AI Slotting Optimization Specialist Interview Questions

A great answer covers how product location affects travel distance, pick rates, labor costs, and throughput - and explains why not all SKUs should be in the same location.

Cover the Pareto principle (A = top 20% SKUs generating 80% of picks), how products are grouped by velocity, and how each tier maps to proximity to the shipping dock.

Mention pick transaction logs, SKU master data, order history, location hierarchies, replenishment records, and pick-path timestamps.

Describe each picking strategy and explain how slotting must align with the chosen method - e.g., zone picking benefits from co-locating frequently co-ordered SKUs within the same zone.

Weight/ergonomic limits (heavy items at waist height), hazmat storage regulations, cold-chain temperature zones, product dimensions vs. shelf dimensions, and proximity constraints for fragile items.

Define decision variables (SKU-to-location assignments), an objective (minimize total weighted travel distance), and constraints (capacity limits, zone restrictions, product compatibility, labor balance).

Discuss how SKU velocity changes by season, promotion, and trend; describe using Prophet or LightGBM for SKU-level forecasting and how forecasts feed the optimization model's demand weights.

Affinity measures how frequently products are ordered together; compute via market basket analysis (support/confidence/lift) on order-line data, then use it to co-locate correlated SKUs.

Discuss decomposition strategies (slot by zone independently), hierarchical approaches (cluster SKUs first), warm-starting from current solutions, time limits on solvers, and metaheuristics for large instances.

Build a SimPy/AnyLogic model simulating order arrivals, pick-path routing, and congestion; run the current layout vs. proposed layout and compare KPIs like average pick time, bottleneck identification, and throughput.

Cubic utilization measures how well the 3D volume of storage locations is filled; high utilization reduces facility footprint costs, but must be balanced against accessibility and pick speed.

Quantify reductions in average travel distance per pick, labor hours saved, throughput increase, and reduction in mis-picks; translate to dollar savings using labor rate and throughput revenue impact.

Static = fixed assignments reviewed quarterly; dynamic = continuous re-assignment driven by real-time signals. Dynamic maximizes efficiency but increases re-slotting labor and operational disruption.

Encode these as hard constraints in the MIP: assign-location eligibility matrices, incompatible-pair exclusion lists, and ergonomic zone restrictions based on SKU weight classes.

Randomly assign comparable warehouse zones to treatment (new slotting) and control (current slotting); measure pick rate, error rate, and travel time over a defined period; use statistical significance testing.

State = current inventory levels, pending orders, congestion heatmap; Action = reassign SKU to new location; Reward = negative of total pick travel distance minus re-slotting cost; use PPO or SAC for continuous adaptation.

Cover data pipeline (Kafka streaming pick events), feature store, forecasting service, optimization solver service (Gurobi microservice), WMS API integration layer, monitoring dashboard, and re-slotting scheduling with labor-aware cadence.

Formulate as a two-stage problem: network-level inventory allocation (which SKUs in which DCs) + facility-level slotting; use Benders decomposition or Lagrangian relaxation to handle the coupling between stages.

AMRs change the cost structure: travel cost depends on robot battery, congestion, and charging schedules rather than human walking; model robot utilization constraints and co-locate high-velocity items near AMR staging areas.

Use transfer learning from similar warehouse profiles, bootstrap with industry-standard velocity heuristics, implement a Bayesian optimization approach that updates beliefs as data accumulates, and start with conservative re-slotting frequency.

Build a full simulation model calibrated to the real warehouse; inject historical peak-demand distributions; test multiple slotting strategies under identical conditions; measure bottleneck emergence, labor utilization, and order cycle time.

Define a cost function that includes pick-travel savings (benefit of better slots) minus re-slotting labor cost, inventory movement disruption, and risk of transition errors; optimize the cadence using marginal analysis or simulation.

Discuss robust optimization (optimize for worst-case within uncertainty set), stochastic programming (optimize expected outcome across demand scenarios), and rolling-horizon re-optimization that adapts as actual demand is observed.

Model the warehouse as a graph where nodes are locations and edges represent adjacency or travel cost; use GNNs to learn embeddings that capture spatial context, then use these embeddings as features in the optimization or RL agent.

Build a RAG pipeline (LangChain + vector store of optimization reports) that translates natural language queries ('Why was SKU-12345 moved to Zone B?') into structured queries against the optimization model's outputs and audit logs.

Start with exploratory analysis of pick-path data, check for SKU velocity drift, review recent product introductions and discontinuations, examine seasonal pattern shifts, then model and simulate updated slotting assignments.

