AI-Light Inventory Management: A 3-Signal Reorder Rule for SMBs

If you're an owner running a 60-200 person business with physical product, light packaging, or consumable supplies — and you've never written down your reorder logic — you're almost certainly buying too much of the slow movers and running out of the fast ones. The fix isn't a six-month ERP rollout; it's three signals and an AI that watches them for you.

Why do SMBs systematically over- and under-stock at the same time?

Because reorder logic in most 50-200-employee companies is one of two things: a fixed minimum somebody set 18 months ago, or "Maria checks the shelf and tells us." Both fail the same way — they don't react to changes in sell-through speed or supplier reliability. The Plan was "never run out." The Fact is "we ran out of the top SKU twice and have 9 months of dead stock on three others." The Gap is invisible until inventory count.

Definition: Light inventory management — a reorder discipline using a small number of well-chosen signals and approval-gated automation, designed for SMBs that don't need a full ERP module to control 200-2,000 SKUs.

The mistake most owners make is reaching for an enterprise-grade inventory system. For most SMB SKU counts, that's a 6-figure cost to solve a 5-figure problem.

What are the three signals?

Three numbers, computed per SKU, refreshed daily. None of them require AI to compute — but AI is what makes them comparable across hundreds of SKUs and what surfaces the anomalies.

Signal 1: Days-on-hand

Current stock divided by average daily sell-through over the last 30 days. Expressed in days.

Definition: Days-on-hand — the number of days of demand currently covered by on-hand inventory at the recent average sell-through rate.

This is the "how worried should I be" signal. Anything below the SKU's lead time + buffer is a reorder trigger.

Signal 2: Sell-through rate (and its trend)

Daily units sold for that SKU, smoothed over a rolling window (typically 14-30 days), with the trend direction noted (rising / flat / falling).

The trend matters more than the absolute number. A 30% rising trend on a slow SKU is a bigger reorder signal than a flat fast SKU.

Signal 3: Lead-time variability

For each supplier-SKU pair, the standard deviation of actual lead time over the last 12 deliveries. Suppliers that consistently deliver in 8 days don't need much buffer. Suppliers that deliver anywhere from 7 to 22 days need a fat one.

Definition: Lead-time variability — the spread of actual supplier delivery times around the supplier-stated average; the dominant input into how much safety stock you actually need.

Most SMBs ignore this signal entirely. It's the highest-leverage one of the three because it directly sets buffer size.

How does AI fit in?

Four jobs. Notice none of them are "the AI orders for you."

1. Threshold suggestion. For each SKU, AI looks at history and proposes a reorder point and quantity. The founder or ops lead reviews and approves.
2. Anomaly detection. When sell-through suddenly changes (3× spike, or drops to zero) AI flags it as a non-trend event that shouldn't trigger threshold adjustment.
3. Lead-time monitoring. AI watches supplier delivery times and updates the variability metric. When a supplier's variability degrades, AI surfaces it as a procurement issue, not a reorder issue.
4. Reorder draft. When signals cross thresholds, AI prepares the draft PO — supplier, SKU, quantity, expected arrival — for human approval.

The pattern: AI does the watching and the drafting. The human signs and learns from the suggestion.

Copy/paste reorder-rule template

This goes per SKU, surfaced in the daily ops digest.

SKU: [CODE]   Name: [TEXT]
Supplier: [NAME]   Stated lead time: [N days]   Actual lead-time variability: [σ days]

Current state:
- On-hand: [N units]
- Days-on-hand: [N days]
- Sell-through (14-day avg): [N units/day]   Trend: [rising/flat/falling]

Thresholds (AI-suggested, [HUMAN-APPROVED on DATE]):
- Reorder point: [N units] (= lead time + 1.65 × σ × daily sell-through)
- Reorder quantity: [N units]
- Maximum stock: [N units]

Decision today:
- [ ] No action
- [ ] Reorder draft prepared (see PO #XXXX)
- [ ] Threshold needs re-review — flagged because: [reason]

The "Threshold needs re-review" flag is the one that prevents the system from running on stale numbers. The most common reason it fires: sell-through trend has been rising or falling for 3+ weeks in a row.

Tool tip (AIAdvisoryBoard.me): Inventory is one of the cleanest Plan → Fact → Gap workflows in a physical-product SMB. Plan: per-SKU reorder thresholds, signed off by ops lead. Fact: days-on-hand, sell-through trend, and lead-time variability today. Gap: SKUs sitting below reorder point, SKUs trending toward stockout, suppliers degrading on variability. The 7-day diagnostic at https://aiadvisoryboard.me/?lang=en surfaces this pattern alongside every other operational cadence — the inventory shelf is just one more place where Plan diverges from Fact.

