How AI Changes Inventory Planning
Traditional forecasting uses static models (moving averages, seasonal indices). AI-native forecasting adapts in real time to demand signals, learns from outcomes, and explains its reasoning. That's what Tightly's Smart Replenishment engine does.
Traditional vs. AI forecasting
Classical inventory forecasting relies on statistical models: moving averages (average of the last N weeks of sales), exponential smoothing (recent weeks weighted higher), and seasonal indices (last year's December × growth factor).
These models have known weaknesses:
Lagging signals — they react to what already happened. A viral social post that drives a 10× sales spike isn't in historical data.
One model fits all — the same formula is applied to a fast-moving SKU and a slow-moving SKU, despite their very different demand patterns.
Manual seasonality — a buyer has to manually tell the system "December is peak season." If they forget or the pattern shifts, the model breaks.
AI-native forecasting addresses these by: automatically detecting seasonality patterns, applying different models to different SKUs based on their demand characteristics, and incorporating live demand signals (current sales velocity, not just historical averages).
Latent potential that hasn't been activated yet. Each new demand signal — a sales spike, a supplier delay, a new market — is aji in Tightly's model. The algorithm holds these possibilities and activates the right response at the right time.
Tightly's Smart Replenishment engine
Tightly's core algorithm uses a composite scoring matrix that prioritizes SKUs by replenishment urgency and revenue impact simultaneously. The key metric: Days of Stock (DoH) — how many days of inventory you have on hand based on current sales velocity.
The engine takes live inputs:
- Current stock levels (from Shopify + 3PL sync)
- Sales velocity (not just historical average — the trailing velocity weighted by recency)
- Inbound POs (stock already ordered but not yet received)
- Lead times (per supplier, updated based on actual performance)
- Seasonal events (configured by the buyer)
- Anomaly detection (sudden demand spikes or drops)
Output: a ranked list of replenishment recommendations, from "must order today" to "no action needed," with the reasoning shown for each SKU.
Explainability as a product feature
One of Tightly's stated differentiators is explainable recommendations. For each SKU, the system shows: current DoH, projected DoH at reorder point, lead time to receipt, inbound coverage already in transit.
This matters for a specific reason: buyers don't trust black boxes. If a tool says "order 500 units of SKU-1234 by Friday" with no explanation, a buyer will second-guess it or ignore it. If it says "order 500 units because at current velocity you'll hit zero stock in 18 days, your supplier lead time is 22 days, and there's no inbound PO covering this gap" — the buyer can verify, trust, and act.
As PM, explainability is not just a UX feature — it's a trust mechanism. The adoption rate of AI recommendations is directly tied to the buyer's confidence in the explanation. Tracking "recommendation acceptance rate" is a core model quality metric.
Select any text in the lesson to ask a question. Press Esc to close.