters are actually profitable • What optimal staffing looks like by matter type • How to price competitively in a value-based world The firms winning in 2026 won't just blindly implement AI—they're leveraging years of operational intelligence to transform strategically, not reactively.