How to Do eCommerce Forecasting the Right Way

You place an inventory order every 4 to 6 weeks. The process usually starts with last month’s sales, a seasonality gut-check, and an adjustment based on what’s sitting in the warehouse right now. For a 15-SKU catalog, that works fine. For 150 SKUs across multiple categories, you’re effectively guessing – just with spreadsheets.

The problem isn’t a lack of data. It’s that the data most eCommerce operators use for forecasting is about 30 to 90 days old by the time it influences a decision. Sales history tells you where demand was. The signals that tell you where demand is going are already in your product data – they’re just not surfaced in most analytics dashboards.

Why trailing revenue is a poor forecasting input

Revenue totals are lagging indicators. A product can look perfectly healthy in monthly revenue right up to the month its demand falls off, because the cohort of customers who buy it was quietly shifting weeks before the revenue moved. By the time the drop shows up in your monthly report, you’ve already placed the order.

The specific problem for eCommerce forecasting is timing. You’re placing an order today that will arrive in 4 to 6 weeks, to cover demand 6 to 10 weeks from now. Your trailing 30-day revenue reflects decisions customers made a month ago, under a different promotion calendar, different ad spend, and a different competitive environment. The past is not irrelevant – it’s just insufficient on its own.

What you need are leading indicators: signals that show where demand is moving before it shows up in revenue.

The four product-level signals that predict eCommerce demand

These four data points, tracked at the product and category level, give you a significantly earlier read on demand than trailing revenue does.

Product velocity trend (4-week rolling)

Velocity is not “what did this product sell last month.” It’s how the rate of purchase is changing week over week. A product with $9,000 in monthly sales and a 3-week declining velocity trend is a very different inventory bet than the same revenue with an accelerating trend. The first warrants conservative ordering. The second deserves a buffer.

Four-week rolling velocity, broken down by SKU and category, gives you a directional read in time to act on it.

Add-to-cart rate by product

Cart additions are an upstream conversion signal. They show demand before it completes as a purchase. A product with stable completed sales but a declining add-to-cart rate over two consecutive weeks is signaling that interest is softening before revenue catches up. For a 60-day inventory commitment, that 2-week lead time matters enormously.

In Stormly, the product performance view shows add-to-cart rate per SKU alongside purchase conversion rate, so these upstream signals are visible separately from completed transactions.

Cohort purchase cadence

Customers who made their first purchase 90 days ago typically return on a predictable schedule. If the 90-day cohort is underperforming the 60-day cohort at the same point in their lifecycle, demand from repeat buyers is contracting – even if new customer acquisition is masking it in total sales figures.

A skincare store running a monthly replenishment model might track a specific serum product: customers who first purchased in April showing a 38% 60-day repurchase rate, compared to customers who first purchased in January at 47%. That divergence is a forward-looking inventory signal most operators miss entirely. eCommerce customer retention analytics covers how to track these cohort-level patterns and catch declines before they hit revenue.

Category-level trend

Category trends precede SKU-level changes. If a “women’s activewear” category has been declining for three consecutive weeks, individual SKUs within it are at risk even if their individual numbers look stable. Tracking category velocity before SKU-level data gives you early warning before the product-level drop becomes obvious.

Building a weekly eCommerce forecasting workflow

The practical version of this is a structured weekly process, not a quarterly spreadsheet exercise.

Monday: velocity anomaly review

Before anything else, check which products deviated from their expected velocity pattern over the past 7 days. Stormly’s anomaly detection feed surfaces products where add-to-cart rate, order velocity, or category revenue shifted outside normal bounds automatically – you don’t build a custom report to find them. You’re looking for early warnings that warrant a closer look before the weekly order window closes.

Wednesday: cohort cadence check

Review repeat purchase rates for your top 20 products. Which products are pulling customers back on schedule? Which show declining cohort retention at the 30-day and 60-day marks? This separates durable demand from one-time spike behavior driven by promotions or influencer mentions that don’t repeat.

Products with strong cohort retention deserve heavier forward inventory coverage. Products with declining retention – even if gross sales look acceptable – are higher-risk bets for large orders.

