Your inventory already has the products. Claude reconciles them with judgment.
Claude reconciles every product that comes in even if it arrives with a different name, deduplicates bulk imports, separates what's available from what's reserved and in-transit, and alerts you to stockout risk with coverage and in-transit stock in view. It is a judgment layer on top of the inventory you already have.
1
SKU per product
however the supplier names it
0
blind entries
anything doubtful goes to a person
3
decisions
match · review · create new
3
industries
same engine running
UNDER THE HOOD
Every SKU it assigns comes with its reason.
An incoming product almost never arrives with your exact nomenclature. Claude exposes the decision cascade — SKU, name, similarity, AI — with the confidence score of each step so your team trusts the match.
Sample upload
The upload comes in
Supplier invoice line: "SKU TAL-450-NEG · PERCUSSION DRILL 450W BLACK · 24 units". The normalized SKU matches character-for-character with a product in the catalog.
Step 1 · Exact SKU
TAL-450-NEG → direct match in catalog, confidence 100%
Name cross-validation
"percussion drill 450w" matches the base description → reinforces the match
No ambiguity
single candidate → subsequent steps not evaluated
Inventory action
+24 units to origin warehouse · movement recorded · no human review
Score
100/100
Verdict from Claude
Automatic match. Applied to available stock. Zero intervention.
The upload comes in
Factory XLSX without SKU column: "Self-drilling screw 1/4 x 2 zinc box 500". Catalog has "SELF-DRILL SCREW 6mm x 50mm ZN x500" with no shared SKU.
Step 1 · Exact SKU
no SKU in source row → discarded
Step 2 · Exact name
strings not identical (1/4" vs 6mm) → does not apply
Step 3 · Jaccard similarity
shared tokens {screw, self-drill, zinc, 500} → Jaccard 0.71 ≥ 0.6 threshold
Match confidence
0.71 above threshold → match proposed, flagged for optional audit
Score
71/100
Verdict from Claude
Similarity match. Applied with review flag. Your team confirms with one click.
The upload comes in
Scanned PDF invoice: "SDS+ drill bit for concrete 12x160 — imported brand". No SKU, Jaccard 0.42 with three different candidates. Text judgment not sufficient.
Steps 1-3 unresolved
no SKU · no exact name · Jaccard 0.42 < 0.6 → escalates to Claude
Step 4 · AI with context
Claude reads description + 3 candidates + attributes (diameter, shank type)
Semantic disambiguation
"SDS+" + "12x160" → concrete drill bit SDS-plus shank 12mm x 160mm, candidate 2
Model confidence
reasoned match with textual justification → confidence 0.88, above auto-apply threshold
Score
88/100
Verdict from Claude
Match resolved by AI. Justification attached. Product enters the correct warehouse.
WHAT AI DOES IN YOUR INVENTORY
Five jobs your team should no longer do manually
Each one runs on every product and every import that comes in, with its confidence threshold. Your team receives reliable stock, not spreadsheets to reconcile.
claude.match(incomingProduct)
Finds the correct SKU even if it arrives with a different name
A 4-step cascade: exact SKU → exact name → Jaccard similarity ≥ 0.6 → AI fallback. Every incoming product exits linked to your catalog with a confidence score.
› input
Invoice line: "self-drill screw 1/4x2 zinc x500" (no SKU)
› claude →
step 1 SKU: no match
step 2 name: no exact match
step 3 Jaccard: 0.71 ≥ 0.6 → candidate
→ match: SELF-DRILL SCREW 6mm x500
confidence: 0.71 · flag: optional review
claude.dedupe(xlsxImport)
Bulk import of thousands of rows without duplicating your catalog
Factory XLSX with variable column mapping. Detects rows that already exist, merges variants, and applies prices with margins by client type — without bloating the catalog.
› input
Factory XLSX: 2,840 rows · non-standard columns
› claude →
column mapping: auto-detected
2,840 rows → 2,610 unique products
230 duplicates merged
184 new · 2,426 updated
→ prices by client list applied
claude.reorder(sku)
Alerts you about the stockout before you run out of stock
Projects consumption against real available inventory (minus reserved, plus in-transit) and fires the reorder alert with remaining days of coverage — not when it's already too late.
