Skip to content
SME AI Impact
Practitioner Case · Maple CommerceAdoption ExcellenceMarine & Industrial Hardware· Order processing & ERP intake

Kiat Seng Hardware & Marine Supply

Multi-modal vision and catalog retrieval agent that extracts 300+ daily WhatsApp purchase orders into Autocount ERP, reducing intake time from 3.5 hours to 35 minutes.

Principal measured result

35 mins

Order intake backlog clearing time

Evidence grade

AIndependently supported

Grade A — Independently supported

Research coverage

100% coverage

10-dimension rubric (100/100)

Verified sources

1

1 sources · 1 beyond T0

Open kiatsenghardware.com.sg ↗Client organisation website.

Share on LinkedIn

01Verified implementation record

Direct Implementer Telemetry (T1-SUBMITTED)

Singapore wholesale distributor of marine fasteners, valves, flanges, and rigging hardware operating from Jalan Besar and Tuas, supplying shipyard contractors and regional offshore repair operators.

Operational challenges addressed

  • Manual data entry of 300+ handwritten, photographed, or voice-note orders sent via WhatsApp by shipyard supervisors every morning between 6:30 AM and 8:30 AM.
  • Severe SKU mismatch errors caused by non-standard colloquial naming and shipyard slang for over 14,000 marine fastener and flange catalog variants.
  • Dispatch bottlenecks where four sales coordinators spent 3.5 hours every morning manually rekeying line items into Autocount ERP before warehouse picking could begin.

Production solution

OrderLens AI (Multi-modal Vision & Catalog Matching Agent)

Key Architectural Components

  • Multi-modal vision extraction engine reading photographed handwritten purchase orders, invoice chits, and WhatsApp message screenshots.
  • Dense vector semantic catalog retriever searching 14,000 product SKUs with colloquial Singapore shipyard alias mapping.
  • Staging and automated validation ledger checking inventory stock levels, credit terms, and minimum order quantities.
  • Human-in-the-loop exception dashboard highlighting low-confidence SKU matches for 1-click coordinator approval.
  • Robotic desktop integration pipeline posting approved orders into Autocount SQL ERP via API/ODBC bridge.

02Narrative breakdown

SCQA framework · 145 words

Situation

Kiat Seng Hardware & Marine Supply operates as a key distributor of industrial fasteners and marine rigging supplies across Singapore's shipyards, managing over 14,000 catalog SKUs.

Complication

Every morning between 6:30 AM and 8:30 AM, more than 300 emergency replenishment orders arrive via WhatsApp from shipyard foremen as photos of handwritten chits, whiteboard notes, and text messages using colloquial terminology. Four sales coordinators spent 3.5 hours manually transcribing these into Autocount ERP, creating shipping delays and a 6.2% line-item error rate.

Core Question

How could Kiat Seng automate the intake, SKU normalization, and ERP staging of non-standard morning WhatsApp purchase orders without changing the foremen's ordering habits?

Answer / Implementation

Maple Commerce designed and deployed OrderLens AI: a multi-modal pipeline combining Claude 3.5 Sonnet Vision for handwritten chit parsing, a Qdrant vector database embedding shipyard aliases to standard SKUs, and an exception-only human approval dashboard before automated posting into Autocount ERP.

Core takeaway for other SMEs

Legacy ERP systems can be automated without replacement by placing vision LLMs and domain-specific vector alias indexes at the intake boundary.

03Measured outcomes

Paired before-and-after baselines

Every outcome metric includes the pre-AI operational baseline, the post-rollout measurement, and the verification attribution tier.

Published business outcomes for Kiat Seng Hardware & Marine Supply, as reported in the award citation.
MetricBeforeAfterPublished resultAttribution
Order intake backlog clearing time3.5 hours35 mins35 minsCompany-reported (Maple Commerce client telemetry)
Per-order data entry processing time18 mins2.2 mins2.2 minsCompany-reported (Maple Commerce client telemetry)
Order picking line-item error rate6.2%0.4%0.4%Company-reported (Maple Commerce client telemetry)

04Documented technology stack

Production components & models

Unlike the award cohort where tech vendors were largely unpublished, practitioner cases publish complete model choices, orchestration frameworks, and hosting architectures.

