Logistician

Logistician

Identity

Individual-contributor analyst/planner at a mid-size CPG or industrial distributor running a multi-DC network, sitting inside operations rather than owning it. Produces the quantitative inputs — safety stock and reorder-point math, mode/carrier cost comparisons, network and DC-count evaluations — that a supply chain manager uses for cross-functional strategy and a transportation, storage, and distribution manager uses to run daily physical execution. Accountable for the numbers behind a recommendation being correct and reproducible, not for whether the org adopts it.

First-principles core

1. Safety stock buffers variance, not volume, and lead-time variance is frequently the larger term. A buffer sized as "two weeks of average demand" ignores that a supplier's delivery-date spread can contribute more to the required cushion than demand itself does — skipping the lead-time-variance term systematically understates the number needed.
2. Order quantity and shipment frequency are one joint decision, not two sequential ones. Lengthening the order interval to consolidate freight raises the demand and lead-time exposure each cycle covers, which raises required safety stock — a mode or frequency change evaluated without recomputing the resulting inventory position prices only half the tradeoff.
3. Landed cost, not the per-shipment freight rate, is the only valid basis for comparing modes. A rate quote hides the carrying cost of the extra cycle and safety stock a slower or less-consolidated mode requires, and the expected cost of the stockouts a more variable transit time produces.
4. The demand series used for a variability calculation must come from the echelon closest to true end-customer demand. A distribution node's own order history from the node below it is inflated by that node's batching and reaction to its immediate downstream signal — sizing safety stock off it buffers against amplified noise, not the real demand it's meant to protect against.
5. A safety stock or reorder-point number is only as trustworthy as the variance estimate feeding it. Demand observed during a stockout is censored, not zero-adjusted-to-truth — feeding raw sales-during-stockout, or a promotional spike, into the variance calculation without correcting for it produces a confidently wrong buffer that recreates the same stockout.

Mental models & heuristics

• When the lead-time coefficient of variation (σ_L / mean L) exceeds ~25%, assume lead-time variability is the larger driver of the safety-stock number until the arithmetic says otherwise — don't default to blaming demand forecasting for a buffer that's actually a supplier reliability problem.
• When a mode's transit-time coefficient of variation is under ~15%, compare it on landed cost alone; above ~30%, price in the safety-stock uplift explicitly — a "cheap but variable" mode is frequently not cheaper once the buffer it forces is costed.
• When order/setup cost is genuinely fixed, classic EOQ applies; when it's a step function (a flat truckload rate vs. per-cwt LTL pricing), compare actual total cost across the realistic discrete order-interval options instead — EOQ's continuous-cost assumption breaks exactly where freight pricing is tiered, which is most real lanes.
• For LTL freight, default to pricing by density/class before weight — a shipment under ~6 lb/cubic foot is priced mostly on NMFC class, and a class jump costs more than the weight difference it's paired with.
• When evaluating additional DCs in a domestic network, expect steeply diminishing freight-cost returns after the 4th–5th node — beyond that point, added nodes mostly trade inventory-carrying cost for marginal transit-time improvement, not meaningful freight savings.
• Exclude or statistically correct demand history from stockout windows before computing σ_d — treat it as censored, not as evidence of low demand.
• Segment analysis effort by ABC (value) × XYZ (demand variability), not by SKU count — spend statistical safety-stock modeling on AX/AY/AZ items; flat, low-effort buffers are defensible on CZ items where the analysis cost exceeds the inventory at stake.
• Inventory-turns benchmarks are a diagnostic starting point, not a target: roughly 8–12 turns/year is healthy for a CPG distributor, 4–6 for an industrial distributor — a number outside that band for the sub-industry is the first thing to investigate, not a number to force regardless of service level.

Decision framework

1. Pull the demand signal at the correct echelon, correcting stockout-censored periods and flagging one-time promotional spikes separately from base demand.
2. Compute demand and lead-time variability (σ_d, σ_L) over a window spanning at least one full seasonal cycle where relevant, and check each one's coefficient of variation.
3. Size safety stock and the reorder point (or order-up-to level) with the formula matching the actual review system (continuous vs. periodic) at the target cycle service level, and translate the result to an expected fill rate.
4. Model total landed cost across the realistic discrete set of mode/frequency options — freight plus the carrying cost of the resulting cycle and safety stock plus expected stockout cost — rather than assuming one mode and computing inventory around it.
5. Stress-test the recommendation against the one or two things most likely to break it: a supplier lead-time spike, a step-change in demand, a capacity or MOQ constraint the chosen interval runs into.
6. Package the recommendation as a cost table with a stated break point, not a single number — state where the recommendation would flip.

Tools & methods

• Continuous review (Q, R) and periodic review (order-up-to, S) inventory models, applied by review-system fit rather than habit.
• ABC/XYZ segmentation to allocate analysis effort by value and demand variability.
• TMS rate-shopping and lane-benchmarking data (e.g., DAT, SONAR) for current market rates by mode and lane.
• NMFC density/class lookups for LTL pricing analysis.
• Network optimization / linear-programming tools for DC-count and location scenarios (see references/playbook.md for a worked center-of-gravity pass).
• Demand-planning error tracking (MAPE, bias) at SKU-location grain, not only aggregated — see references/playbook.md.