Before each inventory decision: velocity vs. history comparison

When you’re about to place an order, compare the product’s 4-week velocity trend against its 90-day historical baseline. A product running 35% above its velocity baseline is undersupplied. A product with 20% lower velocity that still shows solid trailing revenue is a stock risk: it looks fine in the rearview mirror but the forward view is already softening.

Making your eCommerce dashboards actually actionable means structuring these three weekly checkpoints, not opening a report and scrolling through numbers hoping something interesting appears.

---

Ready to run product velocity analysis on your store? See your product trends in Stormly – start a free trial.

---

Three decision buckets, not a single number

The goal of eCommerce forecasting isn’t to predict exact demand – it’s to sort products into the right decision bucket so you order in the right direction.

Increase buffer stock: Products with 3 or more consecutive weeks of positive velocity deviation, strong cohort retention, and no category-level headwind. Under-ordering here is the most common and most expensive forecasting mistake. A product earning a new wave of loyal repeat buyers deserves availability.

Maintain baseline: Products with flat velocity, stable cohort cadence, and no anomaly signals. Order to the historical baseline with a 5 to 10% seasonal adjustment. No heroics needed.

Reduce exposure: Products with 2 or more weeks of declining velocity, falling add-to-cart rates, or weak cohort retention even when gross sales look stable. These are candidates for reduced orders, especially if the category trend is also softening.

Converting store data into weekly decisions maps these three decision buckets onto a broader framework that covers not just inventory but also marketing allocation and catalog prioritization.

What Shopify’s native analytics won’t tell you

Shopify’s analytics shows total revenue, order count, and session-level conversion rate. It doesn’t show velocity trends by SKU, add-to-cart rates per product broken out from revenue, cohort purchase cadence by first-purchase category, or anomaly detection for category-level shifts.

This gap compounds as your catalog grows. At 20 products you can track each SKU manually. At 200, the relevant signals are buried in export files no one processes weekly. Why Shopify’s numbers alone are hard to trust for planning gets at a related problem: even when you have the data, the numbers often don’t reconcile across tools, which makes forecasting off of them even harder.

The operators who make consistently good inventory decisions aren’t spending more time in spreadsheets – they have a product analytics layer that surfaces velocity and cohort signals on a consistent cadence without requiring manual report-building.

Using forecasting signals for decisions beyond inventory

The same velocity and cohort data that improves inventory decisions also informs three other high-value choices.

Promotional timing: Discounting a product whose velocity is already accelerating just trains customers to wait for a sale. Discounting a product with a softening velocity trend can reset the demand curve and generate a cohort of new buyers. Knowing which situation you’re in before planning the promo is the difference between protecting margin and burning it.

Catalog rationalization: Products with consistently weak velocity, poor cohort retention, and no category tailwind are holding working capital. Most operators only exit these products when the warehouse forces the decision – by which point the opportunity cost has already been paid. The data to identify them is already in your analytics.

Product sourcing: Merchants developing private-label products can look at which categories have high browse traffic but below-benchmark conversion rates – a gap that signals unmet demand. Layering velocity data on top tells you whether that demand is growing or plateauing, so you’re not investing in a declining category.

Turning eCommerce data into weekly operational decisions covers how to build these three decision types into a repeatable process for store operators and small analytics teams.

Whether this is really “agentic” or just better tooling

There’s a lot of noise in 2026 about AI and agentic analytics. The honest version of what this means for eCommerce forecasting is straightforward: instead of building a new anomaly report every Monday, the tool surfaces anomalies automatically. Instead of manually comparing velocity trends to baselines, the product dashboard shows deviations directly.

Whether agentic analytics is real or just BI with a better interface is worth reading if you want the full argument. For forecasting specifically, the practical impact is that you go from a process that takes 3 hours a week if you do it rigorously to one that takes 20 minutes – which is the difference between actually running the process and having it exist as a good idea in a shared doc somewhere.

Stormly runs the velocity monitoring, anomaly detection, and cohort cadence tracking automatically. The weekly product insights surface in the dashboard without requiring custom report builds. For a store with more than 50 SKUs, that automation is what makes the process consistently executable.

---

The operators who make consistently good inventory and catalog decisions aren’t smarter about forecasting – they’re looking at different data, a few weeks earlier. Product analytics gives you that earlier view.

See your product velocity and cohort data in Stormly – start a free trial.