Present a phased re-slotting plan (prioritize highest-impact SKUs first), calculate break-even timeline showing when travel savings exceed re-slotting labor cost, and offer to pilot in one zone to prove value.

Use historical demand data to build seasonal velocity profiles; implement a pre-season re-slotting algorithm that migrates summer SKUs to prime pick locations 4-6 weeks before peak; include a post-season rollback strategy.

Acknowledge the constraint gap, work with operations to formalize the missing constraint, add it to the optimization formulation, re-run validation simulations, and implement a constraint audit checklist for future deployments.

Add FEFO-aware objective weighting (shortest-expiry items to most accessible locations), integrate lot-level tracking into the data pipeline, encode DEA storage constraints as hard constraints, and add audit trail requirements to the system.

Extrapolate pilot results with scaling factors for zone heterogeneity, calculate labor savings at current wage rates, account for re-slotting labor and transition period inefficiency, model payback period and 3-year NPV.

Assign higher per-minute labor cost to frozen zones, enforce time-based zone access constraints for workers, co-locate highest-velocity frozen SKUs near zone entry points to minimize exposure time, and add safety as a hard constraint.

Model a hybrid warehouse where GTP-eligible SKUs are slotted to pod locations (optimized for robot retrieval) and manual-pick SKUs remain in traditional racking; design a phased migration schedule prioritizing high-velocity items for GTP first.

Implement a confidence-weighted slotting approach where new products get a Bayesian prior based on similar product profiles, with a decay mechanism that transitions to empirical velocity data as pick history accumulates; use an explore-exploit framework.

Treat as a greenfield re-optimization: update the location master, run the full optimization with the expanded layout, plan a phased migration that avoids disrupting current operations, and use the expansion as an opportunity to redesign zone boundaries.

Describe using the CP-SAT or Routing solver, define assignment variables, encode capacity and compatibility constraints, set objective to minimize weighted travel distance, and discuss solver parameters (time limits, workers, search strategies).

Use LangChain with a tool-calling agent that has access to a Python tool for querying the optimization database, a SQL tool for WMS data, and an LLM for natural language interpretation; implement RAG with slotting report embeddings.

Describe DAG structure: extract task (pull pick data from Snowflake), validate task (check data completeness), transform task (compute velocity metrics), optimize task (call Gurobi solver), load task (push slot assignments to WMS via API), notify task (send results to Slack/Teams).

Define OpenAI functions for 'get_sku_velocity', 'run_slotting_optimization', 'compare_slotting_plans', 'generate_roi_report'; map user intent to function calls; return structured results with LLM-generated natural language summaries.

Model entities (orders, pickers), resources (aisles, conveyors), processes (pick, travel, pack); implement two scenarios with different SKU-to-location mappings; run Monte Carlo simulations with stochastic demand; compare KPI distributions.

Package the solver in a Docker container for Lambda; request includes facility ID, time horizon, and constraint overrides; response includes ranked list of SKU-to-location reassignments with expected improvement metrics and confidence scores.

Use Git branching for experiments, store model hyperparameters and constraint configs in YAML, tag releases for production deployments, use GitHub Actions for CI (run backtest on historical data), and maintain an experiment tracking table.

Use document understanding models (LayoutLM, Donut) to extract location IDs, dimensions, and zone labels from floor plan PDFs; fine-tune on a small set of labeled warehouse documents; pipeline the extracted data into the optimization input format.

Set up SageMaker Pipelines with scheduled retraining triggers, automatic model evaluation against a holdout set, conditional deployment if metrics improve, model registry for versioning, and A/B testing with shadow mode before full cutover.

Ingest pick-event streams, WMS snapshots, IoT sensor data (aisle congestion), and order management data; build a unified star schema with fact tables (pick events) and dimension tables (SKU, location, time); create materialized views for velocity calculations and anomaly detection.

Great answers show empathy for domain expertise, describe a pilot-based persuasion strategy, quantify results, and credit the operations team's input in refining the solution.

Look for candidates who describe iterating with stakeholders to add real-world constraints, showed humility about model limitations, and ultimately produced a 'good enough' solution that was operationally feasible.

Strong answers mention specific conferences (INFORMS, MODELS), journals, GitHub repos, practitioner communities, hands-on experimentation, and a habit of reading papers and translating them into practical applications.

Look for diplomatic negotiation skills, data-driven mediation, creative compromise solutions (phased rollout, pilot programs), and clear communication of trade-offs to all parties.

Expect discussion of data cleaning pipelines, anomaly detection, sensitivity analysis, conservative assumptions where data was missing, and transparent communication of confidence levels to stakeholders.