Operations & Supply Chain Analytics: Forecasting and Inventory for Interview Answers
One store loses sales because shelves are empty by Saturday evening; another locks cash in slow-moving stock that sits for weeks. Forecasting and inventory analytics is the difference between those two realities - not perfect prediction, but disciplined preparation for uncertainty.
- Forecasting estimates future demand; inventory policy decides how much stock to hold, where, and when to replenish.
- The real trade-off is service level versus working capital: higher availability usually needs more inventory unless the system becomes faster or smarter.
- Use the core chain: demand data → forecast → inventory policy → replenishment → service and cash metrics → feedback.
- Important methods: moving average, exponential smoothing, causal regression, ML forecasting, and qualitative judgement for launches or shocks.
- Key inventory levers: reorder point, safety stock, EOQ, ABC classification, postponement, pooling, and lead-time reduction.
- Track both forecast and inventory KPIs: MAPE, bias, fill rate, stockout rate, inventory turns, and days inventory outstanding.
- The interview-winning line: do not optimise forecast accuracy alone; optimise business outcomes under uncertainty.
Big Picture
Forecasting and inventory are one connected system. A forecast that does not change reorder points, safety stock, warehouse allocation or supplier planning is just a spreadsheet; inventory without a forecast is expensive guesswork.
Core Explanation
The big idea is simple: demand is uncertain, supply takes time, and customers do not wait. Forecasting reduces uncertainty; inventory absorbs the uncertainty that remains.
Think of it as two linked questions:
- Forecasting question: What demand do we expect by SKU, location and time period?
- Inventory question: Given uncertainty and lead time, how much should we stock to hit service targets without trapping cash?
1. Forecasting: from naive estimates to demand sensing
A forecast is an estimate of future demand for a product, location and time period. Good forecasting separates four patterns:
- Level: the normal average demand.
- Trend: demand rising or falling over time.
- Seasonality: repeated calendar-linked movements, such as Diwali gifting or summer beverages.
- Noise: random variation that should not be over-explained.
Common methods:
- Moving average: smooths recent demand; useful for stable SKUs.
- Exponential smoothing: gives more weight to recent demand; useful when demand shifts gradually.
- Causal models: use drivers like price, promotions, weather or ad spends.
- ML forecasting: handles many signals across SKUs, stores and external variables.
- Judgemental forecasting: used for launches, disruptions, one-off events and management overrides.
2. Inventory: the buffer between uncertainty and service
Inventory is stock held to meet demand, cover supply lead time and protect service levels. But every unit stocked has a cost: capital, storage, damage, expiry, markdowns and obsolescence.
The most important inventory decisions are:
- Reorder point: the inventory level at which a replenishment order is triggered.
- Safety stock: extra stock kept to protect against demand or supply uncertainty.
- Order quantity: how much to order each time.
- Service level: the target probability or frequency of meeting customer demand without stockout.
- Allocation: where stock should sit - plant, regional warehouse, dark store, retailer or dealer.
3. The inventory policy choice depends on uncertainty
Do not use one policy for every SKU. A fast-moving toothpaste SKU, a new fashion style, a spare machine part and a fresh food item need different logic.
4. ABC analysis: focus effort where it matters
ABC analysis classifies items by value contribution, usually based on annual consumption value. A-items deserve tight forecasting, frequent review and senior attention; C-items need simpler controls.
5. Metrics: what to track in forecasting and inventory
Interviewers like this topic because it connects analytics to measurable business outcomes. Always track at least one forecast accuracy metric, one bias metric, one service metric and one working-capital metric.
Worked Example: Reorder Point and Safety Stock
Suppose a store sells an average of 100 units per day. Supplier lead time is 5 days. Daily demand standard deviation is 20 units. For roughly a 95% cycle service level, use a z-value of 1.65.
Step 1: Demand during lead time = average daily demand × lead time = 100 × 5 = 500 units.
Step 2: Lead-time demand standard deviation = daily standard deviation × square root of lead time = 20 × √5 ≈ 44.7 units.
