Historical sales
Daily and weekly movement for each SKU and branch.
02 — Demand Forecasting Agent
Get ahead of demand shifts before they hit availability and stock
Keeper builds a probabilistic forecast for every SKU and location, measures its accuracy continuously, then links it to available inventory, open orders, lead times, safety stock and minimum order quantities. The result is not just a number — it's a date, a quantity and reviewable options.
An unplanned stock-out means lost sales and emergency purchasing at a higher-than-usual cost. This agent gives you enough lead time for a planned order, and links global market and price signals to its forecasts.
Keeper Command CenterIllustrative example of Keeper output
Stock-out risk17↑ 5 new SKUs
Forecast accuracy89%last 12 weeks
Value protected186KSAR estimated
Actual demand and forecast — SKU 184212 weeks
Order before August 7.
Current cover doesn't reach the expected lead time.
Under the hood
How does forecasting work? Machine-learning models that read what drives your demand.
You don't need a data-science team, but you deserve to know how the forecast is built. Here is a straightforward explanation, without hype or oversimplification.
1
Your own history first
Sales of every SKU in every branch: weekday rhythms, start-of-month and salary cycles, and how each category behaves across the year. Keeper relies on your data, your rules and your operational context when producing the forecast.
2
Seasons and occasions — on both calendars
Ramadan, both Eids, National Day and back-to-school season, on the Hijri and Gregorian calendars together. Keeper supports both calendars and local seasons, and tests their impact on each customer's own data instead of assuming a uniform effect across all sectors.
3
Weather
Heat waves, cold spells and rain enter the model as signals when their impact on a given category is proven from your data.
4
Global market signals
Harvests, freight, currencies, ports — events whose impact reaches your shelf weeks later, linked to the affected SKUs and turned into a coverage recommendation.
5
Your prices and promotions
The impact of discounts and price changes on demand is measured SKU by SKU, so a genuine seasonal rise is never confused with a spike caused by a single week's promotion.
6
Handling the effect of stock-out days
Keeper distinguishes periods of zero sales caused by stock-outs, and estimates unobserved demand based on available data, showing a confidence level, instead of treating it as zero demand.
How do the models work? For every SKU, several statistical and machine-learning models are compared, and the best fit is adopted for each SKU individually. Model accuracy is reviewed and retuned according to the agreed data cycle. And a forecast is a probability, not a prophecy; that's why it arrives with its range and confidence level, and accuracy is measured and shown to you periodically in numbers.
01 — Data
What data does Keeper need?
Keeper gathers the operational signals that actually affect the decision, while preserving the source and timestamp of each piece of information.
Available inventory
Actual balance, reserved, and in transit.
Lead times
Promised duration versus what actually happened.
Seasonality and promotions
Ramadan, holidays, campaigns and demand shifts.
Open purchase orders
What was ordered, what is delayed, and what's arriving soon.
Purchasing constraints
Minimum order quantity, pack sizes, and safety stock.
On top of your current systems
What does Keeper add on top of your existing systems?
Your current planning systems keep forecasts and plans per their own settings, and may offer advanced reports. Keeper does not replace these capabilities; it selects and tunes the models, links external signals to the decision, and compares alternatives by their impact.
| Area | Current planning systems | What Keeper adds |
|---|---|---|
| Forecasting | Forecasts and plans per system settings | Selects, tunes and monitors models with explanation and confidence range |
| Seasonality | Calendars and seasonal transactions | Links the local calendar to SKU and location behavior |
| Stock-out data | Handling depends on setup quality | Detects unobserved demand periods and shows their impact |
| External signals | May require additional sources and modules | Links the external signal to the affected SKUs and the decision |
| Decision | Suggested plans or orders | Compares alternatives and explains the impact of quantity, timing and supplier |
Keeper starts with a read-only connection by default. Draft creation or transaction writing can be enabled through specific integrations, permissions, and approval workflows agreed with the company.
02 — How it works
From data to an actionable decision.
Every recommendation follows a clear sequence your team can review: context, analysis, verification, then a proposed action with its reason and impact.
Gather context
Links demand history to inventory, open orders and lead times.
Test the models
Tests several models on your historical data and picks the best fit per category.
Calculate the risk
Calculates the expected stock-out date, probability and available lead time to act.
Actionable recommendation
Proposes the quantity, date and supplier, and prepares the order draft for approval.
03 — Output
This is what your team actually sees.
Keeper shows the priority, the number, the reason and the next step on a single screen. Illustrative example of the output below.
Keeper recommendation
Early alert — SKU 1842
Expected stock-out dateAugust 18
Suggested quantity1,800 units
Order dateBefore August 7
Confidence level87%
Why?
Demand rose 14% over the last 21 days, and the current supplier's average delay is 3.2 days past the agreed date.
Demand rose 14% over the last 21 days, and the current supplier's average delay is 3.2 days past the agreed date.
Every step stays linked to its source, timestamp and responsible user.
04 — Proof
Accuracy is shown, not asked for on blind trust.
Results are measured against your past and current data, and you see what worked, what needs improvement, and the confidence range of every recommendation.
Performance metrics (illustrative example)
Forecast accuracy over the last 12 weeks
89%SKUs at risk of stock-out
17 SKUsEstimated protected sales
SAR 186KModel update
Yesterday 11:40 PM"A forecast doesn't just give a number; it gives your team clear time to act before the problem becomes a reality."
Keeper design principle
05 — Governance
Clear intelligence, control stays in your hands.
Keeper starts with a read-only connection by default, and every action waits for the authorized user's approval before sending or writing to the system.
Backtesting
The forecast is compared against what actually happened before it's activated.
Confidence level
Every recommendation carries a reviewable confidence level.
Reason explanation
The factors that raised or lowered the forecast are shown.
Continuous improvement
Accuracy is reviewed per the agreed data cycle.
06 — Other capabilities
A team of agents working as one chain.
Each agent is independent in its task, but shares context and decisions with the rest of the team so departments don't work in silos.
Before you ask
The questions we hear most.
How does the agent handle local seasons such as Ramadan?
Keeper supports both the Hijri and Gregorian calendars and local seasons, and tests their impact on each customer's own data instead of assuming a uniform effect across all sectors.
How is forecast accuracy measured?
Accuracy is measured continuously against your actual sales at the SKU level, and the metrics and how they are calculated are shown to you periodically, along with a confidence level indicating the reliability range.
Where do global price and market signals show up?
It is a built-in capability of this agent; global commodity and freight price movements feed directly into the demand and cost forecast instead of being a separate report disconnected from the decision.