IEQ Engine: Demand Forecasting & Reorder Points
How IEQ Engine forecasts daily product demand from your stock movements, chooses a forecasting model per product, labels its confidence, suggests reorder points and flags customers due to reorder.
IEQ Engine is the demand forecasting engine inside IEMSuite. IEQ stands for Inventory EQuilibrium: keeping each product's stock in balance, with enough on hand for the demand ahead and not much more. For each product it reads how stock has actually left your warehouse, works out what kind of demand the product has, picks the forecasting method that performs best on that product's own history, and returns three things: a 30-day daily demand forecast with a confidence label, a suggested reorder point for your supplier lead time, and a list of regular customers who are due to reorder.
Availability
IEQ Engine is in beta and is being rolled out gradually. It is switched on per workspace, and it also depends on your plan's forecasting allowance. When it is not enabled, the sidebar item is hidden and opening the page directly shows "IEQ Engine isn't available yet". Contact support if you would like early access. On the Plans & billing page, plans that include it list "IEQ Engine demand forecasting" with their daily request allowance.
Opening IEQ Engine
Open Finance → IEQ Engine in the sidebar
The item appears only when forecasting is enabled for your workspace and your role can view inventory.
Choose a product
Use the product selector. The list contains your products in alphabetical order.
Read the 30-day demand forecast
The chart, confidence badge, demand pattern and model used refresh for the selected product.
Enter your supplier lead time
In the Reorder point card, set Supplier lead time (days) and, if deliveries vary, Lead time standard deviation (days).
Review customers due to reorder
This card covers the whole workspace, not only the selected product.
What It Reads
Demand is built only from stock transactions that take stock out for real use:
| Transaction | Counts as |
|---|---|
| SHIP_OUT | Stock shipped to a customer |
| CONSUME | Materials consumed by a production batch |
| MANUAL_CONSUMPTION | Stock used outside a batch and recorded by hand |
- Not demand: inbound receipts, transfers, adjustments, wastage and returns.
- Zero days count: every calendar day from the product's first recorded demand is included, and days with no movement count as zero. Quiet periods lower the forecast instead of being skipped.
- Window: the forecast uses up to the last 365 days. The current day is excluded until it is complete (days are cut at midnight UTC).
- Products: only active products are forecast. Archived and deleted products are excluded.
Nothing has to be imported or set up: the engine reads the records IEMSuite already keeps.
Demand Patterns
Before forecasting, the engine classifies each product. The result is shown on the forecast card as "[pattern] pattern".
| Pattern | Meaning | Models tried |
|---|---|---|
| intermittent | More than 30% of days are near zero, or demand varies extremely (coefficient of variation above 1.5). | crostons, tsb, moving_average, naive_seasonal |
| trend | A clear upward or downward slope across the history. | prophet, simple_exp_smoothing |
| volatile | Large day-to-day variation (coefficient of variation above 0.5) without a clear trend. | prophet, moving_average, crostons, tsb |
| stable | Fairly even demand. | simple_exp_smoothing, naive_seasonal, moving_average |
The baselines mean_28d and median_28d are added to every pattern.
Forecasting Models
The card shows the winning model under Model used, using these names:
| Model used | What it does |
|---|---|
| naive_seasonal | Weekly pattern: repeats the shape of the most recent week. |
| moving_average | Average of the last 14 days, carried forward. |
| simple_exp_smoothing | Exponential smoothing: a running level weighted toward recent days, tuned per product. |
| crostons | Croston's method: for products that sell in bursts, estimates order size and the gap between orders separately. |
| tsb | Teunter-Syntetos-Babai: a Croston variant that also tracks how often demand occurs, so a fading product fades in the forecast. |
| prophet | Prophet, an open-source model for trend and weekly seasonality, with yearly seasonality when there is enough history. |
| mean_28d | Average of the last 28 days. |
| median_28d | Median of the last 28 days, less affected by one-off large orders. |
How a Model Is Chosen
- Held-back testing. The engine sets aside recent history, fits each candidate on the days before it, and compares the candidate's forecast with what actually happened. Two separate 14-day windows are used: the earlier one picks the model, the later one measures it. The measuring window never influences the choice.
- Score. Each candidate is scored on WMAPE (weighted mean absolute percentage error, counted over every day including zero-demand days) plus an equal penalty for bias, meaning a consistent lean high or low. Consistent under-forecasting causes stockouts, so it is penalised as much as scatter.
- No empty forecasts. A candidate that would forecast zero for a product that clearly sells is never chosen.
