Demand Forecasting for New Products With No Sales History
Introduction
Introducing a new SKU to the Malaysian retail market is a double-edged sword. Launch too cautiously and you face early stockouts, lost revenue and disappointed customers. Overestimate demand and you tie up working capital in slow-moving or dead stock.
For retail decision-makers, from CEOs to procurement managers and finance teams, traditional demand forecasting depends heavily on historical sales data. So how do you forecast demand for a product that has never sold before?
Handling these "cold-start" launches calls for a structured mix of data modelling, category knowledge and flexible inventory planning. This guide explains how Malaysian retailers can project demand for new SKUs with far less guesswork.
The Cold-Start Problem in Demand Forecasting
When a new product reaches store shelves or e-commerce channels, standard statistical methods such as moving averages or simple time-series analysis fall short, because there is no past performance to work from.
Without a structured approach, retail chains often rely on supplier promises or internal gut feeling. This leads to real operational problems:
Overstocking and trapped capital: Cash is locked in unproven inventory, raising holding costs and eroding gross margins.
The supply chain bullwhip effect: Sudden changes to replenishment trigger panic orders and costly last-minute freight.
Inaccurate stock valuation: Unsold or obsolete stock distorts inventory values on the balance sheet, which makes financial reporting and tax filings harder to get right.
4 Proven Demand Forecasting Strategies for New SKUs
To make up for the lack of sales history, demand planners use a combination of quantitative and qualitative proxy techniques.
1. Attribute-Based Modelling and Lookalike Analysis
Instead of treating a new SKU as a complete unknown, the forecasting engine breaks the product down into its core attributes, such as brand, price point, pack size, flavour, colour or function. It then matches these attributes to similar "lookalike" products in your existing range to build a baseline forecast.
For example, if you are launching a new 500ml organic oat milk, the system can analyse the sales velocity of similar plant-based milk SKUs in the same price band to estimate initial uptake.
2. Predecessor Mapping
If the new SKU is a reformulated version, a packaging update or a direct replacement for an older product, the predecessor's sales history becomes the starting point. Planners then adjust the forecast for expected cannibalisation and current market trends.
3. Category-Level Seasonality
Even without product-level history, category trends reveal how shoppers buy throughout the year. Layering in Malaysian retail peaks such as Hari Raya, Chinese New Year, Deepavali and year-end sales events ensures the launch plan accounts for seasonal spikes in demand. Broader retail trade figures from the Department of Statistics Malaysia (DOSM) can also help planners gauge overall market conditions.
4. AI-Driven Demand Sensing
Once the product goes live, the focus shifts from proxies to real-time demand sensing. By tracking early sales velocity, basket add rates and stock drawdown over the first 7 to 14 days, machine learning models continuously refine the longer-term forecast. This is where AI demand forecasting adds the most value, as it reacts to genuine customer behaviour far faster than manual reviews.
Comparing Demand Forecasting Methods for New Products
Forecasting Method | Best Used For | Main Data Inputs | Risk Level |
Lookalike mapping | Completely new brands or category expansions | Product attributes, category benchmarks, competitor pricing | Moderate |
Predecessor analysis | Product upgrades, rebranding or line extensions | Predecessor SKU sales history and substitution rates | Low |
Delphi / expert opinion | Niche, high-value or luxury items | Merchandiser surveys, supplier estimates, focus groups | High (subjective) |
AI demand sensing | Fast-moving consumer goods (FMCG) and fashion | Real-time POS data and early store sales velocity | Low once early sales data is available |
In practice, most retailers combine methods: a lookalike or predecessor forecast sets the launch quantity, and demand sensing takes over once real sales start flowing in.
Reducing Risk with Flexible Inventory Controls
Demand forecasting is only half the job; the other half is inventory control. Because early forecasts carry a wider margin of error, retail systems should apply protective rules during the launch period:
Targeted initial allocation: Send launch stock to high-performing flagship outlets first, rather than spreading it thinly across every branch.
Dynamic safety stock: Adjust reorder points (ROP) and economic order quantities (EOQ) regularly as sales data comes in, rather than relying on fixed settings.
Short replenishment cycles: Agree shorter replenishment windows with suppliers for the first 30 days, so you can respond quickly to a demand surge or slow sell-through.
Why Partner with QR Retail Automation (QRRA)?
QR Retail Automation (Asia) Sdn Bhd (QRRA) helps Malaysian retailers replace guesswork with data science. Through its AI demand forecast service and AgoraCloud Inventory Optimization as a Service (IOaaS), QRRA provides forecasting and inventory intelligence built for retail environments.
QRRA offers two ways to work, depending on your team's needs:
Managed Forecast-as-a-Service (FaaS): QRRA's specialists run the forecasting engine, handle data cleansing and model retraining, and review results before delivering demand plans to your team, so you do not need in-house data scientists.
Enterprise SaaS platform: A self-service platform for your own planners, with customisable models, "what-if" scenario planning, API integration with your data sources and collaborative forecasting.
System-ready outputs: Inventory recommendations such as safety stock, reorder points and order quantities are delivered as parameters ready for your existing ERP or merchandising system.
MSME tax incentive note: Micro, small and medium enterprises can claim a tax deduction of up to RM50,000 per year of assessment for consultancy fees incurred to implement e-invoicing, from year of assessment 2024 to 2027. If you are upgrading your retail systems at the same time, check with your tax adviser which incentives apply to your business.
Frequently Asked Questions
How do you forecast demand for a product with no sales history?
Retailers use proxies such as similar "lookalike" products, a predecessor SKU's sales and category seasonality, then refine the forecast once real sales begin.
What is lookalike demand forecasting?
It matches a new product's attributes, such as brand, price and pack size, to similar existing products and uses their sales to build a baseline forecast.
How soon can AI improve a new product forecast?
Early sales data from the first 7 to 14 days is often enough for demand sensing models to start refining the forecast.
How much stock should I send to stores at launch?
Start with targeted allocations to high-performing outlets and hold back stock centrally, so you can replenish where demand proves strongest.
Do I need in-house data scientists for demand forecasting?
Not necessarily. Managed forecasting services can run the models and deliver demand plans to your team.
Make Your Next Product Launch Smarter
Launching new SKUs does not have to be a gamble. By combining lookalike modelling, early demand sensing and automated inventory optimisation, your business can seize new opportunities while protecting working capital. Better demand forecasting at launch also means fewer markdowns and fewer emergency orders later in the product's life.
To see how this would work with your own product range, request a demo with QR Retail Automation and book a demand forecasting assessment with the team.
Comments