› input
SKU TAL-450-NEG · avail 38 · reserved 12 · in-transit 0
› claude →
consumption: 6.4 units/day (trend ↑)
real available: 26 units
coverage: 4.1 days → below threshold
⚠︎ projected stockout: Thursday
→ suggests reorder: 120 units to supplier A
claude.sync(shopify)
Keeps your stock aligned across warehouses and channels
Bidirectional sync with Shopify and your core via REST API. When an order is delivered it deducts; when a pickup is received it auto-replenishes. Reserved stock is never sold twice.
› input
Event: order #4821 marked delivered (warehouse 2)
› claude →
deduction: -8 units warehouse 2
reserved released: -8
Shopify: stock updated ✓
associated pickup PO → auto-replenish
→ core notified via webhook
claude.parseInvoice(pdf)
Converts a PDF invoice into inventory lines
Scanned or digital supplier PDF. Claude extracts lines, quantities, costs, and links them to your catalog with the matching engine — no manual typing.
› input
PDF: supplier invoice · 17 lines · scanned
› claude →
lines extracted: 17 / 17
14 exact match · 2 fuzzy · 1 by AI
unit costs captured ✓
1 line flagged for review
→ inventory entry pre-assembled
LIVE SYSTEM · Grupo Latitud
This is what it looks like inside an AI-operated system — and how it is governed.
Availability is stock minus what is reserved, plus what is on the way. This tour follows that datum from real position to the purchase that gets signed.
Modules in this tour
- 01Available, reserved and in transit
- 02What is coming is already promised
- 03An alert that already looked at transit
- 04Where the AI stops on its own
- 05The same datum, end to end
Boosty Standard for Operating with AISimulated AI · demo data
THE SAME ENGINE
One inventory. Signals change, judgment doesn't.
We don't rewrite the logic per industry. The same judgment engine reads each business's own inventory. This is already running in production.
Signals specific to the industry
Signals specific to the industry
Signals specific to the industry
Signals specific to the industry
Signals specific to the industry
Signals specific to the industry
LIVE GRID
Warehouse × product, moving right now
Available, reserved and in transit by location. When the real available crosses the minimum, the cell lights up on its own. This is what your team sees, not a static sheet.
450W hammer drill
TAL-450-NEG
reserved: 12
in transit: 0
real 26 · min 25
reserved: 8
in transit: 40
real 96 · min 30
reserved: 4
in transit: 0
real 5 · min 15
Self-drilling screw 6mm x50
TORN-6X50-ZN
reserved: 80
in transit: 0
real 340 · min 200
reserved: 30
in transit: 500
real 650 · min 150
reserved: 12
in transit: 0
real 83 · min 100
SDS+ drill bit 12x160
BRC-SDS-12160
reserved: 6
in transit: 24
real 40 · min 12
reserved: 2
in transit: 0
real 5 · min 10
reserved: 5
in transit: 0
real 26 · min 18
Nitrile glove size M
GUA-NIT-M
reserved: 20
in transit: 0
real 120 · min 80
reserved: 10
in transit: 0
real 50 · min 50
reserved: 4
in transit: 0
real 14 · min 25
MATCHING ENGINE
A product comes in raw. It leaves linked to your catalog.
The 4-step cascade with a visible confidence score. The match stops at the first step that resolves — it does not spend AI if the SKU already matches.
Raw incoming product · Supplier invoice (digital)
SKU TAL-450-NEG · 450W BLACK HAMMER DRILL · 24 units
Step 1 · Exact SKU
conf 1.00TAL-450-NEG → matches character by character
Step 2 · Exact name
not evaluated — already resolved
Step 3 · Jaccard similarity
not evaluated — already resolved
Step 4 · AI fallback
not evaluated — already resolved
Match result
TAL-450-NEG · 450W hammer drill · confidence 1.00
IN THE SYSTEM · MASTER CATALOG
Above the threshold it is recorded on its own. Below it, a person decides.
A supplier load is compared line by line against the master record, with the confidence of each match and the explicit threshold. A new series is not made up: it stays for review.
Real screenshot of the system · demonstration data
STOCKOUT PREDICTION
It warns you on Tuesday, not when you are already at zero.
