Inventory control models

Inventory control models: EOQ, EPQ, shortages, quantity discounts, (Q,r) and (P,S) policies, safety stock, reorder point, service levels, and single-period (newsvendor) decisions.

Drafted with Aria, reviewed by the AiCanCode.org team · Updated . How we write and check our notes. Spotted an error? Use Give Feedback at the bottom of the page.

Why it matters

Inventory ties up cash and decides whether a plant ships on time or stops for want of a part. Good models set when and how much to order so that service levels are met at the minimum total cost, even when demand and lead time vary.

Key ideas

  • Inventory: stock of materials, WIP, or finished goods held to decouple supply and demand. Measured in units (e.g., pieces) and valued in currency (₹).
  • Demand rate (d or D): how fast items are required. D typically denotes annual demand (units/year); d is per-time demand (e.g., units/day). Sign convention: demand is positive and reduces on-hand inventory.
  • Lead time (L): time between placing an order and receiving it. May be constant (deterministic) or random (stochastic).
  • Ordering/setup cost (K): fixed cost per replenishment order or production setup (₹/order or ₹/setup), independent of quantity.
  • Holding/carrying cost (h): cost to carry one unit for one time unit (₹/unit·year). Often computed as i·c where i is annual holding rate (per year) and c is unit purchase price (₹/unit).
  • Shortage/stockout: when demand cannot be met immediately. Two cases:
    • Backordering: unmet demand is queued and filled later (inventory position can go negative).
    • Lost sales: unmet demand is lost permanently.
  • Shortage cost (c_s): penalty for one unit out of stock per unit time (₹/unit·year), or per unit short (₹/unit) in single-period models. Always specify which.
  • Review systems:
    • Continuous review (Q, r): check inventory continuously; when inventory position hits reorder point (r), order a fixed order quantity (Q).
    • Periodic review (P, S): review every P time units; order up to level S; order size is variable: q = S − I.
  • Safety stock (SS): extra inventory to protect against variability. Often computed using a service level target and the variability of demand during lead time.
  • Service levels:
    • Cycle service level (CSL): probability of no stockout in a cycle (or during lead time).
    • Fill rate (β service level): fraction of demand satisfied immediately from stock.
  • Deterministic vs stochastic models: deterministic assume constant D and L; stochastic allow random demand, lead time, or both.

Classical deterministic models (assumptions and results)

  • Basic EOQ (Economic Order Quantity)
    • Assumptions: constant and known demand rate D; instantaneous replenishment; zero lead-time variability (L may be positive but constant); no shortages allowed; constant unit price (no discounts); holding cost h and order cost K are constant per year.
    • Result: balances order and holding costs to minimize total annual cost. Key outputs: optimal order quantity Q*, order frequency, cycle time, minimum annual cost, reorder point r = d·L (no safety stock).
    • Why reality differs: procurement lead times vary; demand is uncertain; vendors offer price breaks; there are storage limits and MOQ constraints.
  • EOQ with planned backorders (complete backordering)
    • Assumptions: basic EOQ assumptions, but shortages are allowed and fully backordered; shortage cost c_s (₹/unit·year). The inventory level cycles between positive on-hand and negative backorder.
    • Result: optimal cycle Q* and maximum backorder S*; average on-hand and backorder levels share the cycle in ratios determined by h and c_s.
  • EPQ (Economic Production Quantity)
    • Assumptions: production at finite constant rate p (units/time), demand rate d < p; replenishment is gradual (items arrive as produced); no shortages; K and h constant.
    • Result: optimal lot size increases by factor 1/√(1 − d/p) compared to EOQ; maximum on-hand inventory is reduced by the factor (1 − d/p) due to simultaneous production and demand.
  • EOQ with quantity discounts (price breaks)
    • Assumptions: price per unit decreases at specified quantity breakpoints; other EOQ assumptions hold; holding cost rate i is often a fixed percentage of price.
    • Result: compute candidate Q at each price tier with its h, adjust to the tier’s feasible range, then pick the Q with the lowest total annual cost including purchase cost.

