Capacity Planning and Optimization

Spend 2 minutes reading this post and I'll give you back my notes on Capacity planning in system design interviews, which took me 12+ months to create. Capacity planning is one of the most overlooked yet critical parts of system design. It’s the difference between a system that scales smoothly and one that crumbles under unexpected load.  - In interviews, candidates often throw out random numbers.   - In real-world engineering, inaccurate estimates can cause outages, cost overruns, and poor performance.  Let’s break down how to approach capacity planning properly, with real insights from large-scale distributed systems.  ► Capacity Planning in Interviews: The Checklist  You don’t need exact numbers, but you do need a thought process. Here’s what a structured answer looks like:  1️⃣ Estimate Traffic & Workload      - Number of users per day/month/year      - Requests per second (RPS) at peak load      - Read vs. write ratio      - Data growth over time  2️⃣ Estimate Storage Requirements      - How much data each user generates      - How frequently it needs to be stored      - What kind of storage (SQL, NoSQL, object storage)  3️⃣ Compute & Memory Requirements      - How much CPU is required for each request?      - How much RAM do we need for caching?      - Can we optimize with compression?  4️⃣ Network & Bandwidth Needs      - How much data transfer happens per request?      - Do we need CDNs or caching layers?  5️⃣ Scaling Strategy      - Do we scale vertically (bigger machines) or horizontally (more machines)?      - When do we auto-scale, and how do we handle failovers?  6️⃣ Failure Scenarios & Contingency Planning      - What happens when a database node fails?      - How do we handle spikes in traffic (Black Friday problem)?      - How do we ensure high availability?  This is what interviewers want to see, not memorized numbers, but structured problem-solving.  ► Capacity Planning in the Real World: What Actually Happens  1. You’re Not Working With Theoretical Numbers, — You’re Working With Live Data  - In real-world systems, capacity planning is an ongoing process, not a one-time calculation.  - Engineers constantly monitor metrics (latency, error rates, disk utilization) to adjust resources dynamically.  2. Capacity Planning is Business-Driven  - Your system doesn’t just scale infinitely, there are cost constraints.  - You work with finance teams to optimize cloud costs instead of over-provisioning servers.   - Example: Netflix doesn’t just store all videos forever; they tier storage based on popularity. 

From Siloed Projections to System-Wide Planning: How We Built Singapore’s Healthcare Capacity Framework 3 years ago, our healthcare demand projections were done in silos. Today, we have a coherent, system-wide framework that links demand to infrastructure, manpower, and budget planning. Honoured by the recognition on the work done by the team. Here’s the transformation journey. The Challenge We Faced Demand for each care setting is projected independently, using different assumptions and methodologies. 2023: Building the Foundation Introduced more granular inputs: added parameters e.g. functional impairment levels and family support in long-term care projections. Linked patient flows: Connected across settings (e.g. ED visits to acute inpatient to community hospital). 2024: Achieving System Coherence The coordination challenge: Working across 8+ divisions (IPP, HSD, PCC, APO, MP&S, HF) while handling new policy simulations & evolving capacity decisions. The solution: Set up Capacity Planning Committee (CPC) as single decision platform, replacing piecemeal EXCO discussions. The breakthrough: Obtained approval for our projections alignment framework: • Single baseline model across all projections • Common parameters where models intersect • Systematic accounting for care transformation impacts Real impact: Secured approval for new hospital beds through white space activation and new hospital sites. 2025: Advanced System Modelling Healthier SG simulation: Collaborated with Duke-NUS to quantify HSG’s long-term impact on healthcare demand and costs - answering our persistent questions. Disease-based projections: Piloted new method for mental health services, endorsed and used for service planning Tight deadline delivery: Completed baseline and care transformation projections across all settings that should have taken a few years to complete within one year. The Framework That Changed Everything Our Long-Term Capacity Planning Framework now seamlessly connects: • Demand drivers (population aging, functional impairment) • Care settings (from acute to community to home-based care) • Resource planning (manpower, infrastructure, budget) Policy interventions like HSG, right-siting efforts, and palliative care strategies are incorporated. Key Lessons Learned 1. Coordination is as important as methodology - The CPC structure solved more problems than technical improvements alone 2. Resilience matters - When our HSG model wasn’t endorsed initially, we went back to fundamentals and rebuilt stakeholder confidence 3. Granular parameters drive better insights - Moving from broad assumptions to specific factors like family support levels improved accuracy The result? A coherent planning system that helps Singapore prepare for demographic transitions while optimising resource allocation across the entire healthcare continuum. What challenges are you facing in system-wide planning and coordination across multiple stakeholders?

