Pilot participants and use cases
33 organisations from ~10 geographies and industries participated in the pilot. The use cases spanned a broad range of functional areas and LLM usage archetypes. Almost all were already in production, though mostly with humans in the loop.
2.1 Participant profile
GenAI applications from 17 organisations were put to the test during the pilot:
Each of these organisations was paired with 1 (or 2) of 16 specialist firms that provide software and/or services to test GenAI applications. In some cases, the “pairing” was done by the participants themselves, whereas in others, AIVF helped match deployers with testers.
About half of these 33 organisations were based in Singapore. The remaining came from 8 other jurisdictions– Canada, France, Germany, Hong Kong, Switzerland, Taiwan, UK, US.
2.2 Use cases
Background
16 of the 17 use cases were already live in production.
7 of them were in beta or/or rolled out to a limited group of users.
A majority were targeted at specialist users inside an organisation (e.g., software engineers at NCS). 5 were customer/ citizen-facing.
A human was “in the loop” in more than 2/3rd of the cases. Even in the remaining 5, there was significant human involvement outside the immediate workflow of the application.
Full list of use cases
| # | Tester(s) | Deployer | Use case |
| 1 | Advai | CheckMate | On-demand Scam and Online Fact-checker |
| 2 | AIDX | Fourtitude | Customer Service Chatbot (“Assure.ai”) for public sector and utility clients |
| 3 | AIDX | Synapxe | HealthHub AI Conversational Assistant |
| 4 | AIDX Aiquris | ultra mAInds | No-code AI-powered Retrieval Augmented Generation platform for Enterprise search and data connectivity |
| 5 | Fairly | MIND Interview | AI-enabled Candidate Screening and Evaluation tool |
| 6 | Guardrails PRISM Eval | CAG | AskMax Virtual Concierge Chatbot |
| 7 | Knovel | HTX | Productivity Co-pilot |
| 8 | LatticeFlow | Confidential | Investment Insights for Relationship Managers |
| 9 | Parasoft | NCS | AI-enabled Coding Assistant for refactoring code |
| 10 | PwC | SCB | Client Engagement Email Generator for Wealth Management Relationship Managers |
| 11 | PwC | UOB | Internal Q&A Chatbot |
| 12 | Quantpi | Unique | Investment Research Assistant |
| 13 | Resaro | MSD | Confidential |
| 14 | Resaro | Tookitaki | FinMate Anti-Money Laundering Assistant |
| 15 | Softserve | CGH | Medical Reports Summarisation |
| 16 | Verify AI | Confidential | Public Road Safety Chatbot |
| 17 | Vulcan | High-tech Manufacturer | Multi-lingual Internal Knowledge Bot |
2.3 Patterns of LLM usage
Across the 17 applications, LLMs were used in diverse ways.
The top 3 usage patterns were Summarisation, Retrieval Augmented Generation and Data Extraction from unstructured sources. These patterns align with the focus of many of these applications on staff productivity improvement.
LLMs were also used to power multi-turn chatbots, and to help translate between languages. Relatively few used LLMs as part of agentic workflows – yet.
The table below maps each of the 17 applications to the different LLM usage modalities:
