TL;DR & Quick Summary
Nearly every inventory system advertises forecasting. Very few predict demand.
The distinction sounds pedantic and decides whether your reorder points ever get better:
- Forecasted stock — projects your position from what is already committed. On-hand, plus incoming POs, minus outstanding sales orders. Answers "when do I run out, given what's in the system?"
- Forecasted demand — predicts orders that do not exist yet, from history, seasonality and trend. Answers "what will customers want?"
Most systems do the first and call it forecasting. That is not dishonest, but it is routinely misread — and if your system only projects committed orders, a human is still setting every threshold by hand.
- Key Takeaway: Ask what inputs the forecast uses. Confirmed orders and stock means projection. Demand history and seasonality means prediction. The word "forecast" tells you nothing on its own.
- Get Started: Want to know which one your current stack does, and whether it needs to change? Schedule a Strategy Call with Cogniq AI or explore our predictive analytics services.
On specifics. Vendor capabilities and tier availability change frequently, and features often sit behind editions rather than being present or absent outright. Everything below is sourced to vendor documentation at the time of writing and should be re-verified before purchase. The taxonomy is what stays true; the feature list will move.
The Three Tiers
| Tier | What it does | Human effort required |
|---|---|---|
| 1. Static rules | Reorder at fixed min/max thresholds | You set and maintain every threshold |
| 2. Projected position | Forward-looking stock from committed orders | You still set thresholds; system times the order |
| 3. Predicted demand | Statistical forecast of future demand | System proposes thresholds; you review |
Most businesses believe they are at tier three and are actually at tier two. That gap is where inventory problems live, because tier two automates the timing of an order while leaving the quantity logic entirely to human judgment that rarely gets revisited.
Tier 2: Projected Position — What Most Systems Do
Odoo is a clean example, and its documentation is unusually explicit about it.
Odoo's reordering rules maintain stock between a minimum and maximum threshold. The order quantity is:
Order Quantity = Maximum − Forecasted Stock
Where forecasted stock is calculated from three inputs: current on-hand inventory, incoming supply (confirmed purchase and manufacturing orders), and outgoing demand (sales orders, component consumption, internal transfers) (Odoo documentation).
Every one of those inputs is a transaction that already exists. The system is not predicting anything — it is looking at commitments already in the database and projecting them forward.
This has a genuine advantage that gets undersold: it is fully auditable. Any projected shortage traces back to specific orders driving it, and the system flags shortages often weeks before they reach the floor (Odoo Skillz). When a planner asks "why is it telling me to order 400 units?", there is a traceable answer. Statistical models rarely offer that.
The limitation is equally clear. The minimum and maximum are yours. If demand doubles seasonally, the rule does not notice — it just triggers more often, later than it should, and you discover the problem as a stockout. Nothing in the system reviews whether your thresholds still make sense.
Tier 3: Predicted Demand
NetSuite Demand Planning predicts future inventory needs from historical demand, seasonality, open opportunities and sales forecasts, identifying when to reorder and in what quantity. It generates supply plans that automatically create purchase orders, transfer orders and work orders from calculated demand (NetSuite).
Note the inputs: historical demand and seasonality. That is prediction, not projection.
Cin7 ForesightAI applies machine learning and predictive analytics to forecast sales trends up to 24 months ahead, and calculates optimal reorder points and replenishment quantities automatically, suggesting purchase orders in-product (Cin7).
The important word there is "calculates reorder points" — the thresholds themselves become an output rather than an input. That is the actual dividing line between tier two and tier three.
SAP has embedded AI assistance across its supply chain portfolio, including refined forecast outcomes and actionable recommendations (Cin7 overview). SAP's supply chain footprint is large enough that "does SAP forecast?" is not answerable in general — it depends entirely on which modules you have licensed, and that is a question for your account team rather than a blog post.
The Question to Ask Any Vendor
One question separates the tiers, and it works regardless of what the marketing says:
"What inputs does the forecast use?"