Stochastic control policies (service levels and safety stock)

  • Demand during lead time: If demand per time is random with mean μ_d and standard deviation σ_d, and lead time L is constant, then demand during lead time has mean μ = μ_d·L and standard deviation σ_L = σ_d·√L (assuming independent daily demands). If both demand per time and lead time are random and independent, with mean lead time μ_L and standard deviation σ_Lt, then mean lead-time demand is μ_d·μ_L and variance is σ_d²·μ_L + μ_d²·σ_Lt².
  • Continuous review (Q, r) with Normal demand during lead time
    • Assumptions: lead-time demand is Normal with mean μ_L and standard deviation σ_L; target CSL.
    • Result: reorder point r = μ_L + z·σ_L, where z is the standard Normal variate for the target CSL (one-sided). Safety stock SS = z·σ_L.
    • Fill rate β can be linked via expected shortage per cycle: β = 1 − E[shortage]/Q, with E[shortage] for Normal derived by the standard loss function (see formulas). Choose r to meet CSL or β; then choose Q by EOQ or capacity constraints.
  • Periodic review (P, S)
    • Demand uncertainty accumulates over (L + P). For Normal demand, S = μ_d·(L + P) + z·σ(L+P), with σ(L+P) computed the same way as for lead time but over the longer horizon.
  • Poisson approximation
    • For intermittent demand (low-rate events), lead-time demand can be modeled as Poisson(λ = expected lead-time demand). CSL is met by choosing r so that P{N ≤ r} ≥ CSL using Poisson tables.

Single-period (newsvendor) model

  • Newsvendor
    • Assumptions: one selling season; order once before demand is realized; leftover inventory has a salvage value; unmet demand is lost; costs: overage Co (₹/unit) and underage Cu (₹/unit).
    • Exact statement: Optimal order quantity Q* satisfies F(Q*) = Cu / (Cu + Co), where F is the cumulative distribution of demand. For Normal demand, Q* = μ + z*·σ with z* found from the same critical ratio.

Policy selection and practical notes

  • High-value, erratic items: consider periodic review with higher safety stocks, or newsvendor for true single-season items.
  • Stable, high-volume items: continuous review EOQ/(Q, r) is efficient.
  • Classification tools (policy enablers): ABC analysis (by annual rupee usage), VED (Vital–Essential–Desirable) for spares. These guide which items justify sophisticated control.
  • Real-world frictions: minimum order quantities, batch sizes, capacity constraints, multiple items per order (joint setups), supplier calendars, and transportation economies can all shift the simple EOQ decision. Always re-check assumptions before applying a formula.

Outline of key derivations (what exams test)

  • EOQ: total annual cost TC(Q) = ordering cost + holding cost = K·D/Q + h·Q/2. Differentiate w.r.t Q, set derivative to zero, solve Q* = √(2KD/h). At Q*, ordering cost equals holding cost; minimum TC is √(2KD·h).
  • EPQ: average on-hand inventory over a cycle of size Q is Q·(1 − d/p)/2 because inventory builds at rate (p − d) during production and is consumed at rate d otherwise. Substitute into TC(Q) = K·D/Q + h·Q·(1 − d/p)/2 and minimize for Q*.
  • EOQ with backorders: split the cycle into inventory period (with average on-hand (Q − S)/2) and backorder period (with average backorder S/2). TC(Q, S) = K·D/Q + h·(Q − S)²/(2Q) + c_s·S²/(2Q). Set partial derivatives to zero to get Q* and S* relationships.
  • Continuous review safety stock: with Normal lead-time demand, choose r so that P{X ≤ r} = CSL, hence r = μ_L + z·σ_L; expected shortage per cycle is tied to z by a standard loss function; fill rate follows from E[shortage]/Q.
  • Newsvendor: expected profit is concave in Q; setting derivative to zero produces the critical fractile Cu/(Cu + Co).

