Skip to main content
Book a demo
Horizon · Demand forecasting for retail & manufacturing

Demand forecasts your planners actually use.

For Heads of Planning, Supply Chain, and Operations at retailers, manufacturers, and FMCG brands. Horizon forecasts 2,000+ time series at 85-92% weekly accuracy, simulates promotions and shocks, and writes replenishment recommendations back into your ERP. Its agent layer executes the routine ones within approval rules your team sets, and escalates the rest. Median deployment: 12 weeks.

Book a demoSee capabilities
HORIZON · DECISIONS
Today's recommendation
Confidence 92% · Updated 4 min ago
Runner Edge · Multi
SKU 47-A · 8 colorways
Weekly demand · 12 weeksActualForecast
−6w−3wnow+3w+6w
Stock-on-hand · units · no reorder
3.2k2.4k1.7k9500
Reorder 1,240 units to cover the seasonal peak in 4 weeks.
2,000+
Time series forecasted simultaneously
85-92%
Weekly accuracy (1-MAPE, SKU × channel)
3
Core verticals: retail, CPG, manufacturing
12 wk
Median deployment to operational MVP
The difference

Why Horizon, not a planning suite or AutoML.

Buyers evaluating Horizon are usually deciding between a planning-suite RFP, an AutoML build, or holding out with spreadsheets. Here's the trade-off in one frame.

Planning suite / AutoML / BI dashboard
  • 12-month rollout, $1M+ license, multi-year vendor lock-in.
  • One hierarchy level. Bottom-up and top-down forecasts never reconcile.
  • Generic AutoML: accuracy without context, no promotion lift modeling.
  • Locked into the vendor's data model and reporting cadence.
  • BI dashboards report what already happened.
  • Implementation owned by an SI partner you'll never see again.
Horizon
  • 12-week median deployment, phased contract with kill clause.
  • Hierarchical reconciliation across SKU × customer × region.
  • Promotion lift modelled separately from organic demand. Every forecast cites its signals.
  • Plugs into your ERP and writes recommendations back into your planners' workflow.
  • Recommends the reorder, the transfer, and the production schedule, not just the chart. Executes routine reorders and transfers itself, inside your approval rules.
  • Built and operated by ZAAI. Same team after go-live as before.
Capabilities

Five capabilities. One engine.

Connect your data once. Add capabilities as your team graduates from forecasting to simulation to automated decisions. Same data layer, same observability, same integration surface across all five.

01
01 · DEMAND INTELLIGENCE

Multi-level demand forecasting

For Heads of Buying & Planning

Forecasts at any hierarchy level: SKU × customer × week, product × store × day, or aggregate. Integrates sales history, promotions, seasonality, weather, and market signals. Hierarchical reconciliation with weighted least-squares. Deployed for 2,000+ time series simultaneously with 16-week horizons.

  • SKU × customer × week granularity (or any other)
  • Hierarchical reconciliation across product/region/customer levels
  • Promotion lift modeling separated from organic demand
  • Drift monitoring + automated retraining
  • MAPE, RMSE, weighted accuracy reported per dimension
  • Explainable forecasts. Every prediction cites its signals
02
02 · SCENARIO ENGINE

What-if simulation

For Planning & Finance

Test promotion impact, pricing changes, supplier delays, and demand shocks before committing. Compare outcomes across revenue, margins, inventory, and cash flow. Used by banking clients for dose-response curve estimation and by FMCG for promotion planning.

  • Promotion impact on inventory and margin
  • Pricing elasticity / dose-response per segment
  • Supplier delay and demand-shock scenarios
  • Multi-week horizon with confidence bands
  • Side-by-side scenario comparison
  • Decision memo export
03
03 · RESOURCE OPTIMIZATION

Replenishment & production planning

For Operations & Supply Chain

Generates replenishment recommendations, inter-store transfers, and production schedules. Deployed for retailers with 42 stores and manufacturers aligning factory capacity with demand forecasts. Optimizes for service levels, holding costs, and lead times.

  • Per-location reorder recommendations with confidence
  • Lead-time aware safety stock calculations
  • Inter-store and inter-warehouse transfer suggestions
  • Open-to-buy budget tracking by category
  • Production scheduling aligned to factory capacity
  • Approval workflow for high-value or high-risk orders
04
04 · PROACTIVE ALERTS

Root cause analysis & alerts

For Operations & Field Teams

Classifies stockout root causes (delayed restocking, incorrect stock levels, lack of rotation) and alerts teams before problems materialize. Deployed for FMCG distributors to flag stores at high stockout probability before it happens in-store.

