How to Forecast Inventory Demand (2026 Step-by-Step for Multi-Channel Brands)
Forecasting inventory demand isn't guesswork — it's a six-step process most teams skip the first two of. Here's the operator's playbook for FBA + AWD + Shopify + Walmart, with the math, the inputs, and where most forecasts go wrong.
Quick Answer
Forecasting inventory demand is the six-step process of (1) pulling per-channel sales history, (2) calculating daily velocity per SKU per channel, (3) applying seasonality and trend adjustments, (4) factoring in lead time and safety stock, (5) producing a reorder-point + order-quantity recommendation, (6) reviewing and converting to PO decisions weekly.
The output: for every SKU, you know exactly when to reorder and how much to order, with the math behind each decision visible. For multi-channel brands, step 2 (unifying per-channel velocity) is where focused forecasting tools deliver the most value over spreadsheets.
The 6-step inventory demand forecasting process
Pull per-channel sales history
For every SKU you carry, pull the last 90 days of unit-level sales data per channel. Amazon FBA via Seller Central API. Shopify via store admin or app. Walmart via WFS reports. 3PL/DTC via order management system. Don't blend channels at this step — keep them separate so you can spot per-channel patterns.
Most teams skip this step. They use aggregate totals from QuickBooks or a single-channel view, which hides per-channel velocity differences. Same SKU sells 50 units/day on Amazon and 8 units/day on Shopify — treating them as "58 units/day combined" loses the allocation information that matters.
Calculate daily velocity per SKU per channel
For each channel, compute three trailing windows: 7-day (recent trend), 30-day (current pace), 90-day (baseline). Blend them with weighting that favors recent data but doesn't over-react to noise:
This gives you a velocity number that adjusts to recent acceleration or deceleration without flipping on a single big day. For brands with clear seasonality, replace the weighting with same-period-last-year comparisons during seasonal windows.
Apply seasonality and trend adjustments
If the SKU has multi-year history, layer in seasonality: this month vs same month last year, adjusted for current-year trend. If the SKU is new (under 12 months), use category seasonality as a proxy. For promo-driven brands, isolate promo periods from baseline so you don't over-forecast based on a sale month.
Common mistake: treating Q4 lift as permanent demand acceleration. Q4 spikes pull demand forward; January is usually down 30-40% vs October-November baseline. Forecast that into your post-Q4 ordering or you'll be sitting on inventory in February.
Factor in lead time and safety stock
Lead time = days from PO placement to inventory received. Safety stock = buffer days above the forecast to absorb demand variability + supplier delays. Both should be per-supplier-per-SKU when possible, with brand-level defaults as fallback.
The math from our forecasting fundamentals guide:
Produce per-SKU reorder recommendations
For every SKU, output: days of supply remaining, reorder point hit (yes/no), recommended order quantity, channel allocation breakdown (how much to FBA, AWD, Shopify, Walmart, 3PL). Sort by "days of supply remaining" ascending so the urgent SKUs surface first.
For Amazon-heavy brands, add per-FNSKU 2026 fee math here: flag SKUs approaching the 28-day low-inventory-fee threshold, flag SKUs approaching the 181-day aged surcharge, flag AWD shipments that should be released to FBA.
Review and convert to PO decisions weekly
Forecast output without a weekly review cadence is just a report nobody reads. The discipline that separates brands who win at forecasting from brands who buy software and ignore it: every Monday, a human looks at the recommendations, sanity-checks 3-5 outliers, and converts the top SKUs into actual POs.
This is the step managed-service tools (like SKU Compass Tier 2) productize — the analyst does the Monday review on your behalf and surfaces just the decisions you need to approve.
What changes for multi-channel brands
If you sell on more than one channel, step 2 (per-channel velocity) and step 5 (channel allocation) become the most important steps. Single-channel forecasting can use simple trend math; multi-channel requires unification logic that most spreadsheet templates can't handle cleanly.
| Channel | Velocity input | 2026 fee adjustment |
|---|---|---|
| Amazon FBA | Per-FNSKU sales velocity | Low-inventory fee + aged surcharge factored in |
| Amazon AWD | FBA velocity drives AWD-to-FBA release schedule | $0.48/cu ft storage vs FBA rate decides upstream-vs-FBA allocation |
| Shopify | Per-product sales velocity from store admin | No marketplace fee adjustment; standard reorder math |
| Walmart WFS | Per-item velocity from WFS reports | 270-day aged window vs FBA's 181 = WFS for slower SKUs |
| 3PL / DTC | Per-SKU velocity from order management system | No platform fee adjustment; operator workflow |
The unification problem: same SKU sells at different velocities on each channel. Treating them as one number averages away the per-channel signal. Treating them as five separate forecasts misses the cross-channel substitution effect (when Amazon stocks out, Shopify demand can absorb some of it).
