Skip to content
SME AI Impact
Innovation ExcellenceRetail· Order processing

Limteh Project

Limteh Project reads personalised gift orders with AI and cut rework on orders by 80%.

Principal published result

80% less

Rework on orders

Evidence grade

BCanonical source supported

Grade B — Canonical source supported

Research coverage

90% coverage

Ten evidence dimensions, not an AI maturity score

Sources

3

3 sources · 2 beyond T0

Open limtehproject.com ↗Live site, not an official page of this showcase.

Share on LinkedIn

01The official record

Award citation, verbatim in substance
Homepage of Limteh Project, captured 2026-10-18
Limteh Project's homepage, captured 2026-10-18. Shown for identification and comment; it may differ from the award-winning implementation.Open live site ↗

Creates customised corporate gifts, gift sets and personalised presents

Published key challenges

  • Complex personalised orders. Personalisation details from customers through order notes, customer chats and messages in different languages
  • Labour-intensive order processing. Staff manually process around 2,500 orders each month, creating a daily production bottleneck
  • Costly errors. Order mistakes resulted in reprints, reshipping, wasted materials and unhappy customers

Milo: AI-powered Order Automation

  • AI order interpretation: Reads orders and chats across marketplaces to identify each customer’s personalisation requirements
  • Automated production preparation: Turns customer requirements into production instructions and automates print-file preparation
  • AI-powered verification: Cross-checks order details and flags uncertainties for staff review
  • Continuous improvement: Learns from staff corrections to improve order processing over time
  • Business and marketing automation: Automates sales reporting, stock alerts, profit-and-loss dashboards and creation of marketing content

02SCQA case study

356 words
SSituation

Limteh Project creates customised corporate gifts, gift sets and personalised presents. Each order carries customer-specific text, names, logos or artwork. The business runs on marketplaces and chat channels, so the personalisation instruction arrives as free text in whichever channel the customer used.

CComplication

Personalisation details arrived through order notes, customer chats and messages, sometimes in different languages. Staff processed around 2,500 orders a month by hand, which created a daily production bottleneck. Order mistakes produced reprints, reshipping, wasted materials and unhappy customers. In a made-to-order business, an error found after printing cannot be corrected cheaply, because the material is already consumed. Staff time was also the constraint on customer service, because the same people answered enquiries, prepared production files and chased artwork approvals, so a busy week in production became a slow week for customers.

QQuestion

How could the company interpret thousands of free-text personalisation requests accurately enough to send straight to production, without a person retyping every order?

AAnswer

Limteh built Milo, an AI order automation system that carries personalised orders from sales to production. It reads orders and marketplace chats to identify each customer's personalisation requirements. It turns those requirements into production instructions and automates print-file preparation. A verification step cross-checks order details and flags uncertain orders for staff review, which keeps a person on the ambiguous cases rather than on every case. The system learns from staff corrections, so accuracy improves with use. Beyond production, it automates sales reporting, stock alerts, profit-and-loss dashboards and marketing content creation. The published result is 80% less rework on orders, more than 20 hours saved each week, and up to $5,800 saved monthly. The pattern is a narrow, high-value extraction task with a human exception queue, rather than a general assistant. The design also keeps a person on the ambiguous cases rather than on every case, which is what makes the throughput gain compatible with made-to-order accuracy at this volume. The monthly saving is small in absolute terms, which is precisely why it matters here: at this order volume the fix had to be cheap enough to justify itself quickly, and a large platform investment would not have paid back.

Implementation takeaway

In made-to-order production, accuracy is cheaper than speed. Route uncertain orders to a person before the printer starts.

03Verified business results

Company-reported
Published business outcomes for Limteh Project, as reported in the award citation.
MetricPublished resultAttribution
Rework on orders80% lessCompany-reported; published in the IMDA/SBF award citation
Staff timeMore than 20 hours saved each weekCompany-reported; published in the IMDA/SBF award citation
CostUp to $5,800 saved monthlyCompany-reported; published in the IMDA/SBF award citation

The award citation publishes relative changes and absolute volumes, not paired before-and-after values. Baseline figures are therefore not shown, and no improvement percentage has been calculated from them.

04Technology stack

Evidence status per layer
Technology stack for Limteh Project, with evidence status for each layer.
LayerImplementationVendor / in-houseEvidenceSource
AI modelLanguage model for order interpretation and verificationNot disclosedunknownStatedT0-AWARDS
ProcessingProduction instruction generation and automated print-file preparation for heat press and engravingIn-house, unnamedin-houseStatedT0-AWARDST1-LIMTEH-SITE
IntegrationMarketplace chat and order channels across Shopee, Lazada and e-commerce storefrontShopee, Lazada, ShopifyvendorStatedT0-AWARDST1-LIMTEH-SITE
FeedbackLearning from staff correctionsIn-house, unnamedin-houseStatedT0-AWARDS
ReportingSales reporting, stock alerts and P&L dashboardsIn-house, unnamedin-houseStatedT0-AWARDS

A layer marked inferred includes a documented basis. Layers without a named vendor are left unnamed rather than filled with a plausible product.

