How to Use AI to Improve Forecast Accuracy for New and Slow-Turning Products

Q: Why is forecasting new or slow-moving products so difficult?

New and slow-turning products have little or no history, which makes predictions unreliable. AI helps overcome this by using peer comparisons, category signals, and adaptive models that learn quickly as data arrives.

Q: How does using an ensemble of models improve forecast accuracy?

AI runs multiple statistical and machine learning models simultaneously and selects the best-performing one for each SKU. This ensemble approach reduces forecast error and ensures predictions adapt to product behavior.

Q: What is analogous forecasting and how does AI make it more effective?

AI uses product attributes to find similar items and uses their sales patterns to seed forecasts for new SKUs. This creates a more accurate starting point and aligns forecasts with familiar business logic.

Q: How does transfer learning help when there’s no product history at all?

Transfer learning trains on broad category data—such as seasonality, pricing, and lifecycle trends—and applies that knowledge to new products, allowing accurate forecasting even with zero sales history.

Q: Why does modeling the product lifecycle improve forecast accuracy?

AI models forecast demand along the product’s lifecycle, including launch spikes, seasonal ramps, and short-term peaks, resulting in more realistic and timely sales expectations.

Q: What is true demand, and how does AI uncover it?

AI removes periods when products were unavailable or out of stock to reveal genuine customer demand. This prevents false zeroes from distorting the forecast and improves accuracy.

Q: How often should forecasts be updated for new products?

AI models retrain weekly or as new data appears, refining predictions continuously. This ensures early forecasts evolve quickly based on actual demand signals instead of static assumptions.

Q: What can planners do when there’s almost no data available?

Start conservatively with baseline sales, category-level curves, and controlled replenishment. Use small pilots or pre-orders to collect early signals and let AI refine the forecast once data is available.

Q: How does AI forecasting help planning teams overall?

AI streamlines forecasting by connecting product data, analog patterns, and historical performance into one system. It reduces manual work, increases confidence, and enables more strategic, data-driven decisions.

Forecasting new or slow-moving products is hard. You have little history, unpredictable demand, and high pressure to get the buy right. If you over-forecast, you eat markdowns. If you under-forecast, you lose sales and frustrate customers.

AI gives you a better way to handle this work. It helps you use the data you do have, even when it’s sparse. It finds patterns across products, categories, and past launches. And it learns fast as new data arrives.

Below is a practical, step-by-step guide on how to improve forecast accuracy for items with limited or inconsistent history.

Key AI-Enabled Strategies for Improved Forecast Accuracy

1. Use an Ensemble of Models

New products follow different patterns. A single model won’t fit everything.

AI fixes this by testing a set of models and choosing the best for each SKU.

You run statistical models, machine learning models, and hybrid methods at the same time. The system picks the strongest performer or blends a few of them. This avoids the “one model does it all” trap.

Research shows ensemble approaches reduce forecast error by 36% with sparse retail data. They track fashion launches, luxury items, and long-tail SKUs with more stability.

This also gives you confidence when speaking to finance or merchandising. You’re not relying on guesswork. You’re using a tested model selection process.

2. Borrow Signal from Similar Products (Analogous Forecasting)

When a SKU has little history, you look at its closest peers. AI takes this idea and scales it.

The system finds comparable products at a granular level. Style, silhouette, price point, material, category, or past trend behavior. It groups items with similar patterns and uses that history to seed the new item’s forecast.

You get a smarter starting point than a manual best guess.

You can also anchor the forecast to percentiles within the group. If an item is more of a niche piece, you set it at the lower bound. If it’s a hero style, you set it higher.

For planning leaders, this lets you defend decisions with data your merchants already understand: “This is based on the closest 30 products we sold last season.”

Tools like Toolio make this step faster because product attributes and performance are already structured in a clean hierarchy.

3. Leverage Machine Learning with Category Wide Data (Transfer Learning)

Sometimes the issue isn’t just limited history. It’s no history.

Transfer learning solves this by training models on broader category data and then applying that learning to the new item.

This teaches the AI about seasonality, holiday spikes, trend curves, and category demand drivers. The model brings this knowledge into the cold-start forecast, even when the new SKU has zero data.

This is especially helpful for luxury, furniture, or high-end beauty where sales are intermittent. The model uses patterns learned from hundreds or thousands of related products instead of guessing.

For planning leaders, this means new items start closer to the truth. And as data arrives, the model adjusts quickly.

4. Model the Life Cycle and Launch Curve

New items behave differently in their first weeks than they do later. Your forecasts should reflect that.

AI models can build curves for:

Launch-week spikes
Seasonal ramps
Promotional lift
Short life-cycle demand

If your team pushes a big launch, you adjust the curve to reflect the expected lift. If the product is seasonal, you shape the forecast to the selling window.

This reduces the back-and-forth between planning, merchandising, and marketing. You can agree on a realistic selling profile up front and let the model handle the fine-tuning.

5. Clean the Data to Reveal “True Demand”

Slow-moving items often show long stretches of zero sales. But zero sales don’t always mean zero demand. Sometimes the item was out of stock or not displayed.

