Recommended order
24 units
Coca-Cola 500ml · ABC Distribution
Why?
- Current stock is expected to fall below safety stock in 2 days.
- Supplier lead time: 3 days.
- Expected demand during lead time: 19 units.
- Safety stock: 8 units.
84%High confidence
Soffen AI
AI that shows
its work.
Not one giant language model. A set of small, specialised engines that run on your device, predict what is likely to happen, and explain every recommendation in numbers you can check.
What “offline AI” means
Forecasts that work
without the internet.
Core forecasting and recommendations run on the device. A cloud LLM is not required for the inventory forecast — which keeps behaviour predictable and your data private.
Local statistical forecasting
Moving and weighted averages, recent trend and day-of-week effects.
Small local models
Lightweight on-device models only where they measurably help.
Rules engine
Safety stock, case packs, minimum orders and supplier constraints.
Local product data
Your own sales history — not someone else's averages.
If a language model is used, its only job is to turn structured results into natural language. It does not invent inventory numbers, and on-device and cloud AI are always labelled so you know when information leaves the device.
Architecture
Eight steps from data to decision.
Each engine has one job. If any of them is unavailable, inventory, sales, suppliers, orders and reports keep working.
01
Local database
Products, sales, deliveries and adjustments on your device.
02
Data normalisation
Fills gaps, flags missing days and unexplained adjustments.
03
Feature engine
Recent averages, trend, day-of-week effects, lead times.
04
Forecast model
Expected daily demand for each product.
05
Stockout engine
When stock is likely to run out.
06
Reorder engine
How much to order, rounded to case packs.
07
Confidence engine
How reliable each prediction is — and why.
08
Insight generator
Plain-language insights you can act on.
The result reaches you as an insight, a recommendation, or an answer from the assistant.
Explainability
Every recommendation
answers “why?”
Trust comes from evidence. Each recommendation lists the numbers it was built from, and you can accept, edit or dismiss it. Nothing is ever ordered silently.
Confidence
Never presented as certainty.
Every prediction carries a confidence level based on measurable factors. When the data is thin, Soffen says so — and lowers its confidence.
Water 1.5L
91%High confidence
90 days of consistent sales, reliable counts.
Orange juice 1L
73%Medium confidence
Demand spiked last week; recent trend less stable.
Protein bar (new)
41%Low confidence
Only 5 days of reliable sales data. Initial estimate uses comparable products in the category.
Confidence goes down when there is…
- Little sales history
- Inconsistent sales
- Frequent stock adjustments
- Missing days
- Unusual demand spikes
- Uncertain supplier lead time
Forecast horizons
1, 3, 7, 14 and 30-day forecasts. Short-term forecasts generally receive higher confidence than long-term ones. Example for a product with steady sales:
Typical confidence by horizon: 1 day about 95%, 3 days 88%, 7 days 80%, 14 days 68%, 30 days 52%.
Insight types
Seven kinds of insight.
Short, specific, actionable. Written the way a careful store manager would say it.
- Stockout risk
“Likely to run out in 3 days.”
- Demand increase
“Sales are 18% above the recent baseline.”
- Demand decrease
“Sales have declined for 14 days.”
- Overstock
“Current inventory is significantly above recent demand.”
- Slow moving
“This product has had very low sales over the last 30 days.”
- Expiry risk
“24 units may remain when this product reaches its expiry date.”
- Supplier timing
“Current stock may not last until the next expected delivery.”
- Not enough data
“I don't have enough data to make a reliable prediction yet.”
Try it
The forecast, in your hands.
Target stock = daily demand × (lead time + 2-day review) + safety stock. The order is the difference from current stock, rounded up to a full case.
Choose a product
Coca-Cola 500ml
CriticalExpected stockout
~2.9 days
at 6.2 units/day
Recommended order
24units
4 cases of 6 · ABC Distribution
AI confidence
84%High confidence
Based on 30 days of sales data.
Likely to run out ~0.1 days before a delivery ordered today could arrive.
- Projected stock
- With recommended order
- Safety stock
- Reorder point
- ⚠ Stockout
Coca-Cola 500ml: 18 units in stock, selling 6.2 per day. Expected stockout in ~2.9 days. Supplier lead time 3 days, safety stock 8 units, reorder point 27 units. Recommended order: 24 units, arriving on day 3. Likely to run out ~0.1 days before a delivery ordered today could arrive. Confidence 84%.
Tip: hover, tap, or focus the chart and use ← → to read exact values.
Why this recommendation?
- Current stock18 units
- Average sales6.2/day
- Recent demand+18% vs 30-day avg
- Supplier lead time3 days
- Demand during lead time19 units
- Safety stock8 units
Guardrails
What the AI will never do.
Invent numbers
Stock levels, sales, lead times, prices and forecasts come only from your store data.
Order silently
Recommendations are suggestions. You review, edit and approve every purchase order.
Change prices
Expiry suggestions never change prices unless you explicitly set up an integration.
Overwrite inventory from a guess
Future shelf scanning will always ask you to review before stock changes.
Claim to “learn” without learning
Accepted, edited and dismissed recommendations are stored locally; we'll only say the model learns when it actually uses them.
Become a single point of failure
If AI is unavailable, the store keeps running and your inventory data stays safe.
Know what's next.
Before it's gone.
Start with your own products or explore the demo store. No account, no card, no internet required.