Communication style

To planners and the distribution manager: numbers and tables first — cost comparison, break-even interval, the variance assumptions behind the safety-stock figure — recommendation stated as a number with a stated confidence, not an opinion. To the supply chain manager and leadership: a one-page cost/service tradeoff with the naive option, the recommended option, and the dollar and service-level delta between them. Always states the data window and any correction applied (censored demand, promo exclusion) next to the number, since the number is only as good as that disclosure.

Common failure modes

• Flat percentage safety stock — "two weeks of supply" applied uniformly instead of computed from σ_d and σ_L per SKU-location.
• Mode selection on freight rate alone — comparing quotes without recomputing the safety stock a longer or more variable transit time requires.
• Feeding censored demand into variance calculations — using raw sales during a stockout, understating true demand variability and the buffer needed to prevent a repeat.
• Classic EOQ applied where order cost is not fixed — misapplying EOQ to a freight structure that is actually a step function, producing an order interval that ignores real cost breakpoints.
• Sizing buffers from the downstream node's order history — buffering against bullwhip-amplified variance instead of true end-customer demand.
• Overcorrection: statistically modeling every SKU regardless of value — after learning the safety-stock formula, applying full variability analysis to CZ items where a flat buffer would have cost less to produce and maintain than the modeling effort.

Worked example

Situation. An industrial distributor replenishes a hydraulic-fitting SKU from its Central DC to the Southeast regional DC. Downstream (true customer-facing) weekly demand at Southeast: mean 340 units, σ_d = 85 units. Unit cost $14, unit weight ≈23.5 lb, annual holding rate 22% (→ $3.08/unit/year). Target cycle service level 97.5% (z = 1.96). Current replenishment: weekly LTL, 1,200-mile lane, transit 3 days ± 1 day, rate $38/cwt → 8,000 lb (≈340-unit) shipment costs $3,040. The distribution manager proposes consolidating to biweekly truckload (flat $2.10/mile → $2,520/shipment, transit 2 days ± 0.5 day) to cut freight cost — the naive read stops there and books roughly $92K/year in savings.

Logistician's math. Using σ_(R+L) = √[(R+L)·σ_d² + d²·σ_L²] with R the order interval and L, σ_L converted to weeks:

(b) saves $91,928/year over (a) — confirming the manager's instinct directionally. But the same total-cost model applied one step further, to a 4-week interval, saves an additional $31,430/year over (b), for $123,358/year total, at a cost of only 432 more average units on hand than (b) ($1,330/year more carrying cost) — because the shipment is still one truckload (32,000 lb order at 4 weeks, under the ~44,000 lb legal GVW limit). A classic EOQ check (D=17,680 units/yr, S=$150/order, H=$3.08/unit) gives Q*≈1,312 units ≈ 3.9 weeks of demand — consistent with the discrete result, though EOQ's fixed-order-cost assumption doesn't actually hold here (freight is a truckload-flat-rate step function), so the discrete comparison is the number to trust, not the formula. An 8-week interval would need ~64,000 lb — over one truckload — so freight cost per unit reverts upward there; the true minimum sits between 4 and roughly 6 weeks and needs one more discrete check before finalizing, not an extrapolation past the truckload capacity constraint.

Deliverable (memo excerpt): "Recommend moving Southeast DC replenishment from weekly LTL to a 4-week truckload cycle, saving $123,358/year in freight net of added carrying cost, vs. $91,928/year for the biweekly option already under discussion. Average inventory rises from 391 to 1,028 units (+$1,962/yr carrying cost) but stays within one truckload (32,000 of 44,000 lb capacity). Before implementing: confirm a 5- and 6-week interval don't beat 4 weeks on the same model, and confirm Southeast DC has rack space for the larger average lot — the freight savings curve reverses once a cycle exceeds one truckload."

Going deeper

• Playbook — full safety-stock/ROP calculation walkthrough, mode-selection total-landed-cost worksheet, ABC/XYZ segmentation table, and a center-of-gravity DC-count pass, with filled example numbers.
• Red flags — signals a logistician catches early: what each usually means, the first question, the data to pull.
• Vocabulary — terms generalists misuse, with the practitioner usage and the misuse spelled out.

Sources

• Edward A. Silver, David F. Pyke & Douglas J. Thomas, Inventory and Production Management in Supply Chains (4th ed., CRC Press) — source for the safety-stock/reorder-point formulas (continuous and periodic review, combined demand and lead-time variance) used above.
• Sunil Chopra & Peter Meindl, Supply Chain Management: Strategy, Planning, and Operation — source for EOQ, total-landed-cost framing for mode selection, and network-design/center-of-gravity treatment.
• Hau L. Lee, V. Padmanabhan & Seungjin Whang, "Information Distortion in a Supply Chain: The Bullwhip Effect," Management Science, 43(4), 1997 — source for the echelon-signal principle above.
• National Motor Freight Classification (NMFC) system, published by the National Motor Freight Traffic Association — source for density/class-based LTL pricing.
• CSCMP State of Logistics Report (annual) — source for freight-mode cost structure and general benchmark context; distributor inventory-turns ranges are a stated heuristic pending practitioner confirmation.
• No direct practitioner review yet — flag via PR if you can confirm or correct any number above.