AI Demand Forecasting for Ecommerce Inventory

If you have already decided that spreadsheet-based reordering is no longer reliable, the next question is how much of the forecasting process to automate. This guide helps you prepare the data, select between Inventory Planner, Katana and Cin7 Core, run a controlled test and turn the output into a purchasing routine. By the end, you will have a forecast process that informs reorder quantities without removing human judgement.

What you need before you start

A useful forecast starts with clean sales history and realistic supply settings. Gather at least six to twelve months of Shopify orders where possible, current stock, open purchase orders, supplier lead times, minimum order quantities, returns, cancellations, stockout periods and known promotions. Separate discontinued products and one-off clearance activity so they do not distort normal demand.

For an ecommerce operations manager running a 200-SKU online store, the old process means manually checking Shopify sales history and guessing reorder quantities, causing stockouts on bestsellers and excess stock on slow movers. The new process flags reorder points, estimates seasonal spikes and produces a purchasing list for human review.

Step 1: Clean the history before forecasting

Do not ask a forecast model to explain data that the business itself has not reconciled. Match every Shopify variant to one stable SKU, confirm that returns and cancellations are represented correctly, and record periods when a product showed zero sales because it was unavailable rather than unwanted. A stockout can look like falling demand unless the tool knows the item could not be purchased.

Mark promotions, influencer campaigns, bulk orders and clearance events. These are useful signals, but they should not become the permanent baseline for ordinary weeks. For supplier data, use actual delivery performance where available rather than the optimistic lead time written on the original agreement.

Step 2: Choose the right forecasting setup

Inventory Planner is the best starting point when the main problem is ecommerce purchasing rather than replacing the whole inventory system. Its Shopify integration imports product and sales data, then uses patterns, seasonality and lead times to create replenishment recommendations. It supports safety-stock settings, multi-location planning and estimates for new products based on similar items, although pricing needs a quote and the vendor says go-live averages about four weeks.

Katana is stronger when the store makes, assembles or bundles products. Katana Core starts at $299 USD/month and Planning and Forecasting is $249 USD/month as of July 2026, making the setup $548 USD/month for the whole team, or $6,576 USD a year before other add-ons. That is easier to justify when the platform also manages materials, production and Shopify fulfilment.

Cin7 Core is better when forecasting sits inside a wider inventory-control project. Standard starts at $349 USD/month for five users as of July 2026, with ForesightAI sold as an add-on whose price is not public. The forecast report covers up to 12 months by product and location but needs at least six months of sales history in Cin7 Core.

Step 3: Connect Shopify and map every SKU

SKU mapping is the control point that prevents orders, stock and forecasts from drifting apart. Connect a test or restricted-access account first, import products, then verify a sample across variants, bundles, locations and archived items. A blue shirt in three sizes needs three correctly mapped variants, not one product name that hides the stock difference.

Choose which system owns each data field. Shopify may remain the storefront, while the inventory platform becomes the source of truth for available stock and purchasing. Document the direction of stock, order and fulfilment sync so staff do not make manual corrections in both systems and create a second discrepancy.

Step 4: Configure lead times, safety stock and seasonality

Forecast accuracy alone does not produce a sensible reorder quantity. The tool also needs supplier lead time, review frequency, minimum order quantity, case-pack size, target service level and safety stock. Configure these at SKU or supplier level where the differences matter, rather than applying one rule to all 200 products.

Use separate treatment for seasonal products and promotions. A Christmas item, a permanent bestseller and a newly launched colour variant should not share the same demand logic. Inventory Planner can use similar-product history for new items, while Katana expects manual input until enough real history develops, so staff still need to provide commercial context.

Step 5: Run a four-week shadow forecast

Run the system beside the existing process before allowing it to drive purchase orders. Each week, save the forecast, proposed reorder date and proposed quantity for a selected group of products. Compare those recommendations with actual sales, stockouts, supplier arrivals and the order the manager would have placed manually.

Review forecast bias as well as average error. A model that is slightly wrong in both directions may be manageable, while one that consistently underestimates bestsellers creates repeated stockouts. Use the shadow period to correct lead times, campaign flags, bundle treatment and outlier orders before expanding the SKU set.

Step 6: Turn forecasts into controlled reorder decisions

The first useful automation is a prioritised review queue, not unattended purchasing. Ask the platform to surface products that have reached a reorder point, explain the inputs behind the recommendation and group proposed purchases by supplier. The operations manager can then approve, adjust or reject each order with a recorded reason.

Only consider automatic purchase-order creation after the tool has performed reliably across several ordering cycles. Keep approval thresholds for high-value orders, new products, unusually large quantity changes and suppliers with unstable delivery times. The aim is faster, more consistent judgement, not removing accountability.

What success should look like

A successful forecast process improves availability and cash use at the same time. Track stockout rate for top sellers, excess-stock value, weeks of cover, emergency freight or rush orders, supplier-order changes, forecast bias and the hours spent preparing purchase orders. Compare results by product class because a single blended accuracy score can hide poor performance on the products that matter most.

Set a review after eight to twelve weeks. Keep the system if it produces better decisions and a repeatable workflow, even if individual forecasts are not perfect. Rework or abandon it if the team spends more time correcting the model than it previously spent planning stock.

Common problems and practical fixes

• The forecast is too low after a stockout: Mark the unavailable period so zero sales are not treated as zero demand.
• Promotional spikes distort normal weeks: Tag the campaign and enter the next planned promotion separately.
• New products have no history: Use a similar item in Inventory Planner, or a temporary manual forecast in Katana.
• Reorder dates look wrong: Confirm that lead time covers the full period until stock is available for sale.
• Shopify and the planner disagree: Recheck variant and location mapping, bundles, returns and the source of truth for stock.
• Too many recommendations appear: Start with A-class products and suppress discontinued or low-value items.

Implementation checklist

• Reconcile Shopify sales, returns, cancellations and stockout periods.
• Confirm stable SKU and location mapping for active variants.
• Load realistic lead times, order minimums and case packs.
• Select Inventory Planner for forecasting-first ecommerce, Katana for manufacturing, or Cin7 Core for broader inventory control.
• Connect with restricted access and verify samples before full sync.
• Configure safety stock, seasonality, promotions and new-product rules.
• Run a four-week shadow forecast on important products.
• Measure stockouts, overstock, forecast bias and planning time.
• Keep staff approval until several ordering cycles are stable.
• Review vendor permissions, security terms and data handling.

Frequently asked questions

Related reading: our AI governance by region.

How much sales history does ecommerce demand forecasting need?

Six to twelve months is a practical starting range, but a full seasonal cycle is better for products with strong annual peaks. Cin7 ForesightAI specifically states that it needs at least six months of sales history in Cin7 Core. New products still need a similar-item method or a manual starting forecast.

Can AI inventory forecasting work with Shopify?

Yes, all three options can support a Shopify-based workflow, but they do different jobs. Inventory Planner focuses on demand and replenishment, Katana combines Shopify with manufacturing and fulfilment, and Cin7 Core connects forecasting to a broader inventory platform.

Should the system automatically create purchase orders?

Not at the beginning. Start with recommendations that a staff member reviews, then add purchase-order automation only after the data, lead times and forecast behaviour have remained reliable across several cycles. Keep approval controls for high-value, unusual or new-product orders.

Which tool is best for a 200-SKU ecommerce store?

Inventory Planner is the best fit when the main need is Shopify demand forecasting and replenishment. Choose Katana when the business manufactures or assembles stock, and choose Cin7 Core when forecasting is part of a wider inventory, warehouse or accounting-system replacement.