AI systems correct for this by removing weeks when the item wasn’t available.

This prevents the forecast from being dragged down by false zeroes.

This matters for furniture, luxury accessories, or limited-distribution items where shoppers take longer to convert. Once the model sees only “good weeks,” it produces a more accurate base rate.

This also protects you in cross-functional conversations. You can clearly show the difference between “no stock” and “no demand.”

6. Update the Model as Data Arrives

Forecasting new and slow-moving products is not a static process. The model should update as soon as real data appears.

With AI, you don’t wait for mid-season hindsight.

You can retrain weekly.

You can switch models as the pattern becomes clearer.

You can track errors in real time and let the system adjust automatically.

By week six or eight, the forecast looks very different from week one because it’s based on real demand signals, not analogs alone.

This closes the gap between launch assumptions and actual performance faster than manual processes ever can.

In platforms like Toolio, this works cleanly because the system already monitors plan vs actuals and exceptions. You see changes as they happen, not after the fact.

7. Bootstrap Forecasts When There’s Almost No Data

Sometimes you start with nothing: a brand-new category, new brand, or new product line.

In these cases, start simple.

Set conservative baseline sales
Use category-level allocation curves
Anchor presentation minimums
Keep replenishment rules tight

Then let AI take over once a few weeks of sales appear.

Some teams test early demand with pilots, pre-orders, or limited drops. This is a simple way to gather signal fast. Once you have even a handful of data points, you feed them into the model and refine the forecast.

This helps reduce early risk while giving you a clean path to accuracy in the first few weeks.

Accurate AI Forecasts that Power Better Strategy

If you manage newness or slow-sellers, AI gives you a way to work faster with accuracy. You base decisions on peer patterns, category knowledge, clean demand signals, and weekly model updates.

Platforms like Toolio make this easier by storing your product data, analog groups, and performance trends in one connected system. You don’t spend hours pulling spreadsheets or aligning hierarchies. You spend your time shaping the forecast and guiding the business.

And that’s the real value. Better forecasts don’t just reduce error. They give you more control, more confidence, and more room for strategy.

If you need help building these forecasting strategies into your planning workflow, Speak to an Expert and see how this would work for your team.

FAQ: AI Forecasting for New and Slow-Moving Products

Why is forecasting new or slow-moving products so difficult?

New and slow-turning products lack historical sales data, making traditional models unreliable. Planners must often guess based on limited signals, which increases the risk of overbuying or underbuying. AI helps overcome this by recognizing patterns across similar products and adapting quickly as new data appears.

How does using an ensemble of models improve forecast accuracy?

No single forecasting method works for every product type. AI systems run multiple models—statistical, machine learning, and hybrid—simultaneously, then select or blend the most accurate for each SKU. This ensemble approach reduces forecast error, especially for items with limited or inconsistent history, and provides more defensible forecasts to share with finance or merchandising.

What is analogous forecasting and how does AI make it more effective?

Analogous forecasting uses similar products to predict demand for new SKUs. AI automates this process by identifying comparable products based on attributes like style, price, material, or trend behavior. It uses those analogs to seed forecasts and lets planners fine-tune results by adjusting percentile ranges. This creates a smart, data-backed starting point instead of a manual estimate.

How does transfer learning help when there’s no product history at all?

Transfer learning trains models on broader category data—seasonality, trends, and lifecycle behavior—then applies that knowledge to new products. This allows AI to generate forecasts for completely new items by leveraging insights from related categories. It’s especially useful in categories like luxury, furniture, or beauty where demand is irregular or early data is scarce.

Why does modeling the product lifecycle improve forecast accuracy?

Every new item follows a launch curve. AI can model launch spikes, seasonal ramps, and promotional lifts to predict sales more realistically. Planners can align forecasts with expected selling windows and lifecycle stages, reducing manual adjustments and improving coordination between planning, merchandising, and marketing teams.

What is “true demand,” and how does AI uncover it?

Slow-moving products often show false zero sales because they were out of stock or unavailable, not because demand was zero. AI corrects this by filtering out unavailable periods, revealing the true demand rate. This prevents models from being skewed by missing data and leads to more accurate forecasts for products with sporadic sales patterns.

How often should forecasts be updated for new products?

Forecasts for new and slow-moving products should update frequently—ideally weekly—as soon as real data appears. AI models automatically retrain and adjust based on performance, switching approaches as demand patterns become clearer. By week six or eight, forecasts are far more accurate because they’ve learned from actual selling behavior rather than assumptions.

What can planners do when there’s almost no data available?

When launching a new category or product line, start simple with conservative sales estimates and tight inventory rules. Use pilots, pre-orders, or limited releases to gather early data, then let AI refine the forecast as results come in. Even a few weeks of data can dramatically improve model accuracy once the system starts learning.

How does AI forecasting help planning teams overall?

AI brings structure, consistency, and speed to forecasting. It reduces manual work, improves confidence in predictions, and helps planners focus on strategy rather than data wrangling. Tools like Toolio integrate product data, analog relationships, and historical performance into one connected system, making it easier to refine forecasts and align teams around data-driven decisions.