Resource planning separates successful firms from those constantly scrambling to meet deadlines 📊 Most finance teams operate in reactive mode, putting out fires instead of preventing them. I've worked with dozens of clients who struggle with this exact problem. They're always stressed, always behind, and wondering why profitability suffers despite working harder than ever. ➡️ CAPACITY PLANNING FOUNDATION You know what I've learned after years of helping firms optimize their resources? It all starts with forecasting your hours correctly. See, when you can predict workload based on historical data and upcoming client needs, you avoid that feast or famine cycle that absolutely crushes profitability. Monthly recurring revenue clients need consistent attention too. Don't make the mistake I see so many firms make by forgetting about them during busy season. Client volume scaling requires a completely different approach. Growing your client base means different staffing patterns and retention strategies. Plan resources based on both current clients and realistic growth projections. ➡️ BUDGET VS ACTUALS Track your planned versus actual resource utilization religiously. Variance patterns tell you exactly where your assumptions are off. Sometimes it's scope creep eating up resources. Sometimes it's inefficient processes slowing everyone down. Sometimes it's just unrealistic estimates from the start. Your resource planning gets better when you learn from what actually happened versus what you expected. Create accountability across your team so everyone understands how their work impacts overall capacity. ➡️ TIME TRACKING Without accurate time data, resource planning becomes pure guesswork. Monitor your billable versus non-billable ratios to understand true capacity. That administrative time still consumes resources and needs planning. Track project profitability in real-time so you can course-correct before it's too late. Waiting until project completion to assess profitability costs money. Use time data to identify productivity bottlenecks. Maybe certain work takes longer than expected, or specific team members need additional training. ➡️ STANDARD OPERATING PROCEDURES Document your repeatable processes and workflows. This dramatically reduces training time for new team members. Consistent processes mean more predictable resource requirements. When everyone follows the same approach, you can actually forecast capacity accurately. ➡️ CLIENT SCOPE DEFINITION Clearly define project boundaries upfront. Scope creep destroys resource planning faster than anything else I've seen. Set realistic client expectations from the start and stick to them. When clients want additional work, have a system to price and resource it properly. === Resource planning isn't glamorous work, but it's what separates profitable firms from those working harder for less money. What's your biggest resource planning challenge?

