Plan further ahead
19% lowerforecast error
See six months ahead.
Lower error than using last month’s sales to forecast six months ahead.
Spark InventoryDemand forecasting software
Powered by Demand ESP
Demand forecasting software
A forecast method for every SKU.
Demand ESP picks a forecasting method for each SKU, corrects for stockouts and promotions, and turns the forecast into draft purchase orders you approve in the app or by text with Sparki.
Onboard with Sparki in app or your own AI assistant over MCP.
What your team gets
A Method Chosen for Each SKU
Demand ESP reads each item’s demand pattern, tests the eligible methods against that item’s own recent months, and keeps the one that forecast it best. You see which method was chosen and why.
Stockout Correction
A month an item was out of stock sold less because there was nothing to sell. Demand ESP scales those months by the days the item was actually in stock, so lost sales do not read as low demand.
Promotions Taken Out of the Baseline
The extra units an occasional promotion sold are measured and removed before the forecast is fitted, so a one-off spike is not mistaken for growth or seasonality.
Seasonality, Even With Short History
Items with a full year or more get their own seasonal shape. Newer items can borrow the seasonal shape of similar items until their own history builds.
How Demand ESP works
Four demand patterns, a method for each
Demand ESP classifies every item from its monthly sales history, then compares the methods that fit its pattern on the item’s own recent months and keeps the most accurate one. When no single method wins, it can use the month by month median of the eligible methods. With too little history to test, it uses the standard method for the pattern. No AI model touches the numbers: the forecast is statistical and repeatable, and Sparki only explains it.
Steady
Demand that holds around a level. Forecast with simple exponential smoothing.
Trending
Demand that is growing or shrinking. Forecast with damped Holt, so a trend eases off instead of running forever.
Seasonal
Demand with a yearly shape. Forecast with Holt-Winters once there are two full seasons of history.
Intermittent
Items that sell in some months and not others. Forecast with TSB, so the forecast falls as an item goes quiet.
Measured on public data
How accurate is Demand ESP?
We test Demand ESP against simple baselines on public sales data and publish how each figure was measured.
Put inventory to work
12% lessstock held
Hold less. Keep orders moving.
Less inventory than last-month forecasting, at the same 98% fill rate.
Get value sooner
Under 1hour to forecast
Bring your files. Start planning.
From export files to a working forecast with Sparki or your own AI assistant.
Powered by Demand ESP
Every SKU. Its own forecast.
Demand ESP reads each item’s demand pattern and automatically selects a forecasting method. See the method chosen and the history behind your forecast.
Forecast and inventory results from public-data testing.
How we measured it
The 19% and 12% figures were measured in September 2026 on an earlier version of our forecasting engine.
A method for each demand pattern
Demand ESP selects statistical methods for steady, trending, seasonal, and intermittent demand, using the available sales history and fallbacks for limited data.
What the 12% measures
In an inventory simulation using 400 Iowa public-data items, average on-hand inventory cost was approximately $4.12M for Spark Inventory versus $4.67M for last-month forecasting at an interpolated 98% fill rate. Both used the same order-up-to policy, a 14-day lead time, lost sales, and a uniform $12 unit cost.
What the 19% measures
Retrospective testing on 799 Iowa items found six-month pooled weighted absolute percentage error (WAPE) of 22.0% for Spark Inventory versus 27.1% for last-month forecasting, a relative reduction of approximately 19%.
What onboarding time measures
Measured from customer export files to a working forecast. Setup time varies with data readiness and scope.
Act on the forecast by text
When the forecast says an item will run short, Sparki can text you the alert and the drafted purchase order. Reply YES with its code to create it. Text with Sparki is included on Solo and up.
See Text with SparkiBring the data you already have
Connect Shopify, Amazon, WooCommerce, QuickBooks Online and more, or import sales history from a file. You can also ask for forecasts from ChatGPT, Claude or another assistant through Spark’s MCP connection.
Complete capability
Everything you need to run demand forecasting
Built into the same system that tracks stock, orders, purchasing, warehouses, and the decisions between them.
A Method Chosen for Each SKU
Demand ESP reads each item’s demand pattern, tests the eligible methods against that item’s own recent months, and keeps the one that forecast it best. You see which method was chosen and why.
Stockout Correction
A month an item was out of stock sold less because there was nothing to sell. Demand ESP scales those months by the days the item was actually in stock, so lost sales do not read as low demand.
Promotions Taken Out of the Baseline
The extra units an occasional promotion sold are measured and removed before the forecast is fitted, so a one-off spike is not mistaken for growth or seasonality.
Seasonality, Even With Short History
Items with a full year or more get their own seasonal shape. Newer items can borrow the seasonal shape of similar items until their own history builds.
New Products and Launches
An item with no sales yet can start from a lookalike group of similar items, or from a reference item you choose, instead of a forecast of zero.
Forecast Ranges, Not Just a Number
Every forecast carries a range from how far the item’s past forecasts missed, and that range feeds safety stock.
Backtest Your Own Items
Run a backtest on your own catalog: Demand ESP hides recent months, forecasts them from the history before, and scores the result.
Demand Curves and Signals
Add what the history cannot know: a planned launch curve, an expected change in demand, or a known event, layered on top of the statistical forecast.
Reorder Points From the Forecast
The forecast and supplier lead times set each item’s reorder point and safety stock, then turn into draft purchase orders you review.
From signal to action
A decision workflow your whole team can follow
01
Ask the onboarding agent
Use Sparki in app or your own AI assistant over MCP to inspect what exists and bring in the data your team already trusts.
02
See the recommendation
Spark Inventory turns sales, stock, and lead times into a clear next action with the reasoning attached.
03
Approve and move
Review, adjust, and approve in one place. Your team keeps control without rebuilding the analysis every week.
Demand ESP, explained
Demand forecasting questions
How Demand ESP picks a method, what it needs from you, and what it costs.
What forecasting methods does Demand ESP use?
Demand ESP uses statistical methods matched to each item’s demand pattern: simple exponential smoothing for steady demand, damped Holt for trending demand, Holt-Winters for seasonal demand, and TSB for intermittent demand. It compares the eligible methods on the item’s own recent months and keeps the most accurate one, or the month by month median of them when that beats every single method.
How does Demand ESP handle stockouts?
A month an item was out of stock understates demand. Demand ESP counts the days each warehouse actually had the item in stock and scales that month’s sales up to what it would have sold, so the forecast is not dragged down by lost sales.
How much sales history do I need?
Three complete months of sales are enough for a forecast. With six months or more, Demand ESP fits each method’s parameters to the item, and with a year or more it can find the item’s own seasonality. Newer items can borrow the seasonal shape of similar items.
Can I forecast a product that has not sold yet?
Yes. A new item can start from a lookalike group of similar items or from a reference item you choose, and you can add a launch curve or a demand signal for what you expect.
Is there a free plan?
Yes. The Free plan gives one user a per-SKU forecast with seasonal curves, refreshed monthly. Paid plans start at $99 a month with live forecasts, purchasing, unlimited users and Text with Sparki, and include a 14-day free trial.
See your own inventory plan in Spark Inventory
Upload the data you already have and turn it into the next decisions your team needs to make.
The Free plan includes a monthly per-SKU forecast for one user. Paid plans add live forecasts, purchasing and Text with Sparki, with a 14-day free trial. See pricing.