- Baselines win ties. A more complex model must beat the 28-day baseline by at least 2% to replace it, so the choice does not flip on noise.
- Trends. For a product classified as trend, a flat model (moving average, exponential smoothing or a 28-day baseline) cannot win while a trend-capable model is available.
- Refit. The winner is refitted on the full history before producing the 30-day forecast.
- Always current. Selection runs again every time a forecast is requested, so the model follows changes in a product's demand.
The 30-Day Forecast
The 30-day demand forecast card charts expected demand for each of the next 30 days, starting today. Hover a day to see the value in the product's unit. The shaded range is sized from the error the chosen model made on held-back days, so a product that is hard to forecast gets a visibly wider range; with no measured error, the product's own day-to-day variation is used. The lower edge never goes below zero.
Confidence Labels
The badge reads, for example, "medium confidence · 42.0% WMAPE". The percentage is the error the chosen model made on the held-back measuring window for this product.
| Label | Held-back error (WMAPE) |
|---|---|
| high | Below 30% |
| medium | 30% to below 70% |
| low | 70% or more |
| unknown | Less than about 49 days of history: too little to hold back two 14-day windows and still train |
Reorder Point
The Reorder point card suggests the stock level at which to place a new order so that demand during the supplier lead time is covered in about 95% of replenishment cycles.
- Supplier lead time (days): 1 to 180, default 7.
- Lead time standard deviation (days): 0 to 90, default 0. Enter how much delivery time varies.
How it is calculated. The calculation uses the last 90 days of the product's demand.
- When there are at least 20 lead-time-length stretches in that history and the standard deviation is 0, the engine adds up demand over every stretch as long as your lead time and takes the level that covered 95% of them. This follows your product's real ups and downs rather than assuming a bell curve.
- Otherwise it uses the standard formula: average daily demand × lead time + safety stock, where safety stock accounts for demand variation and, if entered, lead time variation.
What the card shows:
- Reorder at - the suggested level, in the product's unit
- Avg. daily demand
- Demand variability - low, moderate or high, judged against the average (± daily variation shown)
- Safety stock - the buffer included above average lead-time demand
- Lead time and Target service level
- Your current minimum - the product's minimum stock level, marked "(below recommended)" when it is lower than the suggestion
Customers Due to Reorder
This card lists customers expected to place their next order of a product within the next 7 days, based on the rhythm of their past orders. It covers the whole workspace and shows up to 8 entries, soonest first. Each entry shows the customer, the product and "Nd until expected reorder"; the badge colour reflects how strong the evidence is.
- Orders from up to the last two years are read. Cancelled orders and quotes are excluded.
- A customer needs at least 5 orders of the product.
- Customers whose order spacing is too irregular to predict are left out.
- Customers already past their usual date are left out, and so are those silent for more than three times their usual gap.
- Strength of evidence: high needs at least 8 orders with evenly spaced intervals; medium needs at least 8 orders or fairly even spacing; everything else is low.
When nobody qualifies, the card shows "Nothing urgent".
Daily Request Limit
Each workspace has a daily allowance of forecast requests set by its plan, counted over a rolling 24 hours. Loading the page and choosing a product makes several requests: one for the forecast, one for the reorder point (and another each time you change the lead time), and one for the customer list. When the allowance is used up the page shows "Too many forecast requests today - please try again tomorrow."
Data and Security
- Forecasts run on IEMSuite's own forecasting service. Your records are not sent to an outside AI provider.
- The service has read-only access to the database and cannot change your records.
- It accepts requests only from the IEMSuite app, which checks the signed-in user's permissions and limits every request to that user's workspace.
- Viewing forecasts requires the inventory read permission.
Messages and Troubleshooting
| Message | What to do |
|---|---|
| Not enough history yet / Not enough recorded demand history for this product yet | The product has fewer than 14 days of shipment or consumption history. Wait for more history. |
| No reorder point yet | The product has no recorded shipment or consumption in the last 90 days. |
| Too many forecast requests today | The daily allowance is used up. Try again after the 24-hour window. |
| This feature isn't available on your current plan | Your plan does not include forecasting, or it is not enabled for your workspace. |
| Forecasting is temporarily unavailable | The forecasting service could not be reached. Try again in a moment; contact support if it persists. |
Limitations
- Forecasts follow past demand. They cannot anticipate events missing from your history, such as a new large customer, a promotion or a supplier problem.
- New products and products that rarely move get wide ranges and low or unknown confidence until history builds up.
- The forecast horizon is fixed at 30 days and is calculated per product.
- Reorder points use a fixed 95% service level target.