🌟 Day 4 – Forecasting Basics How do you know how many calls, chats, or emails to expect tomorrow? 👉 That’s Forecasting—the foundation of Workforce Management (WFM). At its core: Forecast = History (baseline) + Trend + Seasonality + Event Adjustments + Judgment 🔍 Why Forecasting matters (in plain terms) Without a forecast, everything else is guesswork: Capacity Planning: You can’t know how many people you need. Scheduling: You don’t know which hours need extra coverage. Real-Time Management: You can’t tell if you’re off-track or on-target. Reporting: You can’t measure if the plan was realistic. 🧭 What goes into a good forecast - Baseline History- Start with apples-to-apples data (same channel, same handle type). Use the closest comparable days (e.g., last 6–8 Mondays for next Monday). - Trend - Are volumes growing or shrinking month over month? Apply a gentle up/down adjustment (e.g., +2% MoM). - Seasonality - Intra-week: Mondays heavier than Fridays? Intra-day: 11:00–13:00 peak every day? Keep a pattern profile so you can shape the daily forecast by 15/30-minute intervals. - Events & External Drivers - Holidays, promos, product launches, price changes, outages, weather. Each can add/subtract volume. Use an uplift/deflation percentage based on past, similar events. - Judgment & Business Intel - Talk to Marketing, Product, and Ops. Numbers + context beats numbers alone. 🧪 Mini example (numbers you can follow) Baseline: Last 4 Mondays ≈ 10,000 calls This Monday is a holiday: Past similar holiday = +15% uplift Marketing email scheduled 10:30: Past emails add +8% for 2 hours Day total: 10,000 × 1.15 = 11,500 base for the day Apply short, time-boxed +8% uplift 10:30–12:30 to those intervals only. Shaping by intraday pattern (illustration): If 12% of Monday’s calls typically arrive 11:00–12:00, that hour ≈ 11,500 × 12% = 1,380 calls (then layer the +8% marketing effect inside that window). You now have a time-sliced forecast (by 15/30/60-min intervals), not just a day total—this is what schedulers need. 🎯 How to check if your forecast is any good MAPE (Mean Absolute Percentage Error): Average error size. Bias (Over/Under): Do you consistently over- or under-forecast? Hit Rate: % of intervals within a target error band (e.g., ±10%). Track these by channel and by interval, not just daily totals. A perfect day can still hide ugly peaks. 📌 Takeaway: Forecasting is educated prediction—never perfect, always essential. Get close, shape it by interval, adjust for real-world events, and learn fast from misses. That’s how you keep customers happy, and costs controlled. #WorkforceManagement #WFM #Forecasting #ContactCenter #CustomerExperience #BusinessEfficiency #Scheduling #CapacityPlanning #RTA #OperationsExcellence #Analytics #DataDriven

See how easily you can project monthly volumes, predict your business's revenue patterns with precision and plan your production and budget accordingly. Understanding and calculating the seasonality of your revenue can transform how you manage your financial planning. Why Measure Average Volume Demand? Measuring the average volume demand helps you identify patterns in your demand over different periods. By recognizing these patterns, you can adjust your forecasts and budgets to reflect more accurate expectations, preventing potential issues like overcapacity or underproduction. Steps to Calculate Average Seasonality: 1. Collect Data: Gather historical revenue data for multiple years. 2. Calculate Monthly Averages: Determine the average revenue for each month across the years. 3. Compute Overall Average: Find the overall average revenue across all months and years. 4. Determine Seasonal Indices: Divide each monthly average by the overall average to get the seasonal index for each month. Benefits of Applying Seasonal Indices: • Prevent Overcapacity: By anticipating peak periods, you can manage resources better and avoid production bottlenecks. • Optimize Production: Ensure that production schedules align with demand, reducing waste and improving efficiency. • Enhanced Forecast Accuracy: More precise forecasts lead to better financial planning and decision-making. This technique is not only useful when creating monthly budgets and forecasts, but also when crafting long range plans. When we apply the monthly seasonality to the yearly projection, we are able to achieve a granularity that will show us more clearly other aspects of our plan that we are not able to see from the yearly perspective. The capacity constraint is one example. In this case, I have this insight even years ahead to either increase capacity, improve capacity distribution along the year (if possible) or even plan better the volume production. To help you get started, I've created an Excel template for calculating seasonality. You can download it from the link below and integrate it into your budgeting process. https://buff.ly/44WU3tV