Good vs bad reorder logic

Bad: "Reorder 200 units of SKU-A whenever on-hand drops below 50." (Fixed min, fixed quantity, no signal awareness.)

Better: "Reorder when days-on-hand falls below lead time, in the quantity that brings us to lead time × 2 of cover." (Reacts to sell-through.)

Good: "Reorder when days-on-hand falls below lead time + 1.65 × supplier variability, in the quantity that brings us to (lead time × 2 + variability buffer). Threshold reviewed by ops lead monthly, AI-suggested." (Reacts to sell-through AND supplier reliability AND keeps a human in the loop.)

The good version is what almost no SMB does today — not because it's hard, but because nobody assembled the three signals into one decision rule.

Manager scan (2-minute digest example)

• Plan: 240 SKUs under active reorder logic, thresholds reviewed monthly
• Fact: 7 SKUs below reorder point today (drafts ready for approval), 3 SKUs trending toward stockout in 10 days
• Gap: Supplier B's lead-time variability degraded from σ=2 to σ=6 over last 3 deliveries — procurement follow-up needed
• Plan: zero dead stock over 180 days
• Fact: 14 SKUs sitting above 180 days-on-hand
• Gap: 11 of those 14 are seasonal — flag for end-of-season review; 3 are structural overstocks — schedule clearance
• Plan: monthly threshold review by ops lead
• Fact: last review was 7 weeks ago
• Gap: review meeting not on calendar — book this week

Micro-case (what changes after 7-14 days)

A 70-person specialty-foods distributor had 380 SKUs with reorder rules last updated 14 months prior. After standing up the three-signal pipeline, the first week surfaced two patterns: a top-12 SKU set that had been quietly stocking out for 4-6 days every cycle (sell-through had risen 28% since the last threshold review) and a long-tail of 23 SKUs holding 8+ months of cover (sell-through had fallen but reorder kept firing at the old minimum). The ops lead approved AI-suggested new thresholds for 67 SKUs the first week, took two weeks to walk through the others. By the end of week six, stockout days had dropped from a typical 9-12 per month to under 2, and total inventory value had dropped roughly 18% — without losing fill rate.

Note on this case: This example is illustrative — based on typical patterns we observe with companies of 30-500 employees, not a single named client. Specific numbers are rounded approximations of common ranges, not guarantees.

Tool tip (AIAdvisoryBoard.me): Inventory is one of those workflows where "small AI in the right spot" beats "big AI deployed everywhere." You don't need a demand-forecasting model trained on five years of seasonality — you need three signals refreshed daily and a human approval gate. The same Plan → Fact → Gap discipline that runs the daily company digest also runs the per-SKU reorder discipline. See how the 7-day diagnostic works at https://aiadvisoryboard.me/?lang=en.

FAQ

Should AI just auto-place the orders? For most SMBs, no. The cost of one wrong auto-PO is high (supplier annoyance, capital tied up, returns hassle) and the cost of a 5-second human approval is near zero. Keep the human signing for at least the first 90 days. After that, you can consider auto-approving low-risk SKUs.

What about seasonal SKUs? Two changes. First, the sell-through window should match the season (don't use a 30-day window for a 14-day season). Second, the AI should flag seasonal SKUs as a separate class so the human can override the trend signal when the trend itself is seasonal expectation, not anomaly.

Do we need this if our SKU count is under 50? Probably not — a spreadsheet and Maria check is fine. The breakpoint is usually around 100-150 SKUs, when no single person can hold the inventory state in their head reliably. Above that, the cognitive cost of "no system" exceeds the cost of building a light one.

What about supplier-managed inventory (VMI)? VMI is a separate decision and a good option for the slow tail. The three-signal rule still applies — you just hand the supplier the signals (or contract them to watch the signals) instead of running the reorder yourself.

Conclusion

Inventory failure modes in SMBs aren't exotic — they're "we never updated the thresholds" and "we never watched the supplier reliability." Three signals fix both. AI suggests, the founder approves, the system runs daily.

Pick your top 50 SKUs by revenue. Compute the three signals. Have AI propose thresholds. Approve in a 60-minute session. Watch what changes in the next two stock cycles.

If you want a system that surfaces the Plan → Fact → Gap automatically across the company — including the per-SKU reorder discipline — see how the 7-day diagnostic works at https://aiadvisoryboard.me/?lang=en.