  • Stockout root-cause classification
  • Probability-of-stockout alerting per SKU × store
  • Anomaly detection across operational signals
  • Configurable thresholds and notification channels
  • Field-team-ready alerts with recommended action
05
05 · AGENT LAYER

Autonomous decision execution

For COOs & Operations Leadership

Turns recommendations into executed decisions. The agent layer coordinates the reorder, the transfer, and the schedule change across your ERP and the systems around it, executing within rules your team sets. Routine reorders and transfers run overnight; everything else waits for sign-off. In manufacturing, the same layer keeps production schedules and material flows aligned as forecasts move.

  • Executes routine reorders, transfers, and schedule updates automatically
  • Budget thresholds, quantity limits, and approval flows set by your team
  • Escalation to a person for anything above threshold
  • Writes to your ERP; reaches WMS and MES over the same REST API
  • Full audit trail on every executed action
  • Override any decision at any time
Deployment evidence

Where Horizon runs in production today.

Anonymised production deployments. Each shipped end-to-end with monitoring, retraining, and knowledge transfer to the customer team.

Industry
Scale
Timeline
Capability
FMCG · Beverage Distributor
SKU × store level, 14-day horizon
23 weeks
Stockout prediction + root cause analysis on GCP
Industrial Manufacturer
2,000+ time series, 16-week horizon
12 weeks
Hierarchical multi-level forecasting + production planning
Fashion Retailer
1,700 SKUs, 42 stores
62 weeks (4 phases)
Financial + sales + inventory + production planning, end-to-end
Retail Distributor
Multi-channel, B2B + B2C
40 weeks (3 phases)
Conversational analytics + demand forecasting
Methodology

How a Horizon engagement runs.

Five phases. Phased contracts so you evaluate before committing further. 100% production deployment rate across engagements.

  1. Weeks 1–2

    Requirements & architecture

    Requirements and Architecture Document. Data source mapping across ERP, banking, e-commerce, and production systems.

  2. Weeks 3–6

    Data analysis & feature engineering

    Data Readiness Report. Quality assessment, feature pipelines, baseline benchmarking.

  3. Weeks 7–12

    Model development & benchmarking

    Operational MVP with trained models, data pipelines, and back-testing results.

  4. Weeks 13–16

    Testing, validation, monitoring

    Backtesting evidence, monitoring dashboards, alert configuration, drift detection.

  5. Weeks 17+

    Knowledge transfer & stabilization

    Training sessions, documentation, hypercare period. Your team operates the system independently.

How we measure success

  • MAPE·Mean Absolute Percentage Error per SKU and per customer
  • RMSE·Root Mean Squared Error across forecast horizons
  • Weighted accuracy·Business-value-weighted accuracy by volume
  • Coverage·Stability of predictions across time horizons
  • Bias analysis·Systematic over/under-forecasting detection
Industries

Where Horizon is in production today.

Three primary verticals with the deepest deployment evidence. Horizon's data layer and modeling stack transfer cleanly across adjacent industries. See the full list under Solutions.

FMCG & Beverages

On-shelf availability prediction across hypermarkets. SKU × store demand prediction with root-cause classification for stockouts. Deployed on GCP with BigQuery and Vertex AI.

  • Stockout prediction at SKU × store
  • Root cause classification (3+ categories)
  • Proactive field-team alerts
  • BigQuery + Vertex AI pipeline
See how it works

Manufacturing

Weekly demand forecasts for 40+ end products across 50+ customers. Hierarchical forecasting at SKU-customer, SKU, and aggregate levels. Integrated with ERP and production planning.

  • Hierarchical reconciliation
  • Production capacity alignment
  • ERP integration (SAP, Primavera)
  • Multi-customer aggregation
See how it works

Fashion & Retail

Sales and treasury forecasting, inventory optimization across 42 stores, and production scheduling for 1,700+ active SKUs with 6-month lifecycles.

  • Pre-season buy planning
  • Mid-season replenishment
  • Treasury reconciliation (135K+ tx/year)
  • Production scheduling integration
See how it works
Integrations

Plugs into the systems your team already uses.

Horizon ingests signals at the inner ring, pulls from your ERPs in the middle, and reaches the commerce, cloud, and banking systems your buyers and planners already run on the outer ring. ERP write-back runs in production today; when the action lives in a WMS or MES instead, Horizon gets there over the same REST API.

Connected SystemsERPsSignals
HorizonAgent
Sales
Stock
Promotions
Pricing
SAP
Sage
PHC
REST API
Shopify
Banking
Horizon FAQ

Common questions about Horizon.

See Horizon on your data.

Book a 30-minute demo. We'll walk through your data shape and show you what Horizon would produce on it.

Book a demoEmail the team