05Implementation diagrams

Documented vs illustrative
Documented implementationEvidence-bounded
Marketplaceorders andchatsOrderinterpretationAIverificationStaff reviewof flaggedordersProductioninstructionsand…PrintedpersonalisedorderCorrectionlearningINPUTPROCESSAIHUMANOUTPUT

Components named in the award citation.

Illustrative reference architectureIllustrative only
Orderexport frommarketplaceFieldextractionwith…ValidationrulesExceptionqueuePrinttemplatefillPrint-readyfileINPUTPROCESSAISYSTEMHUMANOUTPUT

A feasible build for another made-to-order business. This is not Limteh's production system.

06Unconfirmed — technical inference

Not attempted

No public technical details were identified beyond the official award description. This case is not supported well enough to justify a technical inference, so none is offered.

07How another SME could build something similar

Advisory, not the winner's method

This section is guidance for other companies. It is not a description of what Limteh Project did, and it is not endorsed by them.

Suitable for

  • · Corporate gifting and personalisation
  • · Print and apparel
  • · Custom manufacturing
  • · E-commerce with order notes

Minimum requirements

  • · Order export with the raw customer text intact
  • · A defined output schema for production
  • · A person available to clear an exception queue daily

Three implementation paths

Level 1

Structured order form

Constrain customer input with a form so personalisation arrives as fields, not prose.

Use when: Customers can be guided at the point of order.

Level 2

Assisted extraction

Extract fields from free text and route low-confidence orders to staff.

Use when: Orders arrive through marketplaces and chat.

Level 3

Automated production handoff

Generate print-ready files automatically, with verification and a correction feedback loop.

Use when: Accuracy is proven and volume justifies integration.

Suggested sequence

  1. 1Collect 500 past orders with the final production instruction as ground truth
  2. 2Define a strict extraction schema
  3. 3Measure field-level accuracy before automating file generation
  4. 4Add the exception queue and track its size
  5. 5Feed corrections back into prompts or examples weekly

Tools and alternatives

  • · Marketplace API or order export
  • · Language model with structured output
  • · Validation rules
  • · Print or production template engine

Risks and PDPA considerations

  • · Hallucinated personalisation text
  • · Mixed-language orders misread
  • · Exception queue growing faster than staff can clear it

KPIs to track

  • Field accuracy against ground truth
  • Rework rate
  • Exception rate
  • Orders processed per staff hour
  • Cost per order

08Research gaps

What remains unknown
  • The foundational LLM model powering Milo is not specified.
  • Specific marketplace API rate limits and webhook architectures are unstated.
  • Detailed error categorization for the human exception review queue is not published.

Are you this company? Submit documentation to upgrade your evidence grade.

If you have published technical documentation, vendor whitepapers, or third-party audits, submit documentation to our research desk. We update evidence grades upon verification.

09Sources

3 records
  1. Infocomm Media Development Authority (IMDA) and Singapore Business Federation (SBF) · 16 October 2026 · Retrieved 18 October 2026 · T0 — Canonical awards source

    Canonical source. Page last updated 16 October 2026. Publishes one profile per winner: company description, three key challenges, named AI solution, solution components and three headline metrics. No individual winner URLs are published.

  2. Infocomm Media Development Authority (IMDA) · 13 October 2026 · Retrieved 19 October 2026 · T1 — Direct primary source

    Official government media release for SMEs Go Digital Day 2026 and inaugural SME AI Impact Awards at Suntec Singapore Convention & Exhibition Centre. Minister Josephine Teo presented awards to 30 SMEs.

  3. Limteh Project Pte Ltd · Retrieved 19 October 2026 · T1 — Direct primary source

    Company website detailing personalised corporate gifts, bespoke drinkware, and made-to-order production workflows.

Researched 18 October 2026 · Updated 18 October 2026 · Award citation retrieved 18 October 2026.

Limteh Project is a winner of the SME AI Impact Awards 2026, organised by IMDA and SBF. This case study is independent analysis and is not an official company publication.

Related case studies

Similar technology and sector implementations

Homepage of IYLA Collective, captured 2026-10-18
Adoption ExcellenceRetail
Winner · Award Cohort 2026

IYLA Collective

IYLA Collective used Claude and Claude Code to quadruple content output and save S$14,000 in annual labour.

Annual recurring labour
More than $14,000 saved
Content generationBCanonical source supported
Homepage of Accountingbpo, captured 2026-10-18
Innovation ExcellenceBusiness Services
Winner · Award Cohort 2026

Accountingbpo

Accountingbpo turned a headcount-bound accounting back office into a supervised AI workflow.

Accounting document processing time
Up to 67% reduction
Document processingBCanonical source supported
Homepage of iAPPS Health Group, captured 2026-10-18
Innovation ExcellenceHealthcare
Winner · Award Cohort 2026

iAPPS Health Group

iAPPS Health Group built an AI-native clinic operating system and lifted doctor productivity up to four times.

Doctor productivity
Up to 4x increase
Clinical decision supportBCanonical source supported