- "On-hand stock, confirmed purchase orders, open sales orders" → tier two. Projection. You still own the thresholds.
- "Demand history, seasonality, trend" → tier three. Prediction.
Then a second question that matters just as much:
"Does it report forecast accuracy back to me?"
A system that predicts demand but never shows its error rate leaves you unable to size safety stock properly, because safety stock should be set against the forecast's error distribution rather than a padded guess. It also leaves you unable to tell whether the model is any good. Our forecast accuracy metrics guide covers what to ask for and how to read it.
Do You Actually Need Tier 3?
Not always, and the honest answer disappoints vendors.
Tier 2 is sufficient when:
- Demand is stable and non-seasonal
- Lead times are short enough to react rather than commit ahead
- The catalogue is small enough for humans to review thresholds properly
- Explainability matters more than accuracy — regulated environments, or teams that will not trust a model
Tier 3 earns its cost when:
- Demand is meaningfully seasonal or promotion-driven
- Lead times force commitment before demand is visible
- The catalogue is too large to maintain by hand
- Stockout cost on key lines is high enough to justify the investment
That last point deserves quantifying rather than assuming. Our stockout cost calculator covers how to work out what running out actually costs you per SKU — and it frequently comes out lower than expected on staples and much higher on high-margin substitutable lines.
A pattern worth noting: many businesses that buy tier three would have got most of the benefit from reviewing their tier two thresholds quarterly. Static min/max values set three years ago against demand that has since doubled are a common and entirely fixable problem. It costs an afternoon, not a licence.
Adding Prediction to a System That Lacks It
Replacing a working ERP to get forecasting is usually the wrong trade. The more common pattern keeps the system of record and adds modelling beside it:
- Pull demand history out via API — sales orders by SKU by period
- Generate forecasts externally, where models can be iterated without ERP release cycles
- Write reorder points or suggested quantities back in, so the ERP stays operational source of truth
Your planners keep working in one system. The modelling happens where it is easy to change.
The feasibility test is two questions about your ERP:
- Can you read a clean demand history out of it? Sales orders by SKU by period, without manual export.
- Can you write reorder points back into it? Some systems expose thresholds as writable fields; others hold them in configuration only reachable through the UI.
If both are yes, this integration is straightforward. If the second is no, you can still generate forecasts and hand planners a suggested-quantity report, but the loop stays manual.
For how the underlying methods work, our predictive analytics for inventory management guide covers reorder points and safety stock, and our AI demand planning guide covers the forecasting approach.
A Short Evaluation Checklist
Before committing to any system or upgrade:
| Check | Why it matters |
|---|---|
| What inputs drive the forecast? | Separates projection from prediction |
| Are reorder points an input or an output? | The real tier 2 / tier 3 dividing line |
| Is forecast accuracy reported back? | Needed to size safety stock and judge the model |
| Can demand history be read via API? | Determines whether external modelling is possible |
| Are reorder points writable via API? | Determines whether the loop can close automatically |
| Does it handle intermittent demand? | Many models fail badly on slow movers |
| Is promotional lift modelled separately? | Otherwise promotions corrupt the baseline forecast |
The last two are where most implementations disappoint. A model that treats a promotional spike as a permanent demand shift will over-order for months afterwards — one of the mechanisms behind the demand amplification covered in our bullwhip effect guide.
Conclusion
"Forecasting" in inventory software describes two genuinely different things, and the pricing page rarely distinguishes them. Projecting committed orders forward is useful, auditable and available nearly everywhere. Predicting demand that does not yet exist is a different capability with a different price and a different failure mode.
Work out which tier you are on by asking what inputs the forecast uses. Then decide whether moving up is worth it — which depends on your seasonality, your lead times, and what a stockout actually costs you, rather than on what the category is supposed to require.
Schedule a Strategy Call with Cogniq AI and we will look at what your current system can read and write before recommending anything, including when reviewing your existing thresholds gets you most of the benefit for none of the cost.
