Stock IQ, from Iqlytics Solutions Pvt Ltd, is a distributor auto replenishment system for FMCG general trade. It replaces the salesman-and-norms way of placing primary orders with a daily, explainable, constraint-aware suggested order that planners approve by exception.

What problem it solves

In general trade, most primary orders are still built from fixed norms like "keep 15 days of stock", adjusted by the sales team at month end. That produces the familiar pattern: stockouts on fast movers during festive build-up, piles of slow movers and near-expiry stock, half-empty trucks, and month-end loading that distorts the demand signal. ARS turns replenishment into a pull system driven by what distributors actually sell to retailers.

How a daily run works

1. Ingest. Pull closing stock and secondary sales from the distributor DMS, in-transit and credit from ERP, retailer orders from the sales app, and the promo and event calendar.
2. Cleanse. Cap spikes (scheme loading, bulk institutional orders) at median + 4 MAD so they don't inflate the baseline. In production, also correct for days the SKU was out of stock, when recorded sales understate demand.
3. Classify. Segment each SKU at each distributor by value (ABC) and by demand pattern using ADI and CV² (smooth, erratic, intermittent, lumpy).
4. Forecast. Smooth and erratic items use damped Holt smoothing with weekday indices. Intermittent and lumpy items use Croston with the SBA correction. Planned events multiply on top. Accuracy is measured in weekly buckets on a 2-week holdout against a flat 28-day average so the team can see forecast value added.
5. Set policy. Periodic review with order-up-to level, protecting demand over review period plus lead time, including lead-time variability.
6. Apply constraints. Round to case multiples, cap cover by shelf life, trim to the distributor's credit headroom by priority, and optionally fill or trim the last truck.
7. Review and release. Planners work the exception list, override with a reason code, approve, and the plan is posted to ERP as a sales order.

The core maths

For each SKU at each distributor, with review period R, mean lead time L, lead-time standard deviation σL, forecast error σd and service-level factor z:

Protection period P = R + L Safety stock SS = z × √( P·σd² + D̄²·σL² ) Reorder point ROP = forecast over L + SS Order-up-to S = min( forecast over P + SS , D̄ × max cover days ) Net position NP = on hand + in transit Suggested order Q = round_to_case_multiple( S − NP )

Max cover days is the smaller of the class limit and the shelf-life share, so a 120-day-life snack can never be ordered beyond about 30 days of cover at the default settings.

Architecture

flowchart LR
  DMS["Distributor DMS
stock, secondary sales"] --> ING
  ERP["ERP
in-transit, credit, prices"] --> ING
  APP["Retailer app and SFA
orders, outlet coverage"] --> ING
  CAL["Promo and event calendar"] --> ING
  ING["Ingestion and data quality"] --> CLN["Cleansing and classification"]
  CLN --> FC["Forecast engine"]
  FC --> POL["Policy engine
safety stock, order-up-to"]
  POL --> CON["Constraint solver
case, shelf life, credit, truck"]
  CON --> WB["Planner workbench
exceptions, overrides, approval"]
  WB --> OUT["Sales order to ERP
and distributor notification"]
  OUT -.-> ING

What a production build adds

The prototype runs in the browser on synthetic data. A production system would run the engine as a scheduled service (Python with statsmodels or a gradient-boosted model for high-volume SKUs, or a planning platform such as SAP IBP or o9 if the company already licenses one), store history in a warehouse, integrate with the DMS and ERP through APIs, and push approved orders back as ERP sales orders. Distributors would see and confirm their suggested order in the distributor app. Role-based access separates planners, area sales managers who approve, and distributors who can request changes.

Rollout

1. Pilot in suggest-only mode, 8 to 10 weeks. 10 to 15 distributors of mixed size and region. Orders still placed manually, but compared line by line with the ARS suggestion.
2. Approve-and-release. ARS orders become the default, planners approve by exception, overrides require a reason code. Tune service levels and cover caps from pilot results.
3. Touchless for clean lines. Plans with no critical exceptions release automatically. Scale region by region.
4. Extend downstream. Use the same engine for retailer-level suggested orders in the sales app and for depot-to-distributor stock allocation during supply shortages.

When the distributor can't pay for the full order

The system first calculates what each SKU needs, then checks it against the smaller of the distributor's credit headroom and the amount they can pay now. If the order doesn't fit, it is redistributed using the distributor's own SKU priority list (tier and rank) and minimum order quantities, so a line is either skipped or ordered at least at its MOQ. Four methods are offered side by side with their effect on stockout risk and cover, and the planner picks one. Held-back quantity isn't lost: it stays in the need and comes back automatically in the next run.

What to measure

Distributor fill rate and out-of-stock SKU-days on must-stock lines, inventory days at distributors, forecast WMAPE and bias, share of order lines released without edits, override rate and the accuracy of overrides versus the system, truck utilisation, and near-expiry and damage write-offs. Measure against the pilot's control distributors, not against last year, so seasonality doesn't flatter the result.