Why hospital capacity is a data problem, Not a space problem. How smarter forecasting unlocks hidden capacity. Hospitals don’t usually run out of beds they run out of visibility. On paper, a facility may show 90% or more occupancy, yet studies in healthcare operations reveal that 10–20% of beds can remain effectively unusable due to discharge delays, staffing mismatches, or poor patient flow coordination. In many systems, up to 30% of hospital costs are tied to inefficiencies in scheduling, admissions, and length-of-stay management. The real bottleneck isn’t square footage it’s fragmented data. When demand surges, leaders often assume they need more infrastructure, when what they actually need is better forecasting. Smarter forecasting transforms hospital capacity from a static number into a dynamic system. By analyzing historical admissions, seasonal illness patterns, emergency department trends, and discharge timelines, predictive models can forecast patient inflow days or even weeks in advance with over 85% accuracy in many advanced health systems. Research shows that improving patient flow processes alone can increase effective capacity by 5–15% equivalent to adding an entire ward without laying a single brick. Even reducing average length of stay by just 0.5 days can free up thousands of bed-days annually in mid-sized hospitals. The future of each hospital capacity isn’t about expansion it’s about optimization. Real-time dashboards, AI-driven bed management, and predictive discharge planning allow hospitals to see capacity before it becomes a crisis. Systems that invest in data-driven operations report shorter wait times, lower readmission rates, and millions saved in avoidable costs. Hidden capacity already exists inside most hospitals; it’s simply buried under disconnected spreadsheets and delayed decisions. When data starts leading decisions, space stops being the problem and efficiency becomes the cure. #HospitalCapacity #HealthcareInnovation #DataDrivenHealthcare #SmartHospitals #PredictiveAnalytics #AIInHealthcare #HospitalManagement #PatientFlow #HealthcareOptimization #DigitalHealth #HealthcareEfficiency #MedicalTechnology #HealthTech #OperationalEfficiency #FutureOfHealthcare #RealTimeData #HealthcareLeadership #SmartForecasting #HospitalOperations #BetterPatientCare

I ask every OpEx professional I meet the same question: "What's your current capacity and where's your bottleneck?" About 70% can't answer. I assure you it's a serious problem when I get different answers from people in the same company: planning says one thing, production another, and maintenance something else. They talk about utilization rates or efficiency percentages or "running at capacity." But they can't tell me actual pieces per shift at each step. This tells me they're managing by feel, not data. Here's what strong OpEx leaders have ready: 1) A Process Capacity Sheet: Every process step listed with time breakdowns and capacity calculations. Head forming: 45.5 seconds total time = 633 pieces/shift Threads: 21.2 seconds total time = 1,358 pieces/shift Deburring: 30.0 seconds total time = 960 pieces/shift Deburring is the bottleneck at 960 pieces. That's where you focus. 2) A Standardized Operation Combination Table: Visual timeline of work elements and their duration. Shows the sequence and timing of every task. Displays where work overlaps and where gaps exist. Helps you redesign work flow based on actual timing, not assumptions. 3) An Operation Analysis Sheet: Physical diagram of equipment, material flow, and operator movement. Shows how far parts travel and where motion happens. Makes waste visible. Rearranging two workstations based on this diagram cut one client's cycle time by 14%. Why this matters: You can't improve what you can't measure. And you can't measure what you haven't documented. These three documents transform opinions into facts. They answer executive questions with numbers instead of estimates. They separate OpEx professionals who talk about improvement from those who deliver it. Build these for your top three processes this month. Update them when processes change. Then watch how differently people respond when you can answer capacity questions with data. 📌 In my Newsletter, I share the OpEx leadership playbooks I wish someone gave me in my 30's, the exact frameworks that get your initiatives funded, your results noticed, and your career accelerated. 👉 To Subscribe: Click "𝗩𝗶𝗲𝘄 𝗺𝘆 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿" just above this post. and Join 12,500+ OpEx leaders receiving it weekly. Yours, Mohammad Elshahat