Step 3: Safety stock = z × lead-time demand standard deviation = 1.65 × 44.7 ≈ 74 units.
Step 4: Reorder point = expected lead-time demand + safety stock = 500 + 74 = 574 units.
So when inventory falls to around 574 units, the store should reorder. The safety stock is not random extra inventory; it is the cost of protecting service against uncertainty.
Zara is a classic apparel example because it reduces the penalty of forecast error through short design-to-store cycles and frequent replenishment. The primary driver is responsiveness, supported by tight store feedback, controlled production and rapid logistics. The strategic lesson: when you cannot forecast fashion perfectly, shorten the system's reaction time.
Definitions
- Forecasting: estimating future demand for a product, market, location and time period using data, judgement or both.
- Inventory: stock held to meet demand, support operations or buffer uncertainty in supply and demand.
- Safety stock: extra inventory held above expected demand to protect against demand variability or supply delays.
- Reorder point: the inventory level at which a replenishment order should be placed.
- Service level: the target probability or frequency of satisfying demand without a stockout.
- Bullwhip effect: demand variability amplifies upstream as orders move from retailers to distributors, manufacturers and suppliers.
Case Study: Asian Paints and Inventory Postponement
Asian Paints shows how postponement, forecasting and distribution discipline can support massive colour variety without forcing every dealer to stock every shade.
Situation: Decorative paints are operationally difficult because customers want huge colour variety, but stocking every finished shade at every dealer would create enormous inventory complexity. Demand varies by geography, season, housing activity and local preferences.
The move: Asian Paints built a system around postponement: stock standard base paints and tint them closer to the point of sale using dealer-level tinting machines. The primary driver is delayed differentiation - colour is added after demand is clearer. Supporting drivers include a strong distribution network, technology-enabled replenishment, dealer relationships, and disciplined SKU planning.
Outcome and lesson: The company can offer wide variety while reducing the need to hold every finished colour everywhere. The lesson for interviews is powerful: inventory analytics is not only about calculating safety stock; it is also about redesigning the supply chain so uncertainty is handled later, faster and cheaper.
So what: Asian Paints wins not because of one clever formula, but because analytics, network design, dealer capability and postponement reinforce each other.
How AI Changes Operations and Supply Chain Analytics: Forecasting and Inventory
AI does not remove uncertainty; it makes the system respond to more signals faster. Three changes matter most in 2026:
- ML demand forecasting at SKU-location level: models can combine sales history, price, promotions, seasonality, holidays, weather, local events and competitor signals. This is especially useful when thousands of SKU-store combinations are impossible to forecast manually.
- Probabilistic forecasting for inventory: instead of one forecast number, AI can estimate a demand range. That helps planners set safety stock based on risk, not gut feel.
- Exception-based planning: AI can flag the SKUs where forecast error, stockout risk or excess inventory risk is unusually high, so managers focus attention where intervention matters.
Load a company annual report, investor presentation and recent news into NotebookLM. Ask: “Identify evidence of forecasting, inventory, warehousing, distribution, stockout risk and working-capital pressure. Generate five interview questions and model answers using fill rate, inventory turns and safety stock logic.”
Use AI carefully. A model trained on normal demand can fail during disruptions, launches or sudden regulatory changes. The manager's job is to combine model output with business judgement.
Interview Relevance
“A retail chain is facing both stockouts on fast-moving items and excess inventory on slow-moving items. How would you use forecasting and inventory analytics to solve this?”
Frame your answer as a trade-off: “I would not blindly reduce inventory. I would improve availability for A-items while releasing cash from slow-moving and low-criticality SKUs.” That sounds managerial, not textbook.
Common Mistake
The mistake: saying “improve forecast accuracy” as if that alone solves inventory. It costs candidates because operations leaders know even accurate forecasts fail if lead times, order quantities, supplier reliability and safety stock rules are wrong. One-line fix: connect forecast improvement to inventory policy, service level and working capital.
What to Revise Next
Forecasting and inventory teach you how analytics improves physical flow. Next, revise how analytics improves people decisions and risk decisions - the logic of segmentation, prediction, bias and action carries across domains.