Formulas

TC(Q) = K·D/Q + h·Q/2

  • Total annual relevant cost (₹/year) in basic EOQ without discounts or shortages. K: ₹/order; D: units/year; Q: units; h: ₹/(unit·year).

Q* = √(2·K·D / h)

  • EOQ (units). Applies under basic EOQ assumptions (deterministic D, no shortages, constant h, instantaneous replenishment).

n = D / Q*

  • Number of orders per year (1/year). D: units/year; Q*: units.

t_cycle = Q* / D

  • Average cycle length (years). Multiply by working days/year to get days/cycle.

TC_min = √(2·K·D·h)

  • Minimum annual relevant cost (₹/year) at Q*.

r = d·L

  • Reorder point (units) with deterministic demand and constant lead time. d: units/time; L: time.

SS = z·σ_L

  • Safety stock (units) for Normal demand during lead time. z: one-sided standard Normal variate for CSL; σ_L: standard deviation of demand during lead time (units).

r = μ_L + z·σ_L

  • Reorder point (units) with Normal lead-time demand. μ_L: mean demand during lead time (units); σ_L: standard deviation (units).

σ_L = σ_d·√L (if L constant)

  • Standard deviation of demand during lead time (units). σ_d: std dev of demand per unit time (units/√time); L: time. Assumes independent, identically distributed demand per time unit.

Var(lead‑time demand) = σ_d²·μ_Lt + μ_d²·σ_Lt²

  • Variance when both demand per time and lead time are random and independent. μ_Lt: mean lead time (time); σ_Lt: std dev of lead time (time); μ_d: mean demand rate (units/time); σ_d: std dev of demand rate (units/√time). σ = √Var.

EPQ: Q* = √( 2·K·D / ( h·(1 − d/p) ) )

  • Optimal production lot size (units). d: demand rate (units/time); p: production rate (units/time), with p > d.

I_max (EPQ) = Q*·(1 − d/p)

  • Maximum on-hand inventory (units) in EPQ.

EOQ with backorders: Q* = √( 2·K·D·(h + c_s) / (h·c_s) )

  • Optimal cycle size (units) when complete backordering is allowed. c_s: shortage cost per unit per year (₹/(unit·year)).

S* = ( h / (h + c_s) ) · Q*

  • Maximum backorder level (units) in the same model.

Quantity discounts: h = i·c

  • Holding cost per unit per year when holding rate is a fraction i (1/year) of unit price c (₹/unit). Use this h in EOQ by tier.

Newsvendor: F(Q*) = Cu / (Cu + Co)

  • Critical fractile for single-period optimum. Cu: underage cost (₹/unit); Co: overage cost (₹/unit); F: CDF of demand.

Q* (Normal) = μ + z*·σ

  • Newsvendor solution when demand is Normal(μ, σ²); z* satisfies Φ(z*) = Cu/(Cu + Co).

β = 1 − E[shortage]/Q

  • Fill rate (unit service level) in continuous review. E[shortage]: expected units short per cycle; Q: order quantity.

E[shortage] = σ_L·[ φ(z) − z·(1 − Φ(z)) ]

  • Expected shortage per cycle for Normal lead-time demand and reorder point r = μ_L + z·σ_L. φ, Φ: standard Normal pdf and cdf.

Poisson CSL: P{N ≤ r} ≥ CSL, N ~ Poisson(λ = μ_L)

  • Determine r by Poisson tail for intermittent demand; μ_L is expected lead-time demand (units).

S (periodic review) = μ_d·(L + P) + z·σ(L+P)

  • Order-up-to level (units) under periodic review with Normal demand. μ_d: mean demand rate (units/time); L: lead time (time); P: review period (time); σ(L+P): std dev of demand over (L + P).

Worked examples

Example 1 — Direct EOQ with deterministic lead time (no safety stock) Given: annual demand D = 24,000 units/year; ordering cost K = ₹1,200/order; holding cost h = ₹8/(unit·year); lead time L = 9 working days; working days per year = 300.

