How to Forecast Demand for New Products Without Sales History

The cold-start problem

Every formula in this series so far has one thing in common: it needs historical sales data.

Safety stock needs the standard deviation of daily demand. Reorder points need average daily demand and average lead time. EOQ needs annual demand. All of them assume you have at least 60 to 90 days of clean, stable sales history for a SKU.

For your existing catalog, that is fine. The data is sitting in your Shopify order exports.

But what about the product you are launching next month? Or the new colorway of your bestseller? Or the seasonal item you are bringing in for the first time? You have zero history. The formulas produce either nonsense or nothing at all.

This is the cold-start problem, and it is the single biggest gap in most inventory planning tools. They handle mature SKUs well and new launches poorly, which is exactly backwards in terms of risk. Your mature SKUs have predictable demand. Your new launches are where you are most likely to over-order or under-order by a wide margin.

Why this matters more than you think

Consider the stakes:

• Over-order a new product and you tie up cash in inventory that might not sell. If the product underperforms, you are stuck with dead stock and the holding costs that come with it.
• Under-order a new product and you stock out during the launch window, which is often when demand is highest and marketing spend is active. You lose sales, waste ad spend, and miss the momentum window.

For mature SKUs, a 20% forecasting error means you carry a bit more or less buffer than optimal. For a new launch, a 20% error might mean the difference between a successful product introduction and a write-off.

Most inventory apps handle this by simply not addressing it. They wait for data to accumulate and leave you on your own for the first 60 to 90 days. Some apps show a warning like "insufficient data" and offer no alternative. That is not a solution. That is an admission that the tool does not handle the most important planning decision you make.

Three approaches that actually work

There are three practical methods for forecasting demand on a new product. Each has different data requirements and different levels of confidence. You should choose based on what information you actually have.

Method 1: Manual forecast

This is the simplest approach. You set the expected daily demand for the new SKU by hand, based on your own judgment.

Where does that number come from?

• Pre-launch signals. Waitlist signups, email interest, social media engagement on product teasers. If 200 people signed up for a launch notification and your typical conversion rate is 5%, you might expect 10 orders in the first week.
• Category benchmarks. If you sell candles and your average candle SKU sells 3 units per day after the first month, that is a reasonable starting point for a new candle.
• Supplier or industry data. Some suppliers share sell-through data from other retailers carrying the same or similar products.
• Marketing plan. If you are running paid ads at launch, your expected ROAS and average order value give you a rough unit forecast.

The manual forecast is explicitly a guess, but it is a structured guess. You are stating an assumption, planning around it, and then updating it as real data comes in.

When to use it: Truly new products with no comparable SKU in your catalog. First entry into a new category. Products where your judgment or market research is the best available signal.

How to set it up: In your planning spreadsheet or tool, set the Forecast Mode for that SKU to "Manual" and enter your expected average daily demand. Use that number in place of the calculated average when computing safety stock, reorder point, and EOQ. Revisit it every week as real sales data accumulates.

Method 2: Proxy SKU

This is the most powerful method for stores that already have an established catalog. Instead of guessing from scratch, you borrow the demand pattern from an existing similar SKU and optionally scale it.

Examples:

• New color of an existing product. You sell a black tee that moves 8 units/day. You are launching a navy version. Use the black tee as the proxy, perhaps scaled to 60% if you expect the new color to sell slightly less.
• Size extension. You sell a 12oz candle at 5 units/day. You are launching a 6oz version at a lower price point. Use the 12oz as the proxy, scaled to 80% based on your expectation that the lower price drives slightly higher volume.
• Seasonal variant. You sold a pumpkin spice version last fall at 6 units/day during peak. You are launching a gingerbread version this year. Use last year's pumpkin spice as the proxy.

The proxy method works because demand patterns within a product family tend to be correlated. A customer who buys candles from you is likely to consider your new candle. The demand variability, the day-of-week patterns, and the lead time are all similar.

When to use it: New variants, colorways, sizes, or flavors of existing products. Products in the same category as something you already sell. Seasonal items where you have data from a comparable prior season.

How to set it up: Set the Forecast Mode for the new SKU to "Proxy." Select the existing SKU whose demand pattern you want to borrow. Optionally set a scale factor (e.g., 0.6 means you expect 60% of the proxy's demand). The planning model then uses the proxy's average daily demand (scaled) and the proxy's demand variability to calculate safety stock, reorder point, and EOQ for the new SKU.

This is significantly better than a manual guess because you are borrowing not just the average, but the variability. If your proxy SKU has erratic demand (high standard deviation), your new SKU's safety stock will be appropriately higher. If the proxy has stable demand, the buffer will be leaner.

Method 3: Pre-order or soft launch data

If you can generate even a small amount of real demand data before committing to a full inventory purchase, you dramatically reduce your forecasting risk.

• Pre-orders. Open the product for purchase before you have inventory. You collect real orders, fulfill them when stock arrives, and use the pre-order velocity as your demand signal.
• Small test batch. Order a small quantity (even below your ideal EOQ) to test demand. Accept the higher per-unit cost as the price of information. If the product sells through quickly, you have a demand signal. If it sits, you avoided a large commitment.
• Landing page test. Create the product page, drive traffic to it, and measure add-to-cart rate even if you do not fulfill orders yet. This gives you a conversion signal without inventory risk.

When to use it: When the cost of being wrong is high (expensive products, large MOQs, long lead times). When you have time before the launch window to collect data.

