Skip to main content

Bilingual glossary

Canonical glossary of AInventory terms. It is the single source of definition: other pages link here instead of redefining. Technical labels are kept in their original language. Terms from general knowledge not specific to the platform are marked (general concept).

A​

Accuracy​

Measure of how closely the forecast matches actual demand. In AInventory expressed as 1 - MAPE or the equivalent metric in the KPIs view. See also MAPE.

ARIMA​

Autoregressive Integrated Moving Average. A statistical time-series model that captures trend and autocorrelation in demand history. One of the models available in AInventory's statistical cascade.

B​

Baseline (AI)​

Forecast generated automatically by AInventory's AI engine at the start of each cycle. It is the first phase of the process and is not editable by users; it serves as the reference point for computing Forecast Value Added (FVA).

BIAS​

Systematic deviation of the forecast from actual demand: a positive BIAS indicates persistent overestimation; a negative BIAS indicates underestimation. See /docs/referencia/metricas/bias for the full formula.

bootstrap​

(general concept) Statistical resampling technique: multiple samples are drawn with replacement from historical data to estimate probability distributions without assuming a functional form. Used in AInventory's demand simulation when history is short.

C​

CNNQR​

Convolutional Neural Network — Quantile Regression. A deep-learning forecast model that estimates the full distribution of future demand. Available in the first level of the statistical cascade.

Consensus​

Final phase of the planning cycle in which the agreed-upon forecast is consolidated across all areas. In the default configuration it is the phase marked Is final = true. Its value is used as the official input for purchasing and inventory plans.

D​

DeepAR​

Deep-learning forecast model based on recurrent neural networks (RNN) for probabilistic time-series forecasting. Learns global patterns from multiple SKUs simultaneously. Available in the first level of the statistical cascade.

E​

ECM / MSE​

Error Cuadrático Medio / Mean Squared Error. Average of squared forecast errors. Disproportionately penalizes large errors. See /docs/referencia/metricas/mse for the formula.

ETS​

Error, Trend, Seasonality. Exponential smoothing model that captures trend and multiple seasonality using exponentially decreasing weights on historical data. Available in the second level of the statistical cascade.

F​

factor k​

Multiplier that determines how many standard deviations of demand are added to the mean demand during the lead time to calculate safety stock. A higher factor k implies a higher service level and more safety inventory.

fill rate​

Proportion of demand fulfilled without incurring a stockout. Service-level metric. A fill rate of 95 % means 95 % of demanded units were delivered in the period without a stockout.

Forecast Value Added (FVA)​

Metric that measures the value each human intervention adds to the forecast relative to the Baseline (AI). A positive FVA means the adjustment improved accuracy; a negative FVA means it degraded it. Calculated only on closed versions.

forecast​

Quantitative prediction of the future demand of a SKU for one or more periods. In AInventory the forecast is generated by the statistical cascade (Baseline AI) and then adjusted by users in the cycle phases.

H​

holding​

Cost of keeping units in inventory for one period. Includes cost of tied-up capital, physical storage, and obsolescence risk. In AInventory it is a per-SKU input parameter. See Inventory parameters.

L​

lead time​

Replenishment time: number of periods between placing a purchase order and receiving the goods at the warehouse. Key parameter for computing the ROP and safety stock.

M​

MAE​

Mean Absolute Error. Average of the absolute values of forecast errors. Easy to interpret in the same units as demand. See /docs/referencia/metricas/mae.

MAPE​

Mean Absolute Percentage Error. Average of absolute errors expressed as a percentage of actual demand. Allows precision comparison across SKUs of different scales. See /docs/referencia/metricas/mape.

MASE​

Mean Absolute Scaled Error. Scales the MAE by dividing it by the error of a naive reference model (e.g., seasonal naive). A MASE < 1 indicates the model outperforms the naive. See /docs/referencia/metricas/mase.

Monte Carlo​

(general concept) Simulation method that generates thousands of future demand scenarios through random sampling from probability distributions. Used in AInventory to estimate the cost distribution of inventory and select the optimal policy.

MOQ​

Minimum Order Quantity. Purchasing restriction imposed by the supplier. TODO: confirm whether MOQ is an input parameter in AInventory's inventory module.