  1. Optimal order quantity
  • Formula: Q* = √(2·K·D / h)
  • Numbers: Q* = √(2·1,200·24,000 / 8) units = √(57,600,000 / 8) = √(7,200,000) ≈ 2,683.28 units
  • Result: Q* ≈ 2,683 units
  1. Number of orders per year
  • Formula: n = D / Q*
  • Numbers: n = 24,000 / 2,683.28 ≈ 8.94 orders/year
  1. Average cycle time in working days
  • Formula: t_cycle_days = (working days/year) / n
  • Numbers: t_cycle_days = 300 / 8.94 ≈ 33.6 days/cycle
  1. Minimum annual relevant cost (ordering + holding)
  • Formula: TC_min = √(2·K·D·h)
  • Numbers: TC_min = √(2·1,200·24,000·8) = √(460,800,000) ≈ ₹21,466/year
  1. Reorder point (deterministic demand)
  • Formula: r = d·L, where d = D / (working days/year)
  • Numbers: d = 24,000 / 300 = 80 units/day; r = 80·9 = 720 units

Answer: Order Q ≈ 2,683 units every ≈ 34 working days; r = 720 units; minimum relevant annual cost ≈ ₹21,466.

Assumptions: constant demand and lead time; no shortages; instantaneous replenishment.

Example 2 — Quantity discounts with safety stock (continuous review) Given: annual demand D = 12,000 units/year; K = ₹500/order; unit price schedule: ₹50 if Q < 1,000; ₹49 if 1,000 ≤ Q < 2,000; ₹48 if Q ≥ 2,000. Holding rate i = 20% per year of unit price. Lead time L = 10 days. Working days/year = 300. Daily demand standard deviation σ_d = 15 units/day. Target cycle service level CSL = 95% (z ≈ 1.645 for one-sided Normal).

A) Determine the cost-minimizing order quantity considering discounts

  1. Compute h at each price tier
  • h₁ = 0.2·₹50 = ₹10/(unit·year)
  • h₂ = 0.2·₹49 = ₹9.8/(unit·year)
  • h₃ = 0.2·₹48 = ₹9.6/(unit·year)
  1. Compute EOQ per tier and check feasibility
  • Tier 1 (Q < 1,000): Q₁ = √(2·K·D / h₁) = √(2·500·12,000 / 10) = √(12,000,000 / 10) = √1,200,000 ≈ 1,095. Not feasible for this tier; nearest feasible is Q = 999.
  • Tier 2 (1,000 ≤ Q < 2,000): Q₂ = √(12,000,000 / 9.8) ≈ √1,224,489.8 ≈ 1,106.6. Feasible; candidate Q ≈ 1,107.
  • Tier 3 (Q ≥ 2,000): Q₃ = √(12,000,000 / 9.6) = √1,250,000 ≈ 1,118.0. Not feasible; use breakpoint Q = 2,000.
  1. Compute total annual cost for each candidate (includes purchase, order, and holding)
  • Candidate A (Q = 999 at ₹50):
    • Purchase = 12,000·50 = ₹600,000/year
    • Ordering = K·D/Q = 500·12,000 / 999 ≈ ₹6,006.01/year
    • Holding = h·Q/2 = 10·999/2 = ₹4,995.00/year
    • Total A ≈ ₹600,000 + ₹6,006.01 + ₹4,995.00 = ₹611,001.01/year
  • Candidate B (Q = 1,107 at ₹49):
    • Purchase = 12,000·49 = ₹588,000/year
    • Ordering = 500·12,000 / 1,107 ≈ ₹5,420.05/year
    • Holding = 9.8·1,107/2 = 9.8·553.5 ≈ ₹5,424.30/year
    • Total B ≈ ₹588,000 + ₹5,420.05 + ₹5,424.30 = ₹598,844.35/year
  • Candidate C (Q = 2,000 at ₹48):
    • Purchase = 12,000·48 = ₹576,000/year
    • Ordering = 500·12,000 / 2,000 = ₹3,000.00/year
    • Holding = 9.6·2,000/2 = 9.6·1,000 = ₹9,600.00/year
    • Total C = ₹576,000 + ₹3,000 + ₹9,600 = ₹588,600/year

Conclusion: Minimum total annual cost at Q = 2,000 units (price ₹48). This is the optimal lot size.