How to set it up: Run the pre-order or test for at least 7 to 14 days. Calculate the daily demand rate from that data. Switch the SKU's Forecast Mode from Manual to Auto once you have enough data points for your tool's lookback window. In a custom spreadsheet, you might be comfortable switching after 30 days. In SkuClerk, the default Sales History Window is 90 days, meaning Auto mode uses the last 90 days of orders to calculate demand. You can change this in the Settings tab, but the setting is global: it applies to every SKU in your catalog. Shortening it to 30 days to accommodate a new launch would also shorten the lookback for your mature SKUs, which is usually not what you want. The better approach is to keep the 90-day window and leave new SKUs on Manual or Proxy until they have 90 days of history.

The transition plan: Manual to Proxy to Auto

The three methods are not mutually exclusive. They are stages in a lifecycle:

1. Before launch (no data): Use Manual or Proxy forecast mode. Set your best estimate of daily demand.
2. First 30 to 60 days (thin data): Keep Manual or Proxy, but compare your forecast to actual sales weekly. Adjust the manual number or scale factor if reality diverges from your assumption.
3. After 90 days (sufficient data): Switch to Auto. The formulas now have enough history to calculate average demand and standard deviation reliably. Your safety stock, reorder point, and EOQ are now data-driven.

Why 90 days? Most inventory tools, including SkuClerk, default to a 90-day lookback window for calculating demand averages and variability. That window needs to be full of real sales data before Auto mode produces reliable numbers. Switching earlier is possible if your tool lets you shorten the window, but be aware that a shorter window also affects your mature SKUs. Unless you have a tool that lets you set the lookback per SKU, keep new launches on Manual or Proxy until the 90 days are up.

The key discipline is the weekly check-in. Every Monday, look at your new launches and ask: is the actual demand tracking above, below, or in line with my forecast? If it is off by more than 30%, adjust the manual number or scale factor immediately. Do not wait 90 days to discover you were wrong.

What most tools get wrong

Most inventory planning tools treat the cold-start problem as an edge case. It is not. For a growing Shopify store, new product launches are a regular event. If you launch 2 to 4 new products per month, you always have SKUs in the cold-start phase.

Here is what the common approaches miss:

"Insufficient data" warnings with no alternative. Telling you the formula cannot run is not helpful. You still need to place an order. The tool should offer a way to input a manual forecast or borrow from a proxy.

Flat "days of supply" targets. Some tools let you set a target like "keep 30 days of supply." But 30 days of supply for a new product is a guess built on a guess. If your demand estimate is 50% too high, you are carrying 30 days of inflated demand.

Ignoring variability for new SKUs. Even if a tool lets you input a manual demand number, it often does not account for the higher uncertainty inherent in a new product. A new launch should have a higher safety stock multiplier than a mature SKU with stable demand, because your confidence in the forecast is lower.

The proxy method solves this elegantly. By borrowing the demand variability from a similar existing SKU, you get a realistic safety stock buffer without needing 90 days of history for the new product.

A worked example

You are launching a new scented candle (Eucalyptus) alongside your existing bestseller (Lavender).

Lavender candle (existing, 6 months of data):

• Average daily demand: 5 units/day
• Standard deviation of demand: 2.1 units/day
• Average lead time: 21 days
• Safety stock (95% service level): 22 units
• Reorder point: 127 units

Eucalyptus candle (new, zero data):

• Proxy: Lavender candle
• Scale factor: 0.7 (you expect it to sell at 70% of Lavender's rate initially)
• Derived average daily demand: 5 x 0.7 = 3.5 units/day
• Derived standard deviation: 2.1 x 0.7 = 1.47 units/day
• Same lead time: 21 days
• Derived safety stock: 16 units
• Derived reorder point: 90 units

You now have actionable planning numbers for the Eucalyptus candle on day one. No waiting. No guessing at safety stock. No "insufficient data" error.

After 90 days, the Eucalyptus candle has a full window of real sales data. You switch it to Auto mode. The actual average daily demand turns out to be 4.2 units/day (higher than your 3.5 estimate). The formulas recalculate automatically, and your reorder point adjusts upward.

Because you had a reasonable proxy-based plan from day one, you did not stock out during the launch window. And because you checked weekly and saw demand trending above your proxy estimate, you placed a slightly larger second order at week three to bridge the gap.

The spreadsheet setup

If you are building this yourself, you need three things per SKU:

1. A Forecast Mode column. Values: Auto, Manual, or Proxy.
2. A Manual Demand Override column. Only used when Forecast Mode = Manual. You enter your expected average daily demand.
3. A Proxy SKU column and Scale Factor column. Only used when Forecast Mode = Proxy. You enter the SKU ID of the product to borrow from, and a decimal scale factor.

Your calculation columns then use conditional logic:

• If Forecast Mode = Auto: use the calculated average daily demand from order history.
• If Forecast Mode = Manual: use the Manual Demand Override value.
• If Forecast Mode = Proxy: look up the proxy SKU's average daily demand and standard deviation, multiply by the scale factor.

All downstream formulas (safety stock, reorder point, EOQ) work the same regardless of forecast mode. They just receive their demand inputs from different sources depending on the mode.

The bottom line

New product launches are not an edge case. They are a regular, high-stakes planning decision. Any inventory planning system that cannot handle them is incomplete.

The three methods, Manual, Proxy, and pre-order data, give you a structured way to plan inventory for products with no history. The proxy method in particular is underused and powerful: it lets you leverage your existing catalog's demand patterns to make informed decisions about new additions.

The goal is not to predict the future perfectly. It is to have a defensible starting plan that you refine weekly as real data comes in. That is the difference between a launch that runs out of stock in week two and one that has inventory waiting when the orders arrive.