It projects consumption over the real available (net of reserved). When the line crosses the minimum threshold, it fires the reorder alert with days of coverage — before the stockout.
Engine reading
available: 38 units
reserved: −12 units
real available: 26 units
consumption: 2.6 units/day (trend ↑)
coverage: 10.0 days
Projected stockout
Real available crosses the minimum on day 7. With no action, stock hits zero before the fortnight close.
Suggested reorder
120 units to supplier A · 5-day lead time → arrives before the stockout. Alert sent to purchasing over WhatsApp.
IN THE SYSTEM · STOCK ALERTS
An alert that says what it was crossed with.
The shortage against transit, demand and pipeline, with the proposed transfer or purchase. The purchase order is signed by a person.
Real screenshot of the system · demonstration data
CONNECTED STACK
Lives on top of your inventory. Doesn't replace it.
The judgment layer connects to what you already use. If your core or ERP has a REST API, we talk to it.
Claude · Anthropic
Claude PartnerThe judgment engine: matching, dedupe, invoice parsing, and prediction
Supabase
Deno Edge Functions for real-time matching + stock realtime via triggers
Make / n8n
Webhooks to any ERP or legacy core without a modern API
n8n
Orchestration of bulk imports, automatic reorders, and reconciliation
Monday.com
PartnerSynchronized catalogs and reorder board when stock drops below minimum
WhatsApp Business
Stockout alert reaches the purchasing manager where they actually see it
Frequently asked questions about Smart Inventory
No. It's a judgment layer that lives ON TOP of the inventory you already have. If your core or ERP has a REST API we integrate it natively; if it's legacy with no API, we connect it via Make/n8n. Your team keeps working where they already work — only now stock comes in reconciled and reliable.
A 4-step cascade: first exact SKU, then exact name, then Jaccard index similarity with threshold 0.6, and only if none of that resolves, it escalates to Claude with the candidates in context. Each match comes with a confidence score and, if it's below the auto-apply threshold, it's flagged for one-click review.
It's modeled explicitly. Real available is physical minus reserved in pending orders plus confirmed in-transit. Reserved stock is never sold twice and stockout predictions use real available, not gross.
You define the threshold. By default: exact SKU match applies on its own (confidence 100%), fuzzy match comes in with a review flag, and the AI fallback self-applies only if it exceeds your confidence threshold. When you trust the judgment, you raise the automation. It's gradual, not a leap of faith.
Yes. The XLSX import auto-detects column mapping even if it's non-standard, deduplicates against your existing catalog, merges variants, and applies prices with margins by client type. We tell you how many products were new, how many were updated, and how many duplicates were merged.
Yes, it's natively multi-warehouse: warehouse × product grid with available, reserved, in-transit, minimum, and maximum per location. Price lists by client type (wholesale, installer, distributor, retail) are applied with per-SKU overrides.
Yes. Bidirectional sync with Shopify via Edge Function: when an order is delivered it deducts, when a pickup is received it auto-replenishes, and it keeps stock aligned between warehouses and channels so what doesn't exist can't be sold.
It is defined in the assessment. The scope — what gets built first and what waits — comes from what we see in your operation, not from a catalog. Book 30 minutes and we give you the range in writing.
A WORD FROM THE FOUNDER
“An inventory adds up when every incoming product is recognized, even when it arrives under another name.”
An inventory works like a bilingual dictionary: the supplier writes “1/4 screw”, your catalog says “6 mm”, and someone has to know they are the same part. When that translation is done by hand, import after import, the count stops adding up and nobody can say which line drifted.
Smart inventory puts that translation inside the system: the AI reconciles every incoming product against the catalog, states the confidence of each match and sets aside the ones below the threshold. The ones that pass go in; the doubtful ones go to a person with both options in view. The company sets the threshold, and every match is on record.
For the business, that means an inventory that works as the single source of truth for sales, purchasing and finance. Book 30 minutes with me: I’ll show you the matching engine live on one of your imports. How many supplier imports does your team reconcile by hand today?
Gabriel Montiel
CEO · Boosty Digital
LET'S TALK
Ready for an inventory that has judgment?
Schedule a 30-minute assessment. We'll bring a real import of yours and show you Claude's matching engine live. No corporate presentation.