Busy plants aren’t always productive plants. That’s the fastest way to lose money quietly. Most plants look busy. Most machines look utilized. Most dashboards look green. And yet… output stalls, orders slip, and customers feel it first. This visual explains why. Through my experience, I’ve learned a hard truth: Throughput is not the sum of efficiencies—it is controlled by one constraint. What this bottleneck analysis really shows 1️⃣ Capacity Upstream ≠ Throughput Downstream You can widen capacity everywhere: - Faster suppliers - Bigger supermarkets - Higher utilization in Process A None of it matters if one step produces slower than takt. The hourglass doesn’t lie. 2️⃣ Takt Time Is the Customer’s Voice Takt time is not an internal target. It’s the market pulling on your system. When any process: Has capacity < takt Suffers instability or downtime It becomes the constraint—whether you label it or not. 3️⃣ The Bottleneck Is the Revenue Gate Every minute lost at the bottleneck is: - Lost throughput - Lost sales - Lost trust WIP piles up before it. Starvation happens after it. And leaders often chase symptoms in both directions. 4️⃣ Local Optimization Makes the Constraint Worse Speeding up non-bottlenecks: - Increases inventory - Hides the real problem - Creates false confidence The system doesn’t need more effort. It needs constraint focus. 5️⃣ Flow Stops Where Discipline Stops Downtime, stoppages, queues, and withdrawals don’t happen randomly. They happen when: - Capacity planning ignores variability - Flow decisions aren’t constraint-led Management attention is spread evenly instead of intentionally Why this matters High-performing plants don’t ask: “How do we improve everything?” They ask: “What limits us right now—and how do we protect it?” Because when the constraint flows: - Lead time collapses - WIP stabilizes - Revenue follows The rest of the system naturally falls into line. The best operations don’t chase utilization. They design flow around the constraint. If this resonates, happy to exchange notes on real-world impact and ROI. Curious question to leave you with: In most plants, the bottleneck is known—but not addressed. Is that what you see as well?

The Fab Whisperer: Capacity Planning - From Spreadsheets to Self-Learning Models. Last week we looked at the widening gap between silicon demand and fab capacity — the classic setup for another boom-and-bust cycle. Imbalance is inherent in the market. We try to balance it in the way we plan capacity. For an industry that spends hundreds of billions on CAPEX, capacity planning should be science. Yet it often I see frozen spreadsheets, heroic assumptions, and “best-guess” throughput models that quietly drift from reality. Are we building fabs based on models that no longer represent how fabs actually run? Using the wrong model for the wrong purpose? CAPEX Planning ≠ Fab Daily Operations Planning Capacity — deciding what, when, and where to build. Running Capacity — managing flow, bottlenecks, and daily WIP. CAPEX models are strategic: they test economics, demand scenarios, and sensitivity to capacity detractors. Operational models are tactical: they simulate variability, queueing, and dispatch logic. When fabs try to use the same model for both, they end up with bad investments and bad daily decisions. It’s like using a telescope to check your pulse. Most Common Methods of How We Plan Capacity 1. Static Models (Spreadsheet Economics) Quick and transparent — perfect for early CAPEX justifications. But fixed throughput and yield assumptions age fast. Once products, recipes, or WPH shift, the model collapses. 2. Dynamic Simulations (Discrete-Event or Digital Twins based) Capture queues, PM downtime, and rework loops — essential for operational decision-making. Great for optimizing how to run a fab, not what to build next. Powerful but maintenance-heavy; too often abandoned after the big study. The Next Frontier Not mainstream yet but they point to the future: AI-Driven and Hybrid Models. These models will learn from real time fab data, adapt to product mix, and continuously recalibrate effective capacity. They will bridge the gap between planning and operations — a single living model that never goes stale. The barrier isn’t technology — it’s data discipline and trust. The Real Challenge The biggest risk isn’t model complexity — it’s model decay. Assumptions age. Routings evolve. PM cycles shift. By the time the next CAPEX round starts, you’re planning the future based on a fab that no longer exists. What can we do meanwhile Match the model type to the decision horizon. CAPEX → financial sensitivity and long-term. Operations → flow dynamics, variability control, short term. Treat models as living systems, not one-off projects. Assign ownership for keeping assumptions, routings, and rates current. Benchmark quarterly — compare modeled vs. actual effective capacity. Start building the bridge: integrate AI and fab data into planning cycles today. Are your capacity models describing reality — or nostalgia? #TheFabWhisperer #Semiconductor #FabOperations #CapacityPlanning #DigitalTwin #AI #ManufacturingExcellence #FabModeling