Three things that break inventory forecasts
Stale lead time data
If you set lead time once at onboarding (say, 90 days for overseas manufacturing) and never update it, your forecast doesn't reflect supplier reality. Suppliers run faster or slower season-by-season. Track actual lead time per PO and update the per-SKU default every 90 days.
Missing on-order visibility
If you have a 5,000-unit PO landing next Tuesday and your forecast doesn't know about it, the system tells you to order another 5,000 units today. Double-order = sitting on inventory in 60 days. Every open PO must be visible to the forecast.
Demand spike misread as new baseline
Viral mention, competitor stockout, ad spike, one-time wholesale order — all can produce a temporary demand jump that looks like a permanent acceleration in trailing-window math. Always sanity-check spikes against context. If you can't explain why the 7-day velocity doubled, treat it as noise until the 30-day confirms.
The honest caveat
No forecast is perfectly right. Demand variability, supplier delays, channel shifts, and one-off events make perfect forecasting impossible. The goal isn't perfection; it's being directionally right and faster than your competitors at reacting.
The brands that win at forecasting aren't the ones with the fanciest models. They're the ones with clean per-channel data inputs, realistic safety stock policies, and a Monday-morning review cadence that converts forecast outputs into actual PO decisions weekly — not quarterly.
Frequently asked questions
How do you forecast inventory demand?
Six-step process: (1) pull per-channel sales history (90 days minimum), (2) calculate daily velocity per SKU per channel using weighted trailing windows (7/30/90 day), (3) apply seasonality and trend adjustments, (4) factor in lead time and safety stock, (5) produce per-SKU reorder-point + order-quantity recommendations with channel allocation, (6) review weekly and convert to PO decisions. Steps 1-5 are the math; step 6 is the discipline that separates winners from losers.
What is the formula for forecasting inventory demand?
Reorder Point = (Daily Velocity × Lead Time Days) + Safety Stock Units. Order Quantity = (Daily Velocity × Coverage Period) − On-Hand − On-Order. Daily Velocity itself is typically a weighted blend: (0.5 × 7-day avg) + (0.3 × 30-day avg) + (0.2 × 90-day avg) to balance recent trend with stability.
How often should you forecast inventory demand?
The forecast should auto-refresh continuously (daily or hourly) as new sales data lands. The HUMAN review of forecast output should happen weekly: every Monday, look at the recommendations, sanity-check 3-5 outliers, convert top SKUs into POs. Quarterly review is too slow — you'll miss demand shifts. Daily review wastes time on noise.
What data do you need to forecast inventory demand?
Five inputs per SKU: (1) per-channel sales history (90 days minimum), (2) current on-hand inventory by location, (3) on-order quantities with expected receive dates, (4) lead time per supplier per SKU, (5) safety stock policy (brand-level default + per-SKU overrides). Without all five, the forecast is incomplete and will produce wrong recommendations.
How do you forecast demand for multi-channel inventory?
The key shift from single-channel: keep velocity separate by channel, then allocate inventory across channels based on per-channel demand. Same SKU might sell 50/day on Amazon FBA and 8/day on Shopify — treating them as "58/day combined" loses allocation information. Cross-channel substitution effects (FBA stockout pushing demand to Shopify) require unified forecasting tools that most spreadsheet templates can't handle.
What is the difference between inventory forecasting and planning?
Forecasting is the tactical math underneath (predicting demand, computing reorder points). Planning is the broader strategic decision-making (which SKUs to carry, capital allocation, supplier strategy). Forecasting runs continuously; planning runs quarterly. Most tools focus on one or the other; few do both well. Full comparison here.
What software is best for forecasting inventory demand?
For multi-channel brands at the $5M-$50M ARR line, SKU Compass leads on FBA + AWD + Shopify + Walmart unified with 2026 fee math built in. For Shopify-primary brands, Inventory Planner. For Amazon-only at scale, SoStocked. For ERP-shaped operations, Cin7 Core.