N​

Net Flow (FN)​

Inventory position indicator: FN = on-hand inventory + in-transit inventory. Central variable of the buffer status light: compared with SS, ROP, and target stock to determine each SKU's state.

newsvendor​

(general concept) Classic single-period inventory optimization model that balances the cost of excess (holding) against the cost of shortage. In AInventory it forms the third level of the statistical cascade and underpins the safety stock calculation.

NPTS​

Non-Parametric Time Series. A forecasting approach that does not assume a parametric demand distribution; it uses the empirical history directly. Available in AInventory's statistical cascade.

P​

phase​

One stage of the forecast approval workflow within a monthly cycle. Each phase has a code, name, edit window, and assigned users. See Phases and versions.

predict-then-optimize​

(general concept) Two-stage decision paradigm: first a forecast is generated (predict), then that forecast is used as input to an optimization model (optimize) that determines the inventory policy. The central approach of AInventory.

Prophet​

Time-series model developed by Meta that handles trend, multiple seasonality, and calendar effects. Available in the second level of the statistical cascade.

R​

Reorder Point (ROP)​

Inventory position level at which a new purchase order should be placed in order to receive replenishment before stock runs out. Calculated as mean demand during lead time + safety stock.

RMSE​

Root Mean Square Error. Square root of the MSE. Expresses error in the same units as demand and penalizes large errors. See /docs/referencia/metricas/rmse.

Rule 50/50​

Fallback heuristic in the statistical cascade: when no statistical model produces a reliable prediction, the forecast is built as the 50 % weighted average of the historical mean and the most recent demand.

S​

safety stock​

Additional inventory kept as a buffer to absorb demand or lead-time variability. Calculated as a function of factor k, the standard deviation of demand, and the lead time.

shortage​

Economic penalty per unit that cannot be delivered on time due to lack of inventory. In AInventory it is a per-SKU input parameter reflecting the cost of a stockout.

single source of truth​

(general concept) Data architecture principle whereby a given piece of data has a single authoritative system of record. In AInventory, the glossary is the single source of truth for term definitions; Supabase is the single source of truth for process data.

SKU​

Stock Keeping Unit. The minimum unit of inventory tracking: uniquely identifies a product in a warehouse context. In AInventory, all parameters, forecasts, and KPIs are computed at the SKU level.

SofIA​

AI assistant integrated into AInventory. Detects the active module and offers contextual quick actions (analysis, interpretation, recommendations). Responds via streaming with Markdown formatting. See SofIA quick actions.

stage-gate​

(general concept) Process management model in which each stage must be completed and approved ("gate") before moving to the next. AInventory's phase workflow (Baseline → Sales → Marketing → Consensus) implements a stage-gate process for forecast approval.

statistical cascade (four levels)​

Pipeline of models that AInventory executes to generate the Baseline (AI). The four levels are: (1) global deep-learning models (DeepAR, CNNQR), (2) per-series classical models (ARIMA, ETS, Prophet), (3) newsvendor model with bootstrap, and (4) Rule 50/50 as fallback. The system selects the optimal level per SKU.

stockout​

Situation in which demand exceeds available inventory, resulting in lost sales or delivery delays. In the buffer status light, corresponds to the "Out of stock" and "Depleted" states.

buffer status light​

System of five states (Out of stock, Depleted, Reorder, Optimal, Excess) that classifies the inventory position of each SKU by comparing the Net Flow with SS, ROP, and target stock. See Buffer states.

T​

target stock​

Maximum desirable inventory level for a SKU. In the buffer status light, a Net Flow exceeding the target stock indicates excess inventory (state "Excess").

tenant​

Company or business unit that operates in isolation within the AInventory platform. Each client lives in an isolated data space, with its own phases, versions, SKUs, and parameters. One tenant's data is not visible to others.

V​

version (Open/Closed)​

A time snapshot of the planning process that freezes the forecast state. An Open version is in active editing; a Closed version is immutable and is the only state on which KPIs and FVA are calculated. See Phases and versions.

W​

WAPE​

Weighted Absolute Percentage Error. Variant of MAPE that weights errors by demand volume, reducing the impact of low-volume SKUs. See /docs/referencia/metricas/wape.

wQL​

Weighted Quantile Loss. Metric for evaluating probabilistic forecasts: measures how well calibrated quantile estimates are (e.g., P10, P50, P90). Relevant for the deep-learning models in the cascade. See /docs/referencia/metricas/wql.