B) Reorder point with safety stock for CSL = 95%

  1. Compute average daily demand and lead-time demand variability
  • Mean daily demand: d = D / 300 = 12,000 / 300 = 40 units/day
  • Standard deviation during lead time (L constant): σ_L = σ_d·√L = 15·√10 ≈ 15·3.1623 ≈ 47.43 units
  1. Safety stock and reorder point
  • SS = z·σ_L = 1.645·47.43 ≈ 78.03 units
  • r = d·L + SS = 40·10 + 78.03 ≈ 400 + 78.03 = 478.03 units
  • In practice, round up to avoid CSL shortfall: r ≈ 479 units.

Answer: Order Q = 2,000 units at the ₹48 tier; set r ≈ 479 units to achieve CSL ≈ 95%.

Assumptions: i is constant across tiers; demand per day independent and stationary; lead time constant; Normal approximation for safety stock.

Common mistakes

  • Using h as a percentage (e.g., 20) instead of a fraction (0.20), or forgetting that h must be in ₹/(unit·year), not ₹/unit.
  • Mixing calendar days and working days; compute d = D/(working days/year) if lead time is in working days.
  • Ignoring purchase cost when evaluating quantity discounts; the cheapest unit price can dominate even if Q is far from the EOQ.
  • Confusing CSL and fill rate β; setting r from β without converting via expected shortage.
  • Using z for a two-sided confidence instead of one-sided CSL; for 95% CSL, z ≈ 1.645, not 1.96.
  • Treating σ_L = σ_d·L instead of σ_d·√L; variance scales with time, not standard deviation.
  • Applying EOQ when shortages are allowed, without switching to the backordering model (incorrect Q* and missing S*).
  • Using EPQ formula with p ≤ d (violates feasibility); or swapping d and p in 1 − d/p.

For GATE ME

  • Conceptual MCQs on assumptions: when EOQ applies, why quantity discounts change h, when to prefer (Q, r) vs (P, S).
  • Numericals on EOQ, EPQ, and EOQ with shortages: compute Q*, S*, I_max, cycle time, and TC_min.
  • Discount problems: evaluate candidate Q at price breaks, compute total annual cost including purchase cost, and pick the minimum.
  • Safety stock and reorder point: demand Normal during lead time; compute r = μ_L + z·σ_L; sometimes both demand and lead time vary—practise the combined-variance formula.
  • Single-period newsvendor: identify Cu and Co, compute critical fractile, and find z or Q* for Normal demand.
  • Service-level conversions: given β and Q, find r via expected shortage, or vice versa. Practise using the Normal loss function values (tables are often provided or implied).
  • Intermittent demand: Poisson lead-time demand; select r to meet a target CSL via tail probabilities.

Practise checklist:

  • Derive EOQ from first principles without memorizing.
  • Remember which terms change with discounts (h and purchase cost) and which do not (K, D).
  • Be comfortable with z-tables and reading one-sided probabilities.
  • Keep units consistent and carry them through every step.

Quick check

  1. In basic EOQ, what happens to Q* if K is quadrupled and h is unchanged?
  2. For Normal lead-time demand with CSL = 97.5%, what z should you use for safety stock?
  3. In newsvendor, if Cu = ₹30 and Co = ₹10, what is the critical fractile?

Answers: 1) Q* doubles (since Q* ∝ √K). 2) z ≈ 1.96 (one-sided). 3) 0.75 (Cu/(Cu + Co) = 30/40).

Finished this topic? Mark it so your progress, study plan and readiness keep up.

Stuck on something here?