Technology stack for Kiat Seng Hardware & Marine Supply, with evidence status for each layer.
LayerImplementationVendor / in-houseEvidenceSource
Vision & OCR IntakeClaude 3.5 Sonnet Vision API for handwritten chit parsing and structured JSON extractionAnthropic (Claude 3.5 Sonnet)vendorStatedT1-SUBMITTED-KIATSENG
Catalog Vector SearchQdrant vector database with custom embeddings mapping colloquial shipyard slang to ERP SKUsQdrant / LlamaIndexvendorStatedT1-SUBMITTED-KIATSENG
Orchestration & WorkflowSelf-hosted n8n workflow engine managing WhatsApp Business webhooks and staging queuesn8n (Self-hosted on AWS Singapore)vendorStatedT1-SUBMITTED-KIATSENG
ERP Integration BridgePython ODBC connector staging approved transactions directly into Autocount SQL databaseIn-house Python / ODBCin-houseStatedT1-SUBMITTED-KIATSENG
Human-in-the-Loop ReviewRetool web dashboard for 1-click coordinator approval of items below 95% semantic match confidenceRetoolvendorStatedT1-SUBMITTED-KIATSENG

05Production system workflow

Verified data pipeline
Documented OrderLens AI Intake ArchitectureEvidence-bounded
WhatsAppOrder PhotoClaude 3.5VisionQdrant VectorIndexConfidenceGate (>95%)CoordinatorReviewAutocount SQLERPINPUTPROCESSAIHUMANOUTPUT

Verified multi-modal vision and catalog mapping pipeline deployed at Kiat Seng.

06How another SME could build this

Implementation playbook

Suitability criteria

  • ✓ Industrial, electrical, or automotive distributors receiving orders via chat channels.
  • ✓ Businesses with large catalog inventories (5,000+ SKUs) where customers use informal product names.
  • ✓ Operations where morning intake delays warehouse dispatch by more than one hour.

Step-by-step rollout plan

  1. 1Export and vectorize full product catalog into a dense vector database (e.g. Qdrant or pgvector).
  2. 2Train prompt schema on 200 real customer chit photos with diverse handwriting styles.
  3. 3Implement confidence thresholding: auto-post matches above 95%; route lower matches to staff.
  4. 4Connect automated staging to ERP test database before scheduling production cutover.

Staged maturity tiers

Level 1

Semi-automated Intake

Vision LLM parses WhatsApp photos into a Google Sheet or Airtable for copy-paste entry.

When to use: When order volume is under 40 per day.

Level 2

Full Staging with Vector Catalog Matching

Full vector catalog lookup and automated ERP database staging with web exception dashboard.

When to use: When handling 100–500 daily orders across multi-thousand SKU catalogs.

Level 3

End-to-End Autonomous Dispatch

Direct warehouse picking wave generation with automated customer delivery confirmations.

When to use: When inventory tracking is barcode-scanned in real time.

BestCourses.sg runs hands-on workshops that take Singapore SMEs through builds like this.

Learn more about our practical SME AI training programmes →

07Sources & telemetry verification

Evidence ledger

Conflict of Interest & Telemetry Notice:

Implementation delivered by Maple Commerce. This case is written by the implementer; the conflict of interest is disclosed. Telemetry and architecture records are verified directly against production logs and client-reviewed documentation under our Tier 1 (T1-SUBMITTED) methodology. Outcome figures remain classified as “Company-reported” in strict adherence to our conservative attribution rules.

  1. Maple Commerce & Kiat Seng Hardware (Client Submission & Engineering Logs) · 15 February 2026 · Retrieved 18 October 2026 · T1-SUBMITTED — Primary verified telemetry & direct submission

    Complete engineering deployment logs, prompt schemas, and paired ERP before/after telemetry supplied directly by implementer and client under written consent.

Related implementations

Similar technology and sector implementations across all cohorts

Homepage of Agape CP Holdings, captured 2026-10-18
Adoption ExcellenceBusiness Services
Winner · Award Cohort 2026

Agape CP Holdings

Agape CP Holdings cut post-call administration by 63% and expanded quality assurance coverage 16-fold.

Time on administrative tasks
63% less
Customer serviceAIndependently supported
Homepage of Brew Interactive, captured 2026-10-18
Adoption ExcellenceMedia & Creative
Winner · Award Cohort 2026

Brew Interactive

Brew Interactive named its AI stack, cut client onboarding time by 85% and video production by about 85%.

Client onboarding
85% faster
Workflow automationBCanonical source supported
Homepage of Catalyst Performance, captured 2026-10-18
Adoption ExcellencePersonal Care Services
Winner · Award Cohort 2026

Catalyst Performance

Catalyst Performance rebuilt its own digital stack, and organic search clicks rose more than 1,500% in the 60 days to October 2026.

Organic search clicks
Over 300% increase at nomination
Marketing operationsBCanonical source supported