Every inventory decision collapses into one question: how many units will I sell between now and the moment my next delivery lands? Answer it well and cash flows. Answer it badly and you oscillate between stockouts and dead stock forever.
Most WooCommerce stores answer it with gut feel, a whiteboard, or a spreadsheet formula written once and never revisited. Vendors selling “AI forecasting” answer it with a black box nobody can interrogate. Professional inventory control sits between those extremes: a transparent statistical method, tuned per product, combined with lead times, open commitments and supplier constraints.
This guide shows you exactly how that works, including a worked example you can verify with a calculator. By the end you will be able to audit any forecasting tool on the market, including WiseStock.
1. The Question Forecasting Actually Answers
Reorder forecasting is not “predicting the future” in a mystical sense. It is estimating demand over a specific window: your cover period, which equals supplier lead time plus your review cycle.
If your supplier takes three weeks to deliver and you review suggestions weekly, every forecast you act on must cover four weeks of demand. Tools that quote you a “monthly sales prediction” without connecting it to lead time and review cadence are producing trivia, not purchase quantities.
Hold that frame: rate of demand, multiplied by cover period, adjusted for risk, minus what you already have or have incoming. Everything below is elaboration.
2. The Four Methods, From Naive to Professional
Method 1: Gut feel and static min/max
“Reorder when we hit 20, buy 100.” Set once, in year one, against demand that no longer exists.
Failure mode: decays into fiction as demand shifts. Excellent below roughly 50 slow SKUs, dangerous beyond that.
Method 2: Simple moving average
Average the last N weeks, reorder against that average. Better, because it references actual sales.
Failure mode: every week in the window carries equal weight, so a promotion from eleven weeks ago influences today’s order as much as last week’s reality. When the window rolls past an event, the forecast jumps off a cliff. It also ignores trend: a growing product is perpetually under-forecast.
Method 3: Exponential smoothing
A weighted average that gives recent demand more influence than old demand, with a single tunable dial controlling how much more. This is the workhorse of professional inventory systems, and the method inside WiseStock.
Failure mode: handled poorly on intermittent (lumpy, mostly-zero) demand, which is why mature systems pair it with low-demand protection (section 5).
Method 4: Machine learning
Neural or gradient-boosted models trained on rich feature sets: weather, price changes, ad spend, competitor moves.
When it earns its keep: enormous catalogs, dense sales history, external signals, and a data team to maintain it. When it does not: typical SMB catalogs with sparse, noisy history. There, ML overfits to noise, resists explanation, and charges enterprise prices for confidence it has not earned. A well-tuned smoothing model you can explain to your accountant beats a black box you cannot.
3. Exponential Smoothing Without the Academic Fog
The entire method is one line:
F(next) = (a x D(latest)) + ((1 – a) x F(previous))
Where D(latest) is the most recent period’s actual demand, F(previous) is last period’s forecast, and a (alpha, the smoothing constant) is a memory dial between 0 and 1.
- High alpha (0.4 to 0.6): short memory. The forecast chases recent demand aggressively. Right for fast movers with genuine trend shifts, wrong for noisy data.
- Low alpha (0.1 to 0.2): long memory. The forecast barely flinches at any single week. Right for stable or noisy SKUs where spikes are usually noise.
- The 0.2 to 0.4 band is a sane starting point for most catalog items, tuned per ABC class thereafter.
A worked example you can check by hand
A SKU selling weekly, alpha set to 0.3, seeded with week 1’s actual demand:
| Week | Actual demand | Forecast applied |
| 1 | 40 | 40.0 |
| 2 | 44 | 40.0 |
| 3 | 38 | 41.2 |
| 4 | 46 | 40.2 |
| 5 | 50 | 42.0 |
| 6 | 42 | 44.4 |
| 7 | 48 | 43.7 |
| 8 | 52 | 45.0 |
Check week 3 yourself: (0.3 x 44) + (0.7 x 40) = 13.2 + 28 = 41.2. Correct.
After week 8’s demand of 52, next week’s forecast is (0.3 x 52) + (0.7 x 45.0) = 15.6 + 31.5 = 47.1 units per week.
Notice the behavior: demand climbed from 40 to 52 over eight weeks, and the forecast tracked upward from 40 to 47 without whiplash. A moving average would lag further and then jump. Gut feel would have stayed at 40 until a stockout forced a panic order. This adaptive-but-steady behavior is exactly what reorder forecasting for WooCommerce stores needs, and it is why WiseStock exposes alpha as an adjustable setting rather than hiding it.
4. A Forecast Is Only Half the Answer
The rate (47.1 per week) becomes a purchase quantity through four adjustments:
- Multiply by the cover period. Lead time 3 weeks plus weekly review equals 4 weeks: 47.1 x 4 = about 188 units of expected demand.
- Add safety stock for variability in demand and lead time. (The full formula lives in our safety stock playbook, linked from the pillar guide.) Assume 40 units here: target becomes 228.
- Subtract what you have and what is coming. On hand 95, on order 60 against an open PO: 228 – 155 = 73. This step is where spreadsheets catastrophically fail, because they forget the 60 units already on a truck. A professional system counts open commitments automatically.
- Apply supplier constraints. MOQ of 50 and order multiples of 10 round 73 up to 80 units. That is your suggested purchase order quantity.
Four arithmetic steps, each visible, each defensible. That visibility is the entire point.
5. Where Naive Forecasting Breaks: Slow and Intermittent Demand
Now the honest caveat. Smoothing assumes demand arrives in a steady stream. Many catalog items do not behave that way: they sell zero units for six weeks, then nine in one day.
On such SKUs, plain smoothing decays toward zero (guaranteeing a stockout when the lump arrives) or whiplashes after each spike (guaranteeing dead stock afterward). Professional systems handle this with low-demand protection: floors based on cover-period needs for lumpy items, separate treatment of intermittent patterns, and refusal to chase single spikes. WiseStock ships this protection as a core behavior, not an add-on. (A full deep dive on intermittent demand and Croston-style methods is coming to this series.)
Equally important is knowing what not to forecast automatically: seasonal heroes, discontinued lines, and promotion-driven SKUs belong to human planning, not the algorithm. Per-product exclusion from automated suggestions is therefore a first-class feature, not a gap.
6. From Number to Decision: The Digest Workflow
Forecasts die in dashboards. They live in rituals.
A professional replenishment loop delivers suggestions on a cadence that matches your buying rhythm (daily, weekly or monthly digests; weekly suits most SMBs), and every suggestion carries a state: pending, snoozed, dismissed, resolved. States are a workflow language. “Snoozed until the seasonal changeover” and “dismissed, supplier discontinuing” are decisions with memory, so next week’s digest does not re-litigate last week’s judgment.
Approved suggestions convert directly into purchase orders, which then enter the approval and receiving workflow covered in our purchase order approvals guide. Forecast to PO to receipt to ledger, one continuous loop, one audit trail.
7. How to Audit Any Forecasting Tool (Including Ours)
Put these six questions to any vendor, and watch carefully:
- Can I see the method and its inputs? If the answer is “our AI handles it,” you are buying faith, not forecasting.
- Can I adjust the smoothing per product or class? One global setting means one global compromise.
- What happens to slow movers? Ask specifically about intermittent demand. Vendors without low-demand protection will change subject.
- Does it count open purchase orders in cover? Forgetting on-order stock causes duplicate ordering, the most expensive bug in inventory software.
- Can I exclude individual products from automation? Seasonal and discontinued lines demand a human lane.
- Can I export suggestions to compare against actuals later? Accuracy you cannot measure is accuracy you cannot trust.
WiseStock answers yes to all six, because transparency is the product: exponential smoothing with adjustable alpha, low-demand protection, on-order awareness, per-product exclusions, digest cadences you control, and CSV export of everything. Audit us on your own data. We would rather win that comparison than a marketing one.
Quick Answers
Is exponential smoothing really better than AI for my store? For typical SMB catalogs with sparse history, yes in practice: comparable accuracy, far better stability, and every number explainable. ML earns its premium only with dense data and external signals most stores do not have.
What alpha should I start with? 0.2 to 0.4 for most items. Raise it for fast movers with real trend shifts, lower it for noisy or stable lines, then tune per ABC class after a quarter of observed errors.
How often should I review and reorder? Weekly suits most SMB buying rhythms: frequent enough to catch drift, sparse enough to batch orders against MOQs and freight. High-velocity A-items can justify daily digests.
Does this handle seasonality automatically? Honestly, no statistical method handles sharp seasonal turns gracefully on its own. Smoothing adapts with lag, so seasonal heroes run on human planning with exclusions from automation, while the digest keeps the rest of the catalog truthful.
Forecasting you can interrogate. WiseStock brings adjustable exponential smoothing, low-demand protection, digest workflows and native purchase order conversion to WooCommerce, self-hosted and without the SaaS tax. Launching soon with a limited Founding Member Lifetime Deal. [Join the waitlist]