# Skulicity Cuts Forecast Runtime By Over 50% With TimeGPT

> How Skulicity transformed a compute-heavy forecasting pipeline into a scalable, production-ready system with TimeGPT, enabling faster replenishment decisions across dynamic retail supply chains

Canonical HTML: https://www.nixtla.io/success-stories/skulicity

## Customer

- Name: Skulicity
- Industry: Retail Analytics & Supply Chain

With Nixtla's TimeGPT, Skulicity transformed a compute-heavy forecasting pipeline into a scalable, production-ready system, enabling faster replenishment decisions across dynamic retail supply chains without replatforming its Microsoft Fabric environment

## Key results

- **50% Reduction in Forecast Runtime:** Achieved a 50% reduction in forecast runtime on selected workloads, completing forecasts in under an hour
- **500K Unique IDs per Pipeline Run:** Processing 500,000 unique IDs per pipeline run with 3× weekly forecasting cadence at production scale
- **50% Reduction in Compute Costs:** Lowered cloud costs by 50% in a time-based billing environment while expanding forecasting capabilities

## About


Skulicity is a predictive analytics and forecasting platform built for demanding retail supply chains. They help businesses selling to major big-box retailers. Skulicity supports complex planning and replenishment needs across horticulture, perishable grocery, consumer packaged goods, and retail categories where demand volatility, short shelf life, and timing precision are critical

## The Challenge

For Skulicity, forecasting is a core operational function rather than a back-office exercise. The company generates highly granular forecasts, by store, by SKU, by day, that directly inform replenishment decisions for customers operating in fast-moving retail categories. In environments such as horticulture and perishables, where products have short shelf lives and demand is sensitive to external factors, forecast timing is just as important as accuracy
- Microsoft Fabric's 300 MB limit on model artifacts made deploying modern deep learning models difficult
- Growing runtimes of 2-4 hours per client slowed replenishment decisions in time-sensitive retail categories
- Bespoke code maintenance increased operational risk for a lean team managing complex forecasting pipelines

## The Solution

The team evaluated TimeGPT against its existing pipeline using cross-validation across three criteria that mattered most in production: forecast accuracy, compute time, and overall solution complexity. TimeGPT outperformed the internal approach on all three dimensions. It delivered stronger accuracy, ran significantly faster, and required far less bespoke code to operate at scale. Based on these results, Skulicity replaced its internal forecasting pipeline with TimeGPT
- Deployed TimeGPT with custom compressed model artifacts and lazy execution strategy to work within Microsoft Fabric constraints
- Expanded from ~10 to nearly 40 input features without requiring additional infrastructure or longer runtimes
- Standardized forecasting workflows using Nixtla's APIs, reducing technical debt and making pipelines easier to maintain

## Business Outcomes

Transforming forecasting performance and accelerating time to market
- **Faster Forecasting Cycles**
  - Training and inference reduced from 2-4 hours to under 1 hour for select workloads
  - 3× weekly forecasting cadence at production scale
- **Reduced Infrastructure Costs**
  - 50% reduction in compute costs through optimized runtime
  - Better resource utilization with expanded feature sets
- **Accelerated Time to Market**
  - Reduced technical debt with standardized workflows
  - Easier handoffs between team members without specialized time-series expertise

## Testimonial

> "The biggest breakthrough for us was overcoming our deployment constraints without changing our environment. That flexibility made it possible to scale forecasting without rebuilding everything from the ground up"

Vinit Agrharkar, Data Engineer Intern