Inventory forecasting

Inventory forecasting is the method of estimating future demand and purchase volumes to maintain required product availability without excessive warehousing costs. It consists of concrete formulas, metrics and decision rules for each SKU.

How inventory forecasting works

The inventory forecasting process has three steps: data collection, forecasting model, and translating the forecast into operational parameters (reorder point, order quantity, safety stock). In practice this means:

• collect sales history by SKU and channel (for example, Kaspi.kz sales separately for FBS and separately for FBO);
• account for lead time from supplier to warehouse (for domestic deliveries in Kazakhstan typically 2–10 days, for imports — 30–60 days);
• assess seasonality and promo effects (New Year/December, Nauryz in March, back-to-school in August–September, Kaspi/Black Friday sales);
• convert demand forecast into reorder points using the formula: reorder point = average demand during lead time + safety stock.

Key elements and formulas:

• Average demand (D_avg) — typically the mean over the last 3–12 months depending on SKU stability.
• Lead time (LT) — average calendar or business time from placing an order with the supplier to arrival at the warehouse / into fulfilment.
• Safety stock = z * sigma_LT, where z is the z-value for the desired service level (for example, 1.65 ≈ 95%), and sigma_LT is the standard deviation of demand during LT.
• Reorder point (ROP) = D_avg × LT + safety stock.
• EOQ (economic order quantity) Q = sqrt(2DS/H), where D is annual demand, S is the cost to place an order, and H is annual holding cost per unit.

Why inventory forecasting matters for a Kaspi.kz seller

Inventory forecasting directly affects revenue and cost of goods sold. Specific benefits for Kaspi.kz sellers:

• reduced stock-outs: lost sales on Kaspi not only mean an immediate lost order but also lower conversion and reduced visibility in search results;
• optimized working capital: holding excess stock is expensive — warehousing and financing costs rise; a target turnover for many e‑commerce sellers is 6–12 turns per year (equivalent to 30–60 days of stock);
• lower logistics costs: coordinated ordering reduces the need for express shipments and extra fees for urgency;
• smoother work with fulfilment: if you use FBO or third‑party fulfilment, accurate forecasts let you plan shipments in batches and avoid penalties and overcharges for missed SLA or late replenishment;
• better Buy Box and listing stability: steady availability supports rankings and customer trust on the marketplace.

Practical calculation examples for a Kaspi.kz seller

Below are simple worked examples to apply formulas to real SKUs.

Example 1 — Reorder point with safety stock

Assume a SKU with average daily sales D_avg = 5 units/day, average lead time LT = 10 days. Demand volatility during LT has standard deviation sigma_LT = 8 units. Desired service level 95% → z ≈ 1.65.

Safety stock = z × sigma_LT = 1.65 × 8 ≈ 13.2 → round up to 14 units.

ROP = D_avg × LT + safety stock = 5 × 10 + 14 = 50 + 14 = 64 units. So reorder when on-hand inventory drops to 64 units.

Example 2 — EOQ

Annual demand D = 12,000 units (1,000/month). Ordering cost S = 2,000 KZT per order. Annual holding cost H = 300 KZT per unit.

EOQ Q = sqrt(2DS/H) = sqrt(2 × 12,000 × 2,000 / 300) = sqrt(480,000) ≈ 693 units. Order this quantity to minimize total ordering + holding costs, subject to warehouse space and cash constraints.

Example 3 — Adjusting for promotions and seasonality

If a promotion typically increases demand by 150% over baseline for two weeks, model baseline demand separately and add expected uplift for the promo period. If baseline daily demand is 5 units and promo uplift is +7.5 units/day (150% of 5 = +7.5), then during promotion expected daily demand is 12.5. Increase safety stock before the promotion and plan an extra shipment to fulfilment to avoid stock-outs during high-traffic periods (New Year, Nauryz, Back to School, Kaspi sales).

Forecast quality metrics and target values

Important metrics to track:

• MAPE (mean absolute percentage error) or MAE for accuracy;
• Bias to detect systematic over- or under-forecasting;
• Service level / fill rate and number of stockout days for operational performance.

Reasonable benchmarks for a marketplace in Kazakhstan:

• MAPE 10–30% for stable SKUs;
• Bias within ±10%;
• Service level ≥95% for key SKUs. Adjust targets by category — slow movers tolerate higher errors, key items require tighter control.

Practical tips for implementing forecasting

• start simple: implement basic moving averages and ROP calculations for top-selling SKUs before expanding to advanced models;
• segment SKUs: separate by velocity, lead time, margin and seasonality; apply different models and review cadences;
• integrate data sources: combine Kaspi.kz sales, returns, supplier lead times and warehouse receipts to improve inputs;
• use rolling forecasts: update predictions weekly or daily for fast movers and monthly for slow movers;
• measure the financial impact: compare reduced stock-outs and holding costs after forecasting implementation to justify resource allocation;
• document exceptions and promotions: keep a calendar of planned campaigns and known supply disruptions (customs, public holidays in Kazakhstan) and feed them into the forecast.

Mistakes and risks to avoid

• relying only on long-term historical averages — this ignores recent trends and can cause under- or over-stocking;
• not distinguishing channels (FBO vs FBS) — different channels have different lead times, return rates and demand patterns;
• ignoring promotions and seasonality — failing to bump safety stock or plan extra shipments before big sales on Kaspi leads to stock-outs;
• using a single z-value for all SKUs — critical SKUs may need a higher service level than low-margin items;
• not tracking forecast bias — small persistent under-forecasting quickly amplifies stock-outs and lost visibility on marketplace.

Conclusion and practical advice

Inventory forecasting is a balance between availability and cost. For Kaspi.kz sellers, the practical path is: start with clean, channel-separated sales data; apply simple models for ROP and EOQ; explicitly model promotions and lead times; and track MAPE, bias and service level. Iterate — improve data quality, refine models for priority SKUs, and scale successful rules across the assortment.

Practical short advice: prioritize forecasting for the top 20% of SKUs that generate 80